Megatest

# make install CSCOPTS='-accumulate-profile -profile-name $(PWD)/profile-ww$(shell date +%V.%u)'
PREFIX=$(PWD)
CSCOPTS= 
INSTALL=install
SRCFILES = common.scm items.scm launch.scm \
           ods.scm runconfig.scm server.scm configf.scm \
           db.scm keys.scm margs.scm megatest-version.scm \
           process.scm runs.scm tasks.scm tests.scm genexample.scm \
	   http-transport.scm nmsg-transport.scm filedb.scm \
           client.scm gutils.scm synchash.scm daemon.scm mt.scm dcommon.scm \
	   tree.scm ezsteps.scm lock-queue.scm sdb.scm \
	   rmt.scm api.scm tdb.scm rpc-transport.scm \
	   portlogger.scm archive.scm

# Eggs to install (straightforward ones)
EGGS=matchable readline apropos base64 regex-literals format regex-case test coops trace csv \
     dot-locking posix-utils posix-extras directory-utils hostinfo tcp-server rpc csv-xml fmt \
     json md5 awful http-client spiffy uri-common intarweb spiffy-request-vars \
     spiffy-directory-listing ssax sxml-serializer sxml-modifications iup canvas-draw sqlite3

# Special dependencies for the includes
tests.o db.o launch.o runs.o dashboard-tests.o dashboard-guimonitor.o dashboard-main.o monitor.o dashboard.o megatest.o : db_records.scm
tests.o runs.o dashboard.o dashboard-tests.o dashboard-main.o  : run_records.scm
db.o ezsteps.o keys.o launch.o megatest.o monitor.o runs-for-ref.o runs.o tests.o : key_records.scm
tests.o tasks.o dashboard-tasks.o : task_records.scm
runs.o : test_records.scm
megatest.o : megatest-fossil-hash.scm
client.scm common.scm configf.scm dashboard-guimonitor.scm dashboard-tests.scm dashboard.scm db.scm dcommon.scm ezsteps.scm fs-transport.scm http-transport.scm index-tree.scm items.scm keys.scm launch.scm megatest.scm monitor.scm mt.scm newdashboard.scm runconfig.scm runs.scm server.scm tdb.scm tests.scm tree.scm zmq-transport.scm : common_records.scm rpc-transport.scm

# Temporary while transitioning to new routine
# runs.o : run-tests-queue-classic.scm  run-tests-queue-new.scm

megatest-fossil-hash.scm : $(SRCFILES) megatest.scm *_records.scm
	echo "(define megatest-fossil-hash \"$(MTESTHASH)\")" > megatest-fossil-hash.new
	if ! diff -q megatest-fossil-hash.new megatest-fossil-hash.scm ; then echo copying .new to .scm;cp -f megatest-fossil-hash.new megatest-fossil-hash.scm;fi

;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

(declare (unit api))
(declare (uses rmt))
(declare (uses db))
(declare (uses tasks))

;; allow these queries through without starting a server
;;
(define api:read-only-queries
  '(get-key-val-pairs
    get-keys
    test-toplevel-num-items

    get-runs
    get-all-run-ids
    get-prev-run-ids
    get-run-ids-matching-target
    get-runs-by-patt
    get-steps-data
    login
    testmeta-get-record
    have-incompletes?
    synchash-get
    ))

(define api:write-queries
  '(
    ;; SERVERS
    start-server
    kill-server

    ;; TESTS
    test-set-state-status-by-id
    delete-test-records
    delete-old-deleted-test-records
    test-set-status-state
    test-set-top-process-pid
    roll-up-pass-fail-counts
    update-fail-pass-counts

    ;; RUNS
    register-run
    set-tests-state-status
    delete-run
    lock/unlock-run
    update-run-event_time
    mark-incomplete

    ;; STEPS
    teststep-set-status!

    ;; TEST DATA
    test-data-rollup
    csv->test-data

    ;; MISC
    sync-inmem->db

    ;; TESTMETA
    testmeta-add-record
    testmeta-update-field

    ;; TASKS
    tasks-add
    tasks-set-state-given-param-key
    ))

;; These are called by the server on recipt of /api calls
;;    - keep it simple, only return the actual result of the call, i.e. no meta info here
;;
;;    - returns #( flag result )
;;
(define (api:execute-requests dbstruct dat)
  (handle-exceptions
   exn
   (let ((call-chain (get-call-chain)))
     (print-call-chain (current-error-port))
     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))       
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (if (not (vector? dat))                    ;; it is an error to not receive a vector
       (vector #f #f "remote must be called with a vector")       
       (vector                                   ;; return a vector + the returned data structure
	#t 
	(let ((cmd    (vector-ref dat 0))
	      (params (vector-ref dat 1)))
	  (case (if (symbol? cmd)
		    cmd
		    (string->symbol cmd))

	    ;;===============================================
	    ;; READ/WRITE QUERIES
	    ;;===============================================

	    ;; SERVERS
	    ((start-server)                    (apply server:kind-run params))
	    ((kill-server)                     (set! *server-run* #f))

	    ;; TESTS
	    ((test-set-state-status-by-id)     (apply db:test-set-state-status-by-id dbstruct params))
	    ((delete-test-records)             (apply db:delete-test-records dbstruct params))
	    ((delete-old-deleted-test-records) (apply db:delete-old-deleted-test-records dbstruct params))
	    ((test-set-status-state)           (apply db:test-set-status-state dbstruct params))
	    ((test-set-top-process-pid)        (apply db:test-set-top-process-pid dbstruct params))
	    ((roll-up-pass-fail-counts)        (apply db:roll-up-pass-fail-counts dbstruct params))
	    ((update-fail-pass-counts)         (apply db:general-call dbstruct 'update-pass-fail-counts params))
	    ((test-set-archive-block-id)       (apply db:test-set-archive-block-id dbstruct params))

	    ;; RUNS
	    ((register-run)                 (apply db:register-run dbstruct params))
	    ((set-tests-state-status)       (apply db:set-tests-state-status dbstruct params))
	    ((delete-run)                   (apply db:delete-run dbstruct params))
	    ((lock/unlock-run)              (apply db:lock/unlock-run dbstruct params))
	    ((update-run-event_time)        (apply db:update-run-event_time dbstruct params))

	    ;; STEPS
	    ((teststep-set-status!)         (apply db:teststep-set-status! dbstruct params))

	    ;; TEST DATA
	    ((test-data-rollup)             (apply db:test-data-rollup dbstruct params))
	    ((csv->test-data)               (apply db:csv->test-data dbstruct params))

	    ;; MISC
	    ((sync-inmem->db)               (let ((run-id (car params)))
					      (db:sync-touched dbstruct run-id force-sync: #t)))
	    ((mark-incomplete)              (apply db:find-and-mark-incomplete dbstruct params))

	    ;; TESTMETA
	    ((testmeta-add-record)       (apply db:testmeta-add-record dbstruct params))
	    ((testmeta-update-field)     (apply db:testmeta-update-field dbstruct params))

	    ;; TASKS
	    ((tasks-add)                 (apply tasks:add dbstruct params))   
	    ((tasks-set-state-given-param-key) (apply tasks:set-state-given-param-key dbstruct params))

	    ;; ARCHIVES
	    ;; ((archive-get-allocations)   
	    ((archive-register-disk)     (apply db:archive-register-disk dbstruct params))
	    ((archive-register-block-name)(apply db:archive-register-block-name dbstruct params))
	    ((archive-allocate-testsuite/area-to-block)(apply db:archive-allocate-testsuite/area-to-block dbstruct block-id testsuite-name areakey))

	    ;;======================================================================
	    ;; READ ONLY QUERIES
	    ;;======================================================================

	    ;; KEYS
	    ((get-key-val-pairs)               (apply db:get-key-val-pairs dbstruct params))
	    ((get-keys)                        (db:get-keys dbstruct))

	    ;; TESTS
	    ((test-toplevel-num-items)         (apply db:test-toplevel-num-items dbstruct params))
	    ((get-test-info-by-id)	       (apply db:get-test-info-by-id dbstruct params))
	    ((test-get-rundir-from-test-id)    (apply db:test-get-rundir-from-test-id dbstruct params))

	    ((get-count-tests-running)         (apply db:get-count-tests-running dbstruct params))
	    ((get-count-tests-running-in-jobgroup) (apply db:get-count-tests-running-in-jobgroup dbstruct params))

	    ;; ((delete-test-step-records)        (apply db:delete-test-step-records dbstruct params))


	    ((get-previous-test-run-record)    (apply db:get-previous-test-run-record dbstruct params))
	    ((get-matching-previous-test-run-records)(apply db:get-matching-previous-test-run-records dbstruct params))
	    ((test-get-logfile-info)           (apply db:test-get-logfile-info dbstruct params))
	    ((test-get-records-for-index-file)  (apply db:test-get-records-for-index-file dbstruct params))
	    ((get-testinfo-state-status)       (apply db:get-testinfo-state-status dbstruct params))

	    ((test-get-top-process-pid)        (apply db:test-get-top-process-pid dbstruct params))
	    ((test-get-paths-matching-keynames-target-new) (apply db:test-get-paths-matching-keynames-target-new dbstruct params))
	    ((get-prereqs-not-met)             (apply db:get-prereqs-not-met dbstruct params))


	    ((get-count-tests-running-for-run-id) (apply db:get-count-tests-running-for-run-id dbstruct params))
	    ((synchash-get)                    (apply synchash:server-get dbstruct params))

	    ;; RUNS
	    ((get-run-info)                 (apply db:get-run-info dbstruct params))
	    ((get-run-status)               (apply db:get-run-status dbstruct params))
	    ((set-run-status)               (apply db:set-run-status dbstruct params))


	    ((get-tests-for-run)            (apply db:get-tests-for-run dbstruct params))
	    ((get-test-id)                  (apply db:get-test-id dbstruct params))
	    ((get-tests-for-run-mindata)    (apply db:get-tests-for-run-mindata dbstruct params))


	    ((get-runs)                     (apply db:get-runs dbstruct params))
	    ((get-all-run-ids)              (db:get-all-run-ids dbstruct))
	    ((get-prev-run-ids)             (apply db:get-prev-run-ids dbstruct params))
	    ((get-run-ids-matching-target)  (apply db:get-run-ids-matching-target dbstruct params))
	    ((get-runs-by-patt)             (apply db:get-runs-by-patt dbstruct params))
	    ((get-run-name-from-id)         (apply db:get-run-name-from-id dbstruct params))



	    ;; STEPS





	    ((get-steps-data)               (apply db:get-steps-data dbstruct params))

	    ;; MISC
	    ((have-incompletes?)            (apply db:have-incompletes? dbstruct params))
	    ((login)                        (apply db:login dbstruct params))
	    ((general-call)                 (let ((stmtname   (car params))
						  (run-id     (cadr params))
						  (realparams (cddr params)))
					      (db:with-db dbstruct run-id #t ;; these are all for modifying the db
							  (lambda (db)
							    (db:general-call db stmtname realparams)))))

	    ((sdb-qry)                      (apply sdb:qry params))
	    ((ping)                         (current-process-id))

	    ;; TESTMETA
	    ((testmeta-get-record)       (apply db:testmeta-get-record dbstruct params))

	    ;; TASKS 
	    ((find-task-queue-records)   (apply tasks:find-task-queue-records dbstruct params))))))))



;; http-server  send-response
;;                 api:process-request
;;                    db:*
;;
;; NB// Runs on the server as part of the server loop
;;
(define (api:process-request dbstruct $) ;; the $ is the request vars proc
  (let* ((cmd     ($ 'cmd))
	 (paramsj ($ 'params))
	 (params  (db:string->obj paramsj transport: 'http)) ;; (rmt:json-str->dat paramsj))
	 (resdat  (api:execute-requests dbstruct (vector cmd params))) ;; #( flag result )
	 (res     (vector-ref resdat 1)))

    ;; This can be here but needs controls to ensure it doesn't run more than every 4 seconds
    ;; (rmt:dat->json-str
    ;;  (if (or (string? res)
    ;;          (list?   res)
    ;;          (number? res)
    ;;          (boolean? res))
    ;;      res 
    ;;      (list "ERROR, not string, list, number or boolean" 1 cmd params res)))))
    (db:obj->string res transport: 'http)))

;; Copyright 2006-2014, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')

(use sqlite3 srfi-1 posix regex regex-case srfi-69 dot-locking format md5 message-digest)
(import (prefix sqlite3 sqlite3:))

(declare (unit archive))
(declare (uses db))
(declare (uses common))

(include "common_records.scm")
(include "db_records.scm")

;;======================================================================
;; 
;;======================================================================

;; NOT CURRENTLY USED
;;
(define (archive:main linktree target runname testname itempath options)
  (let ((testdir (conc linktree "/" target "/" runname "/" testname "/" itempatt))
	(flavor  'plain) ;; type of machine to run jobs on
	(maxload 1.5)   ;; max allowed load for this work
	(adisks  (archive:get-archive-disks)))
    ;; get testdir size
    ;;   - hand off du to job mgr
    (if (and (file-exists? testdir)
	     (file-is-writable? testdir))
	(let* ((dused  (jobrunner:run-job 
			flavor  ;; machine type
			maxload ;; max allowed load
			'()     ;; prevars - environment vars to set for the job
			common:get-disk-space-used  ;; if a proc call it, if a string it is a unix command
			(list testdir)))
	       (apath  (archive:get-archive testname itempath dused)))
	  (jobrunner:run-job
	   flavor
	   maxload
	   '()
	   archive:run-bup
	   (list testdir apath))))))
	  
;; Get archive disks from megatest.config
;;
(define (archive:get-archive-disks)
  (let ((section (configf:get-section *configdat* "archive-disks")))
    (if section
	section
	'())))

;; look for the best candidate archive area, else create new 
;; area
;;
(define (archive:get-archive testname itempath dused)
  ;; look up in archive_allocations if there is a pre-used archive
  ;; with adequate diskspace
  ;;
  (let* ((existing-blocks (rmt:archive-get-allocations testname itempath dused))
	 (candidate-disks (map (lambda (block)
				 (list
				  (vector-ref block 1)   ;; archive-area-name
				  (vector-ref block 2))) ;; disk-path
			       existing-blocks)))
    (or (common:get-disk-with-most-free-space candidate-disks dused)
	(archive:allocate-new-archive-block testname itempath))))

;; allocate a new archive area
;;
(define (archive:allocate-new-archive-block run-area-home testsuite-name dneeded)
  (let* ((adisks    (archive:get-archive-disks))
	 (best-disk (common:get-disk-with-most-free-space adisks dneeded)))
    (if best-disk
	(let* ((bdisk-name    (car best-disk))
	       (bdisk-path    (cdr best-disk))
	       (area-key      (substring (message-digest-string (md5-primitive) run-area-home) 0 5))
	       (bdisk-id      (rmt:archive-register-disk bdisk-name bdisk-path (get-df bdisk-path)))
	       (archive-name  (let ((sec (current-seconds)))
				(conc (time->string (seconds->local-time sec) "%Y")
				      "_q" (seconds->quarter sec) "/"
				      testsuite-name "_" area-key)))
	       (archive-path  (conc bdisk-path "/" archive-name))
	       (block-id      (rmt:archive-register-block-name bdisk-id archive-path)))
	  ;;   (allocation-id (rmt:archive-allocate-testsuite/area-to-block block-id testsuite-name area-key)))
	  (if block-id ;; (and block-id allocation-id)
	      (cons block-id archive-path)
	      #f))
	#f)))

;; archive - run bup
;;
;; 1. create the bup dir if not exists
;; 2. start the du of each directory
;; 3. gen index
;; 4. save
;;
(define (archive:run-bup archive-command run-id run-name tests)
  ;; move the getting of archive space down into the below block so that a single run can 
  ;; allocate as needed should a disk fill up
  ;;
  (let* ((min-space    (string->number (or (configf:lookup *configdat* "archive" "minspace") "1000")))
	 (archive-info (archive:allocate-new-archive-block *toppath* (common:get-testsuite-name) min-space))
	 (archive-dir  (if archive-info (cdr archive-info) #f))
	 (archive-id   (if archive-info (car archive-info) -1))
	 (disk-groups  (make-hash-table))
	 (test-groups  (make-hash-table)) ;; these two (disk and test groups) could be combined nicely
	 (bup-exe      (or (configf:lookup *configdat* "archive" "bup") "bup"))
	 (compress     (or (configf:lookup *configdat* "archive" "compress") "9"))
	 (linktree     (configf:lookup *configdat* "setup" "linktree")))

    (if (not archive-dir) ;; no archive disk found, this is fatal
	(begin
	  (debug:print 0 "FATAL: No archive disks found. Please add disks with at least " min-space " MB space to the [archive-disks] section of megatest.config")
	  (debug:print 0 "       use [archive] minspace to specify minimum available space")
	  (debug:print 0 "   disks: " (string-intersperse (map cadr (archive:get-archive-disks)) "\n         "))
	  (exit 1))
	(debug:print-info 0 "Using path " archive-dir " for archiving"))

    ;; from the test info bin the path to the test by stem
    ;;
    (for-each
     (lambda (test-dat)
       (let* ((item-path         (db:test-get-item-path test-dat))
	      (test-name         (db:test-get-testname  test-dat))
	      (test-id           (db:test-get-id        test-dat))
	      (run-id            (db:test-get-run_id    test-dat))
	      (target            (string-intersperse (map cadr (rmt:get-key-val-pairs run-id)) "/"))
	      
	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (runs:make-full-test-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      (test-physical-path (if (file-exists? test-path) (read-symbolic-link test-path #t) #f))
	      (partial-path-index (if test-physical-path (substring-index test-partial-path test-physical-path) #f))
	      (test-base         (if (and partial-path-index 
					  test-physical-path )
				     (substring test-physical-path
						0
						partial-path-index)
				     #f)))
	 
 	 (if (or toplevel/children
		 (not (file-exists? test-path)))
	     #f
	     (begin
	       (debug:print 0
			    "From test-dat=" test-dat " derived the following:\n"
			    "test-partial-path  = " test-partial-path "\n"
			    "test-path          = " test-path "\n"
			    "test-physical-path = " test-physical-path "\n"
			    "partial-path-index = " partial-path-index "\n"
			    "test-base          = " test-base)
	       (hash-table-set! disk-groups test-base (cons test-physical-path (hash-table-ref/default disk-groups test-base '())))
	       (hash-table-set! test-groups test-base (cons test-dat (hash-table-ref/default test-groups test-base '())))
	       test-path))))
     tests)
    ;; for each disk-group
    (for-each 
     (lambda (disk-group)
       (debug:print 0 "Processing disk-group " disk-group)
       (let* ((test-paths (hash-table-ref disk-groups disk-group))
	      ;; ((string-intersperse (map cadr (rmt:get-key-val-pairs 1)) "-")
	      (bup-init-params  (list "-d" archive-dir "init"))
	      (bup-index-params (append (list "-d" archive-dir "index") test-paths))
	      (bup-save-params  (append (list "-d" archive-dir "save" ;; (conc "--strip-path=" linktree)
					      (conc "-" compress) ;; or (conc "--compress=" compress)
					      "-n" (conc (common:get-testsuite-name) "-" run-id)
					      (conc "--strip-path=" disk-group))
					test-paths))
	      (print-prefix      #f)) ;; "Running: ")) ;; change to #f to turn off printing
	 (if (not (file-exists? archive-dir))
	     (create-directory archive-dir #t))
	 (if (not (file-exists? (conc archive-dir "/HEAD")))
	     (begin
	       ;; replace this with jobrunner stuff enventually
	       (debug:print-info 0 "Init bup in " archive-dir)
	       (run-n-wait bup-exe params: bup-init-params print-cmd: print-prefix)))
	 (debug:print-info 0 "Indexing data to be archived")
	 (run-n-wait bup-exe params: bup-index-params print-cmd: print-prefix)
	 (debug:print-info 0 "Archiving data with bup")
	 (run-n-wait bup-exe params: bup-save-params print-cmd: print-prefix)
	 (for-each
	  (lambda (test-dat)
	    (let ((test-id           (db:test-get-id        test-dat))
		  (run-id            (db:test-get-run_id    test-dat)))
	      (rmt:test-set-archive-block-id run-id test-id archive-id)
	      (if (member archive-command '("save-remove"))
		  (runs:remove-test-directory test-dat 'archive-remove))))
	  (hash-table-ref test-groups disk-group))))
     (hash-table-keys disk-groups))
    #t))

(define (archive:bup-restore archive-command run-id run-name tests)  ;; move the getting of archive space down into the below block so that a single run can 
  ;; allocate as needed should a disk fill up
  ;;
  (let* ((disk-groups  (make-hash-table))
	 (test-groups  (make-hash-table)) ;; these two (disk and test groups) could be combined nicely
	 (bup-exe      (or (configf:lookup *configdat* "archive" "bup") "bup"))
	 (linktree     (configf:lookup *configdat* "setup" "linktree")))

    ;; from the test info bin the path to the test by stem
    ;;
    (for-each
     (lambda (test-dat)
       (let* ((item-path         (db:test-get-item-path test-dat))
	      (test-name         (db:test-get-testname  test-dat))
	      (test-id           (db:test-get-id        test-dat))
	      (run-id            (db:test-get-run_id    test-dat))
	      (target            (string-intersperse (map cadr (rmt:get-key-val-pairs run-id)) "/"))
	      
	      (toplevel/children (and (db:test-get-is-toplevel test-dat)
				      (> (rmt:test-toplevel-num-items run-id test-name) 0)))
	      (test-partial-path (conc target "/" run-name "/" (runs:make-full-test-name test-name item-path)))
	      ;; note the trailing slash to get the dir inspite of it being a link
	      (test-path         (conc linktree "/" test-partial-path))
	      (test-physical-path (if (file-exists? test-path) (read-symbolic-link test-path #t) #f))
	      (partial-path-index (if test-physical-path (substring-index test-partial-path test-physical-path) #f))
	      (test-base         (if (and partial-path-index 
					  test-physical-path )
				     (substring test-physical-path
						0
						partial-path-index)
				     #f)))
	 
	 ;; CREATE WORK AREA
	 (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat #!key (remtries 2))
	 
	 ;; DO BUP RESTORE



 	 (if (or toplevel/children
		 (not (file-exists? test-path)))
	     #f
	     (begin
	       (debug:print 0
			    "From test-dat=" test-dat " derived the following:\n"
			    "test-partial-path  = " test-partial-path "\n"
			    "test-path          = " test-path "\n"
			    "test-physical-path = " test-physical-path "\n"
			    "partial-path-index = " partial-path-index "\n"
			    "test-base          = " test-base)
	       (hash-table-set! disk-groups test-base (cons test-physical-path (hash-table-ref/default disk-groups test-base '())))
	       (hash-table-set! test-groups test-base (cons test-dat (hash-table-ref/default test-groups test-base '())))
	       test-path))))
     tests)
    ;; for each disk-group
    (for-each 
     (lambda (disk-group)
       (debug:print 0 "Processing disk-group " disk-group)
       (let* ((test-paths (hash-table-ref disk-groups disk-group))
	      ;; ((string-intersperse (map cadr (rmt:get-key-val-pairs 1)) "-")
	      (bup-init-params  (list "-d" archive-dir "init"))
	      (bup-index-params (append (list "-d" archive-dir "index") test-paths))
	      (bup-save-params  (append (list "-d" archive-dir "save" ;; (conc "--strip-path=" linktree)
					      (conc "-" compress) ;; or (conc "--compress=" compress)
					      "-n" (conc (common:get-testsuite-name) "-" run-id)
					      (conc "--strip-path=" disk-group))
					test-paths))
	      (print-prefix      #f)) ;; "Running: ")) ;; change to #f to turn off printing
	 (if (not (file-exists? archive-dir))
	     (create-directory archive-dir #t))
	 (if (not (file-exists? (conc archive-dir "/HEAD")))
	     (begin
	       ;; replace this with jobrunner stuff enventually
	       (debug:print-info 0 "Init bup in " archive-dir)
	       (run-n-wait bup-exe params: bup-init-params print-cmd: print-prefix)))
	 (debug:print-info 0 "Indexing data to be archived")
	 (run-n-wait bup-exe params: bup-index-params print-cmd: print-prefix)
	 (debug:print-info 0 "Archiving data with bup")
	 (run-n-wait bup-exe params: bup-save-params print-cmd: print-prefix)
	 (for-each
	  (lambda (test-dat)
	    (let ((test-id           (db:test-get-id        test-dat))
		  (run-id            (db:test-get-run_id    test-dat)))
	      (rmt:test-set-archive-block-id run-id test-id archive-id)
	      (if (member archive-command '("save-remove"))
		  (runs:remove-test-directory test-dat 'archive-remove))))
	  (hash-table-ref test-groups disk-group))))
     (hash-table-keys disk-groups))
    #t))

;; Not currently used! But, I think it *should* be used!!!
(define (client:logout serverdat)
  (let ((ok (and (socket? serverdat)
		 (cdb:logout serverdat *toppath* (client:get-signature)))))
    ok))

(define (client:connect iface port)
  (case (server:get-transport)
    ((rpc)  (rpc:client-connect  iface port))
    ((http) (http:client-connect iface port))
    ((zmq)  (zmq:client-connect  iface port))
    (else   (rpc:client-connect  iface port))))

(define (client:setup  run-id #!key (remaining-tries 10) (failed-connects 0))
  (case (server:get-transport)
    ((rpc) (rpc-transport:client-setup run-id)) ;;(client:setup-rpc run-id))
    ((http)(client:setup-http run-id))
    (else  (rpc-transport:client-setup run-id)))) ;; (client:setup-rpc run-id))))

;; (define (client:login-no-auto-setup server-info run-id)
;;   (case (server:get-transport)
;;     ((rpc)  (rpc:login-no-auto-client-setup server-info run-id))
;;     ((http) (rmt:login-no-auto-client-setup server-info run-id))
;;     (else   (rpc:login-no-auto-client-setup server-info run-id))))
;; 
;; (define (client:setup-rpc run-id)
;;   (debug:print 0 "INFO: client:setup remaining-tries=" remaining-tries)
;;   (if (<= remaining-tries 0)
;;       (begin
;; 	(debug:print 0 "ERROR: failed to start or connect to server for run-id " run-id)
;; 	(exit 1))
;;       (let ((host-info (hash-table-ref/default *runremote* run-id #f)))
;; 	(debug:print-info 0 "client:setup host-info=" host-info ", remaining-tries=" remaining-tries)
;; 	(if host-info
;; 	    (let* ((iface     (car  host-info))
;; 		   (port      (cadr host-info))
;; 		   (start-res (client:connect iface port))
;; 		   ;; (ping-res  (server:ping-server run-id iface port))
;; 		   (ping-res  (client:login-no-auto-setup start-res run-id)))
;; 	      (if ping-res   ;; sucessful login?
;; 		  (begin
;; 		    (hash-table-set! *runremote* run-id start-res)
;; 		    start-res)  ;; return the server info
;; 		  (if (member remaining-tries '(3 4 6))
;; 		      (begin    ;; login failed
;; 			(debug:print 25 "INFO: client:setup start-res=" start-res ", run-id=" run-id ", server-dat=" host-info)
;; 			(hash-table-delete! *runremote* run-id)
;; 			(open-run-close tasks:server-force-clean-run-record
;; 			 		tasks:open-db
;; 			 		run-id 
;; 			 		(car  host-info)
;; 			 		(cadr host-info)
;; 					" client:setup (host-info=#t)")
;; 			(thread-sleep! 5)
;; 			(client:setup run-id remaining-tries: 10)) ;; (- remaining-tries 1)))
;; 		      (begin
;; 			(debug:print 25 "INFO: client:setup failed to connect, start-res=" start-res ", run-id=" run-id ", host-info=" host-info)
;; 			(thread-sleep! 5)
;; 			(client:setup run-id remaining-tries: (- remaining-tries 1))))))
;; 	    ;; YUK: rename server-dat here
;; 	    (let* ((server-dat (open-run-close tasks:get-server tasks:open-db run-id)))
;; 	      (debug:print-info 0 "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
;; 	      (if server-dat
;; 		  (let* ((iface     (tasks:hostinfo-get-interface server-dat))
;; 			 (port      (tasks:hostinfo-get-port      server-dat))
;; 			 (start-res (http-transport:client-connect iface port))
;; 			 ;; (ping-res  (server:ping-server run-id iface port))
;; 			 (ping-res  (rmt:login-no-auto-client-setup start-res run-id)))
;; 		    (if start-res
;; 			(begin
;; 			  (hash-table-set! *runremote* run-id start-res)
;; 			  start-res)
;; 			(if (member remaining-tries '(2 5))
;; 			    (begin    ;; login failed
;; 			      (debug:print 25 "INFO: client:setup start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
;; 			      (hash-table-delete! *runremote* run-id)
;; 			      (open-run-close tasks:server-force-clean-run-record
;; 					      tasks:open-db
;; 					      run-id 
;; 					      (tasks:hostinfo-get-interface server-dat)
;; 					      (tasks:hostinfo-get-port      server-dat)
;; 					      " client:setup (server-dat = #t)")
;; 			      (thread-sleep! 2)
;; 			      (server:try-running run-id)
;; 			      (thread-sleep! 10) ;; give server a little time to start up
;; 			      (client:setup run-id remaining-tries: 10)) ;; (- remaining-tries 1)))
;; 			    (begin
;; 			      (debug:print 25 "INFO: client:setup start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
;; 			      (thread-sleep! 5)
;; 			      (client:setup run-id remaining-tries: (- remaining-tries 1))))))
;; 		  (begin    ;; no server registered
;; 		    (if (eq? remaining-tries 2)
;; 			(begin
;; 			  ;; (open-run-close tasks:server-clean-out-old-records-for-run-id tasks:open-db run-id " client:setup (server-dat=#f)")
;; 			  (client:setup run-id remaining-tries: 10))
;; 			(begin
;; 			  (thread-sleep! 2) 
;; 			  (debug:print 25 "INFO: client:setup start-res (not defined here), run-id=" run-id ", server-dat=" server-dat)
;; 			  (if (< (open-run-close tasks:num-in-available-state tasks:open-db run-id) 3)
;; 			      (begin
;; 				;; (open-run-close tasks:server-clean-out-old-records-for-run-id tasks:open-db run-id " client:setup (server-dat=#f)")
;; 				(server:try-running run-id)))
;; 			  (thread-sleep! 10) ;; give server a little time to start up
;; 			  (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))))

;; Do all the connection work, look up the transport type and set up the
;; connection if required.
;;
;; There are two scenarios. 
;;   1. We are a test manager and we received *transport-type* and *runremote* via cmdline
;;   2. We are a run tests, list runs or other interactive process and we must figure out
;;      *transport-type* and *runremote* from the monitor.db
;;
;; client:setup
;;
;; lookup_server, need to remove *runremote* stuff
;;
(define (client:setup-http run-id #!key (remaining-tries 10) (failed-connects 0))
  (debug:print-info 2 "client:setup remaining-tries=" remaining-tries)
  (let* ((tdbdat (tasks:open-db)))
    (if (<= remaining-tries 0)
	(begin
	  (debug:print 0 "ERROR: failed to start or connect to server for run-id " run-id)
	  (exit 1))























	(let* ((server-dat (tasks:get-server (db:delay-if-busy tdbdat) run-id)))
	  (debug:print-info 4 "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	  (if server-dat
	      (let* ((iface     (tasks:hostinfo-get-interface server-dat))
		     (hostname  (tasks:hostinfo-get-hostname  server-dat))
		     (port      (tasks:hostinfo-get-port      server-dat))
		     (start-res (case *transport-type*
				  ((http)(http-transport:client-connect iface port))
				  ((nmsg)(nmsg-transport:client-connect hostname port))))
		     (ping-res  (case *transport-type* 
				  ((http)(rmt:login-no-auto-client-setup start-res run-id))
				  ((nmsg)(let ((logininfo (rmt:login-no-auto-client-setup start-res run-id)))
 					   (if logininfo
 					       (car (vector-ref logininfo 1))
 					       #f))))))
		(if (and start-res
			 ping-res)
		    (begin
		      (hash-table-set! *runremote* run-id start-res)
		      (debug:print-info 2 "connected to " (http-transport:server-dat-make-url start-res))
		      start-res)
		    (begin    ;; login failed but have a server record, clean out the record and try again
		      (debug:print-info 0 "client:setup, login failed, will attempt to start server ... start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
		      (case *transport-type* 
			((http)(http-transport:close-connections run-id)))
		      (hash-table-delete! *runremote* run-id)
		      (tasks:kill-server-run-id run-id)
		      (tasks:server-force-clean-run-record (db:delay-if-busy tdbdat)
							   run-id 
							   (tasks:hostinfo-get-interface server-dat)
							   (tasks:hostinfo-get-port      server-dat)
							   " client:setup (server-dat = #t)")
		      (if (> remaining-tries 8)
			  (thread-sleep! (+ 1 (random 5))) ;; spread out the starts a little
			  (thread-sleep! (+ 15 (random 20)))) ;; it isn't going well. give it plenty of time
		      (server:try-running run-id)
		      (thread-sleep! 5)   ;; give server a little time to start up
		      (client:setup run-id remaining-tries: (- remaining-tries 1))
		      )))
	      (begin    ;; no server registered
		(let ((num-available (tasks:num-in-available-state (db:dbdat-get-db tdbdat) run-id)))
		  (debug:print-info 0 "client:setup, no server registered, remaining-tries=" remaining-tries " num-available=" num-available)

		  (if (< num-available 2)

		      (server:try-running run-id))
		  (thread-sleep! (+ 5 (random (- 20 remaining-tries))))  ;; give server a little time to start up, randomize a little to avoid start storms.
		  (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))

;; keep this as a function to ease future 
(define (client:start run-id server-info)
  (http-transport:client-connect (tasks:hostinfo-get-interface server-info)
				 (tasks:hostinfo-get-port server-info)))

;; client:signal-handler

(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*    'http)
(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg
(define *runremote*         (make-hash-table)) ;; if set up for server communication this will hold <host port>
(define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
(define *client-non-blocking-mode* #f)
(define *server-id*         #f)
(define *server-info*       #f)
(define *time-to-exit*      #f)

  '((0 "COMPLETED")
    (1 "NOT_STARTED")
    (2 "RUNNING")
    (3 "REMOTEHOSTSTART")
    (4 "LAUNCHED")
    (5 "KILLED")
    (6 "KILLREQ")
    (7 "STUCK")
    (8 "ARCHIVED")))

(define *common:std-statuses*
  '((0 "PASS")
    (1 "WARN")
    (2 "FAIL")
    (3 "CHECK")
    (4 "n/a")
    (5 "WAIVED")
    (6 "SKIP")
    (7 "DELETED")
    (8 "STUCK/DEAD")
    (9 "ABORT")))

;; These are stopping conditions that prevent a test from being run
(define *common:cant-run-states-sym* 
  '(COMPLETED KILLED WAIVED UNKNOWN INCOMPLETE ABORT ARCHIVED))

;;======================================================================
;; D E B U G G I N G   S T U F F 
;;======================================================================

(define *verbosity*         1)
(define *logging*           #f)

  (debug:print-info 2 "starting exit process, finalizing databases.")
  (rmt:print-db-stats)
  (let ((run-ids (hash-table-keys *db-local-sync*)))
    (if (and (not (null? run-ids))
	     (configf:lookup *configdat* "setup" "megatest-db"))
	(db:multi-db-sync run-ids 'new2old)))
  (if *dbstruct-db* (db:close-all *dbstruct-db*))
  (if *inmemdb*     (db:close-all *inmemdb*))
  (if (and *megatest-db*
	   (sqlite3:database? *megatest-db*))
      (begin
	(sqlite3:interrupt! *megatest-db*)
	(sqlite3:finalize! *megatest-db* #t)
	(set! *megatest-db* #f)))
  (if *task-db*     (let ((db (cdr *task-db*)))

;; return a nice clean pathname made absolute
(define (nice-path dir)
  (normalize-pathname (if (absolute-pathname? dir)
			  dir
			  (conc (current-directory) "/" dir))))















(define (get-cpu-load)
  (car (common:get-cpu-load)))
;;   (let* ((load-res (cmd-run->list "uptime"))
;; 	 (load-rx  (regexp "load average:\\s+(\\d+)"))
;; 	 (cpu-load #f))
;;     (for-each (lambda (l)
;; 		(let ((match (string-search load-rx l)))

(define (get-uname . params)
  (let* ((uname-res (cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
	 (uname #f))
    (if (null? (car uname-res))
	"unknown"
	(caar uname-res))))

;;======================================================================
;; D I S K   S P A C E 
;;======================================================================

(define (common:get-disk-space-used fpath)
  (with-input-from-pipe (conc "/usr/bin/du -s " fpath) read))

(define (get-df path)
  (let* ((df-results (cmd-run->list (conc "df " path)))
	 (space-rx   (regexp "([0-9]+)\\s+([0-9]+)%"))
	 (freespc    #f))
    ;; (write df-results)
    (for-each (lambda (l)
		(let ((match (string-search space-rx l)))
		  (if match 
		      (let ((newval (string->number (cadr match))))
			(if (number? newval)
			    (set! freespc newval))))))
	      (car df-results))
    freespc))
  
;; paths is list of lists ((name path) ... )
;;
(define (common:get-disk-with-most-free-space disks minsize)
  (let ((best     #f)
	(bestsize 0))
    (for-each 
     (lambda (disk-num)
       (let* ((dirpath    (cadr (assoc disk-num disks)))
	      (freespc    (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 50 "disks not a dir " disk-num)
				(debug:print 0 "WARNING: disk " disk-num " at path " dirpath " is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 50 "disks not writeable " disk-num)
				(debug:print 0 "WARNING: disk " disk-num " at path " dirpath " is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 50 "disks not a proper path " disk-num)
				(debug:print 0 "WARNING: disk " disk-num " at path " dirpath " is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-df dirpath)))))
	 (if (> freespc bestsize)
	     (begin
	       (set! best     (cons disk-num dirpath))
	       (set! bestsize freespc)))))
     (map car disks))
    (if (and best (> bestsize minsize))
	best
	#f))) ;; #f means no disk candidate found

;;======================================================================
;; E N V I R O N M E N T   V A R S
;;======================================================================
	      
(define (save-environment-as-files fname #!key (ignorevars (list "USER" "HOME" "DISPLAY" "LS_COLORS" "XKEYSYMDB" "EDITOR" "MAKEFLAGS" "MAKEF")))
  (let ((envvars (get-environment-variables))
        (whitesp (regexp "[^a-zA-Z0-9_\\-:,.\\/%$]")))
     (with-output-to-file (conc fname ".csh")
       (lambda ()
          (for-each (lambda (keyval)

	       (lambda (sheet-name)
		 (let* ((dat-path  (conc refdb-path "/" sheet-name ".dat"))
			(ref-dat   (configf:read-file dat-path #f #t))
			(ref-assoc (map (lambda (key)
					  (list key (hash-table-ref ref-dat key)))
					(hash-table-keys ref-dat))))
				   ;; (hash-table->alist ref-dat)))
		   ;; (set! data (append data (list (list sheet-name ref-assoc))))))
		   (set! data (cons (list sheet-name ref-assoc) data))))
	       sheets)
	      (list data "NO ERRORS"))))))

;; map over all pairs in a three level hierarchial alist and apply a function to the keys/val
;;
(define (configf:map-all-hier-alist data proc #!key (initproc1 #f)(initproc2 #f)(initproc3 #f))
  (for-each 

	    (for-each
	     (lambda (varname)
	       (let* ((valtmp (assoc varname sectiondat))
		      (val    (if valtmp (cadr valtmp) "")))
		 (proc sheetname sectionname varname val)))
	     (map car sectiondat))))
	(map car sheetdat))))
   (map car data))
  data)

;;======================================================================
;;  C O N F I G   T O / F R O M   A L I S T
;;======================================================================

(define (configf:config->alist cfgdat)
  (hash-table->alist cfgdat))

(define (configf:alist->config adat)
  (let ((ht (make-hash-table)))
    (for-each
     (lambda (section)
       (hash-table-set! ht (car section)(cdr section)))
     adat)
    ht))

(define (configf:read-alist fname)
  (configf:alist->config
   (with-input-from-file fname read)))

(define (configf:write-alist cdat fname)
  (with-output-to-file fname
    (lambda ()
      (pp (configf:config->alist cdat)))))
     

			   (itempath   (db:test-get-item-path test))
			   (testfullname (test:test-get-fullname test))
			   (teststatus (db:test-get-status   test))
			   (teststate  (db:test-get-state    test))
			   ;;(teststart  (db:test-get-event_time test))
			   ;;(runtime    (db:test-get-run_duration test))
			   (buttontxt  (cond
					((member teststate '("COMPLETED" "ARCHIVED")) teststatus)
					((and (equal? teststate "NOT_STARTED")
					      (member teststatus '("ZERO_ITEMS" "BLOCKED" "PREQ_FAIL" "PREQ_DISCARDED" "TIMED_OUT" "KEEP_TRYING" "TEN_STRIKES")))
					 teststatus)
					(else
					 teststate)))
			   (button     (vector-ref columndat rown))
			   (color      (car (gutils:get-color-for-state-status teststate teststatus)))

  (let* ((modtime         (dashboard:get-youngest-run-db-mod-time)) ;; (file-modification-time *db-file-path*))
	 (monitor-modtime (if (file-exists? *monitor-db-path*)
			      (file-modification-time *monitor-db-path*)
			      -1))
	 (run-update-time (current-seconds))
	 (recalc          (dashboard:recalc modtime *please-update-buttons* *last-db-update-time*)))
    (if (and (eq? *current-tab-number* 0)
	     (or (> monitor-modtime *last-monitor-update-time*)
		 (> (- run-update-time *last-monitor-update-time*) 5))) ;; update every 1/2 minute just in case
	(begin
	  (set! *last-monitor-update-time* run-update-time) ;; monitor-modtime)
	  (if dashboard:update-servers-table (dashboard:update-servers-table))))
    (if recalc
	(begin	
	  (case *current-tab-number* 
	    ((0) 
	     (if dashboard:update-summary-tab (dashboard:update-summary-tab)))
	    ((1) ;; The runs table is active

	  ;; (dbr:dbstruct-set-run-id! dbstruct run-id)
	  (if local
	      (begin
		(dbr:dbstruct-set-localdb! dbstruct run-id db) ;; (dbr:dbstruct-set-inmem! dbstruct db) ;; direct access ...
		db)
	      (begin
		(dbr:dbstruct-set-inmem!  dbstruct inmem)
		;; dec 14, 2014 - keep deleted records available. hunch is that they are needed for id placeholders
		;; (sqlite3:execute db "DELETE FROM tests WHERE state='DELETED';") ;; they just slow us down in this context
		(db:sync-tables db:sync-tests-only db inmem)
		(db:delay-if-busy refdb) ;; dbpath: (db:dbdat-get-path refdb)) ;; What does delaying here achieve? 
		(dbr:dbstruct-set-refdb!  dbstruct refdb)
		(db:sync-tables db:sync-tests-only inmem refdb) ;; use inmem as the reference, don't read again from db
		;; sync once more to deal with delays?
		;; (db:sync-tables db:sync-tests-only db inmem)
		;; (db:sync-tables db:sync-tests-only inmem refdb)
		inmem))))))

;; This routine creates the db. It is only called if the db is not already ls opened
;;
(define (db:open-main dbstruct) ;;  (conc *toppath* "/megatest.db") (car *configinfo*)))
  (let ((mdb (dbr:dbstruct-get-main dbstruct)))
    (if mdb

	(inmem  (dbr:dbstruct-get-inmem dbstruct))
	(maindb (dbr:dbstruct-get-main  dbstruct))
	(refdb  (dbr:dbstruct-get-refdb dbstruct))
	(olddb  (dbr:dbstruct-get-olddb dbstruct))
	;; (runid  (dbr:dbstruct-get-run-id dbstruct))
	)
    (debug:print-info 4 "Syncing for run-id: " run-id)
    ;; (mutex-lock! *http-mutex*)
    (if (eq? run-id 0)
	;; runid equal to 0 is main.db
	(if maindb
	    (if (or (not (number? mtime))
		    (not (number? stime))
		    (> mtime stime)
		    force-sync)

	(if (or (not (number? mtime))
		(not (number? stime))
		(> mtime stime)
		force-sync)
	    (begin
	      (db:delay-if-busy rundb)
	      (db:delay-if-busy olddb)

	      (dbr:dbstruct-set-stime! dbstruct (current-milliseconds))
	      (let ((num-synced (db:sync-tables db:sync-tests-only inmem refdb rundb olddb)))
		;; (mutex-unlock! *http-mutex*)
		num-synced)
	      (begin
		;; (mutex-unlock! *http-mutex*)
		0))))))

(define (db:close-main dbstruct)
  (let ((maindb (dbr:dbstruct-get-main dbstruct)))
    (if maindb
	(begin
	  (sqlite3:finalize! (db:dbdat-get-db maindb))

     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
     (print "exn=" (condition->list exn))
     (debug:print 0 " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
     (debug:print 0 " src db:  " (db:dbdat-get-path fromdb))
     (for-each (lambda (dbdat)
		 (debug:print 0 " dbpath:  " (db:dbdat-get-path dbdat)))
	       (cons todb slave-dbs))
     (if *server-run* ;; we are inside a server, throw a sync-failed error
	 (signal (make-composite-condition
		 (make-property-condition 'sync-failed 'message "db:sync-tables failed in a server context.")))))

	 ;; (set! *time-to-exit* #t) ;; let watch dog know that it is time to die.
	 ;; (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "shutting-down")
	 ;; (portlogger:open-run-close portlogger:set-port port "released")
	 ;; (exit 1)))
   (cond
    ((not fromdb) (debug:print 3 "WARNING: db:sync-tables called with fromdb missing") -1)
    ((not todb)   (debug:print 3 "WARNING: db:sync-tables called with todb missing") -2)
    ((not (sqlite3:database? (db:dbdat-get-db fromdb)))
     (debug:print 0 "ERROR: db:sync-tables called with fromdb not a database " fromdb) -3)
    ((not (sqlite3:database? (db:dbdat-get-db todb)))
     (debug:print 0 "ERROR: db:sync-tables called with todb not a database " todb) -4)

	       (if (> (length fromdat) batch-len)
		   (begin
		     (set! fromdats (cons fromdat fromdats))
		     (set! fromdat  '())
		     (set! totrecords (+ totrecords 1)))))
	     (db:dbdat-get-db fromdb)
	     full-sel)
	    
	    ;; tack on remaining records in fromdat
	    (if (not (null? fromdat))
		(set! fromdats (cons fromdat fromdats)))

	    (debug:print-info 2 "found " totrecords " records to sync")

	    ;; read the target table
	    (sqlite3:for-each-row
	     (lambda (a . b)
	       (hash-table-set! todat a (apply vector a b)))

;;
;;  run-ids: '(1 2 3 ...) or #f (for all)
;;
(define (db:multi-db-sync run-ids . options)
  (let* ((toppath  (launch:setup-for-run))
	 (dbstruct (if toppath (make-dbr:dbstruct path: toppath) #f))
	 (mtdb     (if toppath (db:open-megatest-db)))
	 (allow-cleanup (if run-ids #f #t))
	 (run-ids  (if run-ids 
		       run-ids
		       (if toppath (begin
				     (db:delay-if-busy mtdb)
				     (db:get-all-run-ids mtdb)))))
	 (tdbdat  (tasks:open-db))
	 (servers (tasks:get-all-servers (db:delay-if-busy tdbdat))))

		   (dbstruct (if toppath (make-dbr:dbstruct path: toppath local: #t) #f)))
	       (debug:print 0 "INFO: Propagating " (length testrecs) " records for run-id=" run-id " to run specific db")
	       (db:replace-test-records dbstruct run-id testrecs)
	       (sqlite3:finalize! (db:dbdat-get-db (dbr:dbstruct-get-rundb dbstruct)))))
	   run-ids)))

    ;; now ensure all newdb data are synced to megatest.db
    ;; do not use the run-ids list passed in to the function
    ;;
    (if (member 'new2old options)
	(let* ((maindb      (make-dbr:dbstruct path: toppath local: #t))
	       (src-run-ids (db:get-all-run-ids (db:dbdat-get-db (db:get-db maindb 0))))
	       (all-run-ids (sort (delete-duplicates (cons 0 src-run-ids)) <))
	       (count       1)
	       (total       (length all-run-ids))
	       (dead-runs  '()))
	  (for-each
	   (lambda (run-id)
	     (debug:print 0 "Processing run " (if (eq? run-id 0) " main.db " run-id) ", " count " of " total)
	     (set! count (+ count 1))
	     (let* ((fromdb (if toppath (make-dbr:dbstruct path: toppath local: #t) #f))
		    (frundb (db:dbdat-get-db (db:get-db fromdb run-id))))
	       ;; (db:delay-if-busy frundb)
	       ;; (db:delay-if-busy mtdb)
	       ;; (db:clean-up frundb)
	       (if (eq? run-id 0)
		   (begin
		     (db:sync-tables (db:sync-main-list dbstruct) (db:get-db fromdb #f) mtdb)
		     (set! dead-runs (db:clean-up-maindb (db:get-db fromdb #f))))
		   (begin
		     ;; NB// must sync first to ensure deleted tests get marked as such in megatest.db
		     (db:sync-tables db:sync-tests-only (db:get-db fromdb run-id) mtdb)
		     (db:clean-up-rundb (db:get-db fromdb run-id))
		     ))))
	   all-run-ids)
	  ;; removed deleted runs
	  (let ((dbdir (tasks:get-task-db-path)))
	    (for-each (lambda (run-id)
			(let ((fullname (conc dbdir "/" run-id ".db")))
			  (if (file-exists? fullname)
			      (begin
				(debug:print 0 "Removing database file for deleted run " fullname)
				(delete-file fullname)))))
		      dead-runs))))
    ;; (db:close-all dbstruct)
    ;; (sqlite3:finalize! mdb)
    ))

;; keeping it around for debugging purposes only
(define (open-run-close-no-exception-handling  proc idb . params)
  (debug:print-info 11 "open-run-close-no-exception-handling START given a db=" (if idb "yes " "no ") ", params=" params)

                                     reviewed    TIMESTAMP,
                                     iterated    TEXT DEFAULT '',
                                     avg_runtime REAL,
                                     avg_disk    REAL,
                                     tags        TEXT DEFAULT '',
                                     jobgroup    TEXT DEFAULT 'default',
                                CONSTRAINT test_meta_constraint UNIQUE (testname));")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS tasks_queue (id INTEGER PRIMARY KEY,
                                action TEXT DEFAULT '',
                                owner TEXT,
                                state TEXT DEFAULT 'new',
                                target TEXT DEFAULT '',
                                name TEXT DEFAULT '',
                                testpatt TEXT DEFAULT '',
                                keylock TEXT,
                                params TEXT,
                                creation_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                execution_time TIMESTAMP);")
       ;; archive disk areas, cached info from [archive-disks]
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_disks (
                                id INTEGER PRIMARY KEY,
                                archive_area_name TEXT,
                                disk_path TEXT,
                                last_df INTEGER DEFAULT -1,
                                last_df_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                creation_time TIMESTAMP DEFAULT (strftime('%','now')));")
       ;; individual bup (or tar) data chunks
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_blocks (
                                id INTEGER PRIMARY KEY,
                                archive_disk_id INTEGER,
                                disk_path TEXT,
                                last_du INTEGER DEFAULT -1,
                                last_du_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                creation_time TIMESTAMP DEFAULT (strftime('%','now')));")
       ;; tests allocated to what chunks. reusing a chunk for a test/item_path is very efficient
       ;; NB// the per run/test recording of where the archive is stored is done in the test
       ;;      record. 
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_allocations (
                                id INTEGER PRIMARY KEY,
                                archive_block_id INTEGER,
                                testname TEXT,
                                item_path TEXT,
                                creation_time TIMESTAMP DEFAULT (strftime('%','now')));")
       ;; move this clean up call somewhere else
       (sqlite3:execute db "DELETE FROM tasks_queue WHERE state='done' AND creation_time < ?;" (- (current-seconds)(* 24 60 60))) ;; remove older than 24 hrs
       (sqlite3:execute db (conc "CREATE INDEX IF NOT EXISTS runs_index ON runs (runname" (if havekeys "," "") keystr ");"))
       ;; (sqlite3:execute db "CREATE VIEW runs_tests AS SELECT * FROM runs INNER JOIN tests ON runs.id=tests.run_id;")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS extradat (id INTEGER PRIMARY KEY, run_id INTEGER, key TEXT, val TEXT);")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS metadat (id INTEGER PRIMARY KEY, var TEXT, val TEXT,
                                  CONSTRAINT metadat_constraint UNIQUE (var));")
       (sqlite3:execute db "CREATE TABLE IF NOT EXISTS access_log (id INTEGER PRIMARY KEY, user TEXT, accessed TIMESTAMP, args TEXT);")
       ;; Must do this *after* running patch db !! No more. 

                     final_logf   TEXT      DEFAULT 'logs/final.log',
                     logdat       TEXT      DEFAULT '', 
                     run_duration INTEGER   DEFAULT 0,
                     comment      TEXT      DEFAULT '',
                     event_time   TIMESTAMP DEFAULT (strftime('%s','now')),
                     fail_count   INTEGER   DEFAULT 0,
                     pass_count   INTEGER   DEFAULT 0,
                     archived     INTEGER   DEFAULT 0, -- 0=no, > 1=archive block id where test data can be found
                        CONSTRAINT testsconstraint UNIQUE (run_id, testname, item_path));")
     (sqlite3:execute db "CREATE INDEX IF NOT EXISTS tests_index ON tests (run_id, testname, item_path);")
     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS test_steps 
                              (id INTEGER PRIMARY KEY,
                               test_id INTEGER, 
                               stepname TEXT, 
                               state TEXT DEFAULT 'NOT_STARTED', 

     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS test_rundat (
                              id           INTEGER PRIMARY KEY,
                              test_id      INTEGER,
                              update_time  TIMESTAMP,
                              cpuload      INTEGER DEFAULT -1,
                              diskfree     INTEGER DEFAULT -1,
                              diskusage    INTGER DEFAULT -1,
                              run_duration INTEGER DEFAULT 0);")
     (sqlite3:execute db "CREATE TABLE IF NOT EXISTS archives (
                              id           INTEGER PRIMARY KEY,
                              test_id      INTEGER,
                              state        TEXT DEFAULT 'new',
                              status       TEXT DEFAULT 'n/a',
                              archive_type TEXT DEFAULT 'bup',
                              du           INTEGER,
                              archive_path TEXT);")))
  db)

;;======================================================================
;; A R C H I V E S
;;======================================================================

;; dneeded is minimum space needed, scan for existing archives that 
;; are on disks with adequate space and already have this test/itempath
;; archived
;;
(define (db:archive-get-allocations dbstruct testname itempath dneeded)
  (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
	 (db           (db:dbdat-get-db dbdat))
	 (res          '())
	 (blocks       '())) ;; a block is an archive chunck that can be added too if there is space
    (sqlite3:for-each-row
     (lambda (id archive-disk-id disk-path last-du last-du-time)
       (set! res (cons (vector id archive-disk-id disk-path last-du last-du-time) res)))
     db
     "SELECT b.id,b.archive_disk_id,b.disk_path,b.last_du,b.last_du_time FROM archive_blocks AS b
        INNER JOIN archive_allocations AS a ON a.archive_block_id=b.id
        WHERE a.testname=? AND a.item_path=?;" 
     testname itempath)
    ;; Now res has list of candidate paths, look in archive_disks for candidate with potential free space
    (if (null? res)
	'()
	(sqlite3:for-each-row
	 (lambda (id archive-area-name disk-path last-df last-df-time)
	   (set! blocks (cons (vector id archive-area-name disk-path last-df last-df-time) blocks)))
	 db 
	 (conc
	  "SELECT d.id,d.archive_area_name,disk_path,last_df,last_df_time FROM archive_disks AS d
             INNER JOIN archive_blocks AS b ON d.id=b.archive_disk_id
             WHERE b.id IN (" (string-intersperse (map conc res) ",") ") AND
         last_df > ?;")
	 dneeded))
    blocks))
    
;; returns id of the record, register a disk allocated to archiving and record it's last known
;; available space
;;
(define (db:archive-register-disk dbstruct bdisk-name bdisk-path df)
  (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
	 (db           (db:dbdat-get-db dbdat))
	 (res          #f))
    (sqlite3:for-each-row
     (lambda (id)
       (set! res id))
     db
     "SELECT id FROM archive_disks WHERE archive_area_name=? AND disk_path=?;"
     bdisk-name bdisk-path)
    (if res ;; record exists, update df and return id
	(begin
	  (sqlite3:execute db "UPDATE archive_disks SET last_df=?,last_df_time=(strftime('%s','now'))
                                  WHERE archive_area_name=? AND disk_path=?;"
			   df bdisk-name bdisk-path)
	  res)
	(begin
	  (sqlite3:execute
	   db
	   "INSERT OR REPLACE INTO archive_disks (archive_area_name,disk_path,last_df)
                VALUES (?,?,?);"
	   bdisk-name bdisk-path df)
	  (db:archive-register-disk dbstruct bdisk-name bdisk-path df)))))

;; record an archive path created on a given archive disk (identified by it's bdisk-id)
;; if path starts with / then it is full, otherwise it is relative to the archive disk
;; preference is to store the relative path.
;;
(define (db:archive-register-block-name dbstruct bdisk-id archive-path #!key (du #f))
  (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
	 (db           (db:dbdat-get-db dbdat))
	 (res          #f))
    ;; first look to see if this path is already registered
    (sqlite3:for-each-row
     (lambda (id)
       (set! res id))
     db
     "SELECT id FROM archive_blocks WHERE archive_disk_id=? AND disk_path=?;"
     bdisk-id archive-path)
    (if res ;; record exists, update du if applicable and return res
	(begin
	  (if du (sqlite3:exectute db "UPDATE archive_blocks SET last_du=?,last_du_time=(strftime('%s','now'))
                                          WHERE archive_disk_id=? AND disk_path=?;"
				   bdisk-id archive-path du))
	  res)
	(begin
	  (sqlite3:execute db "INSERT OR REPLACE INTO archive_blocks (archive_disk_id,disk_path,last_du)
                                                        VALUES (?,?,?);"
			   bdisk-id archive-path (or du 0))
	  (db:archive-register-block-name dbstruct bdisk-id archive-path du: du)))))


;; The "archived" field in tests is overloaded; 0 = not archived, > 0 archived in block with given id
;;
(define (db:test-set-archive-block-id dbstruct run-id test-id archive-block-id)
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (db)
     (sqlite3:execute db "UPDATE tests SET archived=? WHERE id=?;"
		      archive-block-id test-id))))
 

;; (define (db:archive-allocate-testsuite/area-to-block block-id testsuite-name areakey)
;;   (let* ((dbdat        (db:get-db dbstruct #f)) ;; archive tables are in main.db
;; 	 (db           (db:dbdat-get-db dbdat))
;; 	 (res          '())
;; 	 (blocks       '())) ;; a block is an archive chunck that can be added too if there is space
;;     (sqlite3:for-each-row  #f)

;;======================================================================
;; L O G G I N G    D B 
;;======================================================================

(define (open-logging-db) ;;  (conc *toppath* "/megatest.db") (car *configinfo*)))
  (let* ((dbpath    (conc (if *toppath* (conc *toppath* "/") "") "logging.db")) ;; fname)

;;======================================================================
;; D B   U T I L S
;;======================================================================

;;======================================================================
;; M A I N T E N A N C E
;;======================================================================

(define (db:have-incompletes? dbstruct run-id ovr-deadtime)
  (let* ((dbdat        (db:get-db dbstruct run-id))
	 (db           (db:dbdat-get-db dbdat))
	 (incompleted '())
	 (oldlaunched '())
	 (toplevels   '())
	 (deadtime-str (configf:lookup *configdat* "setup" "deadtime"))
	 (deadtime     (if (and deadtime-str
				(string->number deadtime-str))
			   (string->number deadtime-str)
			   7200))) ;; two hours
    (if (number? ovr-deadtime)(set! deadtime ovr-deadtime))
    
    ;; in RUNNING or REMOTEHOSTSTART for more than 10 minutes
    ;;
    ;; HOWEVER: this code in run:test seems to work fine
    ;;              (> (- (current-seconds)(+ (db:test-get-event_time testdat)
    ;;                     (db:test-get-run_duration testdat)))
    ;;                    600) 
    (db:delay-if-busy dbdat)
    (sqlite3:for-each-row 
     (lambda (test-id run-dir uname testname item-path)
       (if (and (equal? uname "n/a")
		(equal? item-path "")) ;; this is a toplevel test
	   ;; what to do with toplevel? call rollup?
	   (begin
	     (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
	     (debug:print-info 0 "Found old toplevel test in RUNNING state, test-id=" test-id))
	   (set! incompleted (cons (list test-id run-dir uname testname item-path run-id) incompleted))))
     db
     "SELECT id,rundir,uname,testname,item_path FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?) AND state IN ('RUNNING','REMOTEHOSTSTART');"
     run-id deadtime)

    ;; in LAUNCHED for more than one day. Could be long due to job queues TODO/BUG: Need override for this in config
    ;;
    (db:delay-if-busy dbdat)
    (sqlite3:for-each-row
     (lambda (test-id run-dir uname testname item-path)
       (if (and (equal? uname "n/a")
		(equal? item-path "")) ;; this is a toplevel test
	   ;; what to do with toplevel? call rollup?
	   (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
	   (set! oldlaunched (cons (list test-id run-dir uname testname item-path run-id) oldlaunched))))
     db
     "SELECT id,rundir,uname,testname,item_path FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > 86400 AND state IN ('LAUNCHED');"
     run-id)
    
    (debug:print-info 18 "Found " (length oldlaunched) " old LAUNCHED items, " (length toplevels) " old LAUNCHED toplevel tests and " (length incompleted) " tests marked RUNNING but apparently dead.")
    (if (and (null? incompleted)
	     (null? oldlaunched)
	     (null? toplevels))
	#f
	#t)))

;;  select end_time-now from
;;      (select testname,item_path,event_time+run_duration as
;;                          end_time,strftime('%s','now') as now from tests where state in
;;      ('RUNNING','REMOTEHOSTSTART','LAUNCED'));

(define (db:find-and-mark-incomplete dbstruct run-id ovr-deadtime)

;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up dbdat)
  ;; (debug:print 0 "WARNING: db clean up not fully ported to v1.60, cleanup action will be on megatest.db")
  (let* ((db         (db:dbdat-get-db dbdat))
	 (count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM tests)+(SELECT count(id) FROM runs);"))
	(statements
	 (map (lambda (stmt)
		(sqlite3:prepare db stmt))
	      (list
	       ;; delete all tests that belong to runs that are 'deleted'

			       (debug:print-info 0 "Records count after  clean: " tot))
			     count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")))

;; Clean out old junk and vacuum the database
;;
;; Ultimately do something like this:
;;
;; 1. Look at test records either deleted or part of deleted run:
;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up-rundb dbdat)
  ;; (debug:print 0 "WARNING: db clean up not fully ported to v1.60, cleanup action will be on megatest.db")
  (let* ((db         (db:dbdat-get-db dbdat))
	 (count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM tests);"))
	(statements
	 (map (lambda (stmt)
		(sqlite3:prepare db stmt))
	      (list
	       ;; delete all tests that belong to runs that are 'deleted'
	       ;; (conc "DELETE FROM tests WHERE run_id NOT IN (" (string-intersperse (map conc valid-runs) ",") ");")
	       ;; delete all tests that are 'DELETED'
	       "DELETE FROM tests WHERE state='DELETED';"
	       ))))
    (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count before clean: " tot))
			     count-stmt)
       (map sqlite3:execute statements)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count after  clean: " tot))
			     count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")))

;; Clean out old junk and vacuum the database
;;
;; Ultimately do something like this:
;;
;; 1. Look at test records either deleted or part of deleted run:
;;    a. If test dir exists, set the the test to state='UNKNOWN', Set the run to 'unknown'
;;    b. If test dir gone, delete the test record
;; 2. Look at run records
;;    a. If have tests that are not deleted, set state='unknown'
;;    b. ....
;;
(define (db:clean-up-maindb dbdat)
  ;; (debug:print 0 "WARNING: db clean up not fully ported to v1.60, cleanup action will be on megatest.db")
  (let* ((db         (db:dbdat-get-db dbdat))
	 (count-stmt (sqlite3:prepare db "SELECT (SELECT count(id) FROM runs);"))
	 (statements
	  (map (lambda (stmt)
		 (sqlite3:prepare db stmt))
	       (list
		;; delete all tests that belong to runs that are 'deleted'
		;; (conc "DELETE FROM tests WHERE run_id NOT IN (" (string-intersperse (map conc valid-runs) ",") ");")
		;; delete all tests that are 'DELETED'
		"DELETE FROM runs WHERE state='deleted';"
		)))
	 (dead-runs '()))
    (sqlite3:for-each-row
     (lambda (run-id)
       (set! dead-runs (cons run-id dead-runs)))
       db
       "SELECT id FROM runs WHERE state='deleted';")
    (db:delay-if-busy dbdat)
    (sqlite3:with-transaction 
     db
     (lambda ()
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count before clean: " tot))
			     count-stmt)
       (map sqlite3:execute statements)
       (sqlite3:for-each-row (lambda (tot)
			       (debug:print-info 0 "Records count after  clean: " tot))
			     count-stmt)))
    (map sqlite3:finalize! statements)
    (sqlite3:finalize! count-stmt)
    ;; (db:find-and-mark-incomplete db)
    (db:delay-if-busy dbdat)
    (sqlite3:execute db "VACUUM;")
    dead-runs))

;;======================================================================
;; M E T A   G E T   A N D   S E T   V A R S
;;======================================================================

;; returns number if string->number is successful, string otherwise
;; also updates *global-delta*

       res))))

;;======================================================================
;; QUEUE UP META, TEST STATUS AND STEPS REMOTE ACCESS
;;======================================================================

;; NOTE: Can remove the regex and base64 encoding for zmq
(define (db:obj->string obj #!key (transport 'http))
  (case transport
    ;; ((fs) obj)
    ((http fs)
     (string-substitute
      (regexp "=") "_"
      (base64:base64-encode 
       (z3:encode-buffer
	(with-output-to-string
	  (lambda ()(serialize obj)))))
      #t))
    ((zmq nmsg)(with-output-to-string (lambda ()(serialize obj))))
    (else obj)))

(define (db:string->obj msg #!key (transport 'http))
  (case transport
    ;; ((fs) msg)
    ((http fs)
     (if (string? msg)
	 (with-input-from-string 
	     (z3:decode-buffer
	      (base64:base64-decode
	       (string-substitute 
		(regexp "_") "=" msg #t)))
	   (lambda ()(deserialize)))
	 (begin
	   (debug:print 0 "ERROR: reception failed. Received " msg " but cannot translate it.")
	   msg))) ;; crude reply for when things go awry
    ((zmq nmsg)(with-input-from-string msg (lambda ()(deserialize))))
    (else msg)))

(define (db:test-set-status-state dbstruct run-id test-id status state msg)
  (let ((dbdat  (db:get-db dbstruct run-id)))
    (if (member state '("LAUNCHED" "REMOTEHOSTSTART"))
	(db:general-call dbdat 'set-test-start-time (list test-id)))
    (if msg

	    (db:general-call dbdat 'top-test-set-running (list test-name))
	    (if (equal? status "LAUNCHED")
		(db:general-call dbdat 'top-test-set (list "LAUNCHED" test-name))
		(db:general-call dbdat 'top-test-set-per-pf-counts (list test-name run-id test-name test-name test-name))))
	#f)
      #f))









(define (db:test-get-logfile-info dbstruct run-id test-name)
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (db)
     (let ((res #f))

	 ;; the synchash calls modify the "data" hash
	 (get-runs-sig    (conc (client:get-signature) " get-runs"))
	 (get-tests-sig   (conc (client:get-signature) " get-tests"))
	 (get-details-sig (conc (client:get-signature) " get-test-details"))

	 ;; test-ids to get and display are indexed on window-id in curr-test-ids hash
	 (test-ids        (hash-table-values (dboard:data-get-curr-test-ids *data*)))
	 ;; run-id is #f in next line to send the query to server 0
 	 (run-changes     (synchash:client-get 'db:get-runs get-runs-sig (length keypatts) data #f runname #f #f keypatts))
	 (tests-detail-changes (if (not (null? test-ids))
				   (synchash:client-get 'db:get-test-info-by-ids get-details-sig 0  data #f test-ids)
				   '()))

	 ;; Now can calculate the run-ids
	 (run-hash    (hash-table-ref/default data get-runs-sig #f))
	 (run-ids     (if run-hash (filter number? (hash-table-keys run-hash)) '()))

	 (all-test-changes (let ((res (make-hash-table)))
			     (for-each (lambda (run-id)
					 (if (> run-id 0)
					     (hash-table-set! res run-id (synchash:client-get 'db:get-tests-for-run-mindata get-tests-sig 0 data run-id 1 testpatt states statuses #f))))
				       run-ids)
			     res))
	 (runs-hash    (hash-table-ref/default data get-runs-sig #f))
	 (header       (hash-table-ref/default runs-hash "header" #f))
	 (run-ids      (sort (filter number? (hash-table-keys runs-hash))
			     (lambda (a b)
			       (let* ((record-a (hash-table-ref runs-hash a))
				      (record-b (hash-table-ref runs-hash b))
				      (time-a   (db:get-value-by-header record-a header "event_time"))

		  (set! colnum (+ colnum 1))))
	      run-ids)

    ;; Scan all tests to be displayed and organise all the test names, respecting what is in the hash table
    ;; Do this analysis in the order of the run-ids, the most recent run wins
    (for-each (lambda (run-id)
		(let* ((run-path       (hash-table-ref (dboard:data-get-run-keys *data*) run-id))
		       (test-changes   (hash-table-ref all-test-changes run-id))
		       (new-test-dat   (car test-changes))
		       (removed-tests  (cadr test-changes))
		       (tests          (sort (map cadr (filter (lambda (testrec)
								 (eq? run-id (db:mintest-get-run_id (cadr testrec))))
							       new-test-dat))
					     (lambda (a b)
					       (let ((time-a (db:mintest-get-event_time a))

				      (iup:attribute-set! (dboard:data-get-runs-matrix *data*)
							  (conc rownum ":" 0) dispname)
				      ))
				;; set the cell text and color
				;; (debug:print 2 "rownum:colnum=" rownum ":" colnum ", state=" status)
				(iup:attribute-set! (dboard:data-get-runs-matrix *data*)
						    (conc rownum ":" colnum)
						    (if (member state '("ARCHIVED" "COMPLETED"))
							status
							state))
				(iup:attribute-set! (dboard:data-get-runs-matrix *data*)
						    (conc "BGCOLOR" rownum ":" colnum)
						    (car (gutils:get-color-for-state-status state status)))
				))
			    tests)))
	      run-ids)

    (let ((updater (hash-table-ref/default  (dboard:data-get-updaters *data*) window-id #f)))
      (if updater (updater (hash-table-ref/default data get-details-sig #f))))

    (iup:attribute-set! (dboard:data-get-runs-matrix *data*) "REDRAW" "ALL")
    ;; (debug:print 2 "run-changes: " run-changes)
    ;; (debug:print 2 "test-changes: " test-changes)
    (list run-changes all-test-changes)))

;;======================================================================
;; TESTS DATA
;;======================================================================

;; Produce a list of lists ready for common:sparse-list-generate-index
;;

sudo netstat -lptu
sudo netstat -tulpn</code></pre>
</div></div>
</div>
</div>
</div>
<h1 id="_reference">Reference</h1>







<div class="sect1">
<h2 id="_the_testconfig_file">The testconfig File</h2>
<div class="sectionbody">
<div class="sect2">
<h3 id="_setup_section">Setup section</h3>
<div class="sect3">
<h4 id="_header">Header</h4>

<td class="icon">
<div class="title">Note</div>
</td>
<td class="content">There is a trailing space after the --</td>
</tr></table>
</div>
</div>
</div>
</div>
<div class="sect1">
<h2 id="_programming_api">Programming API</h2>
<div class="sectionbody">
<div class="paragraph"><p>These routines can be called from the megatest repl.</p></div>
<div class="tableblock">
<table rules="all"
width="70%"
frame="hsides"
cellspacing="0" cellpadding="4">
<caption class="title">Table 2. API Server Management Calls</caption>
<col width="14%" />
<col width="28%" />
<col width="28%" />
<col width="28%" />
<thead>
<tr>
<th align="center" valign="top">API Call                  </th>
<th align="left" valign="top"> Purpose comments   </th>
<th align="left" valign="top"> Returns                 </th>
<th align="left" valign="top"> Comments</th>
</tr>
</thead>
<tfoot>
<tr>
<td align="center" valign="top"><p class="table">(rmt:login        run-id)</p></td>
<td align="left" valign="top"><p class="table"><code>Verify the the version, testsuite area etc. are correct.</code></p></td>
<td align="left" valign="top"><p class="table"><code>#( #t "successful login" )</code></p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
</tr>
</tfoot>
<tbody>
<tr>
<td align="center" valign="top"><p class="table">(rmt:start-server run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>#( success/fail   n/a )</code></p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
</tr>
<tr>
<td align="center" valign="top"><p class="table">(rmt:kill-server  run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>#( success/fail   n/a )</code></p></td>
<td align="left" valign="top"><p class="table"><code>Works only if the server is still reachable</code></p></td>
</tr>
</tbody>
</table>
</div>
<div class="tableblock">
<table rules="all"
width="70%"
frame="hsides"
cellspacing="0" cellpadding="4">
<caption class="title">Table 3. API Keys Related Calls</caption>
<col width="14%" />
<col width="28%" />
<col width="28%" />
<col width="28%" />
<thead>
<tr>
<th align="center" valign="top">API Call                        </th>
<th align="left" valign="top"> Purpose comments   </th>
<th align="left" valign="top"> Returns                 </th>
<th align="left" valign="top"> Comments</th>
</tr>
</thead>
<tfoot>
<tr>
<td align="center" valign="top"><p class="table">(rmt:get-key-val-pairs run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>#t=success/#f=fail</code></p></td>
<td align="left" valign="top"><p class="table"><code>Works only if the server is still reachable</code></p></td>
</tr>
</tfoot>
<tbody>
<tr>
<td align="center" valign="top"><p class="table">(rmt:get-keys run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>( key1 key2 &#8230; )</code></p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
</tr>
</tbody>
</table>
</div>
<div class="sect2">
<h3 id="_megatest_internals">Megatest Internals</h3>
<div class="imageblock graphviz">
<div class="content">
<img src="server.png" alt="server.png" />
</div>
</div>

</div>
</div>
</div>
<div id="footnotes"><hr /></div>
<div id="footer">
<div id="footer-text">
Version 1.0<br />
Last updated 2014-11-25 21:11:28 MST
</div>
</div>
</body>
</html>

Reference
=========







The testconfig File
-------------------

Setup section
~~~~~~~~~~~~~

Header

-----------------
[triggers]
COMPLETED/ xterm -e bash -s -- 
-----------------

NOTE: There is a trailing space after the --

Programming API
---------------

These routines can be called from the megatest repl. 

.API Server Management Calls
[width="70%",cols="^,2m,2m,2m",frame="topbot",options="header,footer"]
|======================
|API Call                  | Purpose comments   | Returns                 | Comments
|(rmt:start-server run-id) |                    | #( success/fail   n/a ) | 
|(rmt:kill-server  run-id) |                    | #( success/fail   n/a ) | Works only if the server is still reachable
|(rmt:login        run-id) | Verify the the version, testsuite area etc. are correct. | #( #t "successful login" ) |
|======================

.API Keys Related Calls
[width="70%",cols="^,2m,2m,2m",frame="topbot",options="header,footer"]
|======================
|API Call                        | Purpose comments   | Returns                 | Comments
|(rmt:get-keys run-id)           |                    | ( key1 key2 ... )       | 
| (rmt:get-key-val-pairs run-id) |                    | #t=success/#f=fail      | Works only if the server is still reachable
|======================


:numbered!:

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

(require-extension (srfi 18) extras tcp s11n)

(use sqlite3 srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest)
(import (prefix sqlite3 sqlite3:))

(use spiffy uri-common intarweb http-client spiffy-request-vars)

(tcp-buffer-size 2048)

(declare (unit fs-transport))

(declare (uses common))
(declare (uses db))
(declare (uses tests))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.

(include "common_records.scm")
(include "db_records.scm")


;;======================================================================
;; F S   T R A N S P O R T   S E R V E R
;;======================================================================

;; There is no "server" per se but a convience routine to make it non
;; necessary to be reopening the db over and over again.
;;

(define (fs:process-queue-item packet)
  (if (not *megatest-db*) ;; we will require that (setup-for-run) has already been called
      (set! *megatest-db* (open-db)))
  (debug:print-info 11 "fs:process-queue-item called with packet=" packet)
  (db:process-queue-item *megatest-db* packet))
      

	 (c2 (map string->number (string-split color2)))
	 (delta (map (lambda (a b)(abs (- a b))) c1 c2)))
    (null? (filter (lambda (x)(> x 3)) delta))))

(define (gutils:get-color-for-state-status state status);; #!key (get-label #f))
  ;; ((if get-label cadr car)
   (case (string->symbol state)
     ((COMPLETED ARCHIVED)
      (case (string->symbol status)
	((PASS)        (list "70  249 73" status))
	((WARN WAIVED) (list "255 172 13" status))
	((SKIP)        (list "230 230 0" status))
	(else (list "223 33 49" status))))
     ;;      (if (equal? status "PASS")
     ;;	  '("70 249 73" "PASS")

	 (link-tree-path  (configf:lookup *configdat* "setup" "linktree")))
    ;; (set! db *inmemdb*)
    (debug:print-info 0 "portlogger recommended port: " start-port)
    (root-path     (if link-tree-path 
		       link-tree-path
		       (current-directory))) ;; WARNING: SECURITY HOLE. FIX ASAP!
    (handle-directory spiffy-directory-listing)
    (handle-exception (lambda (exn chain)
			(signal (make-composite-condition
				 (make-property-condition 
				  'server
				  'message "server error")))))

    ;; http-transport:handle-directory) ;; simple-directory-handler)
    ;; Setup the web server and a /ctrl interface
    ;;
    (vhost-map `(((* any) . ,(lambda (continue)
			       ;; open the db on the first call 
				 ;; This is were we set up the database connections
			       (let* (($   (request-vars source: 'both))
				      (dat ($ 'dat))
				      (res #f))
				 (cond
				  ((equal? (uri-path (request-uri (current-request)))
					   '(/ "api"))
				   (send-response body:    (api:process-request *inmemdb* $) ;; the $ is the request vars proc
						  headers: '((content-type text/plain)))
				   (mutex-lock! *heartbeat-mutex*)
				   (set! *last-db-access* (current-seconds))
				   (mutex-unlock! *heartbeat-mutex*))





















				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ ""))
				   (send-response body: (http-transport:main-page)))
				  ((equal? (uri-path (request-uri (current-request))) 
					   '(/ "runs"))
				   (send-response body: (http-transport:main-page)))
				  ((equal? (uri-path (request-uri (current-request))) 

(define (http-transport:client-api-send-receive run-id serverdat cmd params #!key (numretries 3))
  (let* ((fullurl    (if (vector? serverdat)
			 (http-transport:server-dat-get-api-req serverdat)
			 (begin
			   (debug:print 0 "FATAL ERROR: http-transport:client-api-send-receive called with no server info")
			   (exit 1))))
	 (res        #f)
	 (success    #t)
	 (sparams    (db:obj->string params transport: 'http)))
;;    (condition-case
;;     handle-exceptions
;;     exn
;;     (if (> numretries 0)
;;	 (begin
;;	   (mutex-unlock! *http-mutex*)
;;	   (thread-sleep! 1)
;;	   (handle-exceptions
;;	    exn
;;	    (debug:print 0 "WARNING: closing connections failed. Server at " fullurl " almost certainly dead")
;;	    (close-all-connections!))
;;	   (debug:print 0 "WARNING: Failed to communicate with server, trying again, numretries left: " numretries)
;;	   (http-transport:client-api-send-receive run-id serverdat cmd sparams numretries: (- numretries 1)))
;;	 (begin
;;	   (mutex-unlock! *http-mutex*)
;;	   (tasks:kill-server-run-id run-id)
;;	   #f))
;;     (begin
       (debug:print-info 11 "fullurl=" fullurl ", cmd=" cmd ", params=" params ", run-id=" run-id "\n")
       ;; set up the http-client here
       (max-retry-attempts 1)
       ;; consider all requests indempotent
       (retry-request? (lambda (request)
			 #f))
       ;; send the data and get the response
       ;; extract the needed info from the http data and 
       ;; process and return it.
       (let* ((send-recieve (lambda ()
			      (mutex-lock! *http-mutex*)
			      ;; (condition-case (with-input-from-request "http://localhost"; #f read-lines)
			      ;;					       ((exn http client-error) e (print e)))
			      (set! res (vector
					 success
					 (db:string->obj 
					  (handle-exceptions
					   exn
					   (begin
					     (set! success #f)
					     (debug:print 0 "WARNING: failure in with-input-from-request to " fullurl ".")
					     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
					     (hash-table-delete! *runremote* run-id)
					     ;; Killing associated server to allow clean retry.")
					     (tasks:kill-server-run-id run-id)  ;; better to kill the server in the logic that called this routine?

					     (mutex-unlock! *http-mutex*)
					     (signal (make-composite-condition
						      (make-property-condition 'commfail 'message "failed to connect to server")))
					     "communications failed")
					   (with-input-from-request ;; was dat
					    fullurl 
					    (list (cons 'key "thekey")
						  (cons 'cmd cmd)
						  (cons 'params sparams))
					    read-string))
					  transport: 'http)))
			      ;; Shouldn't this be a call to the managed call-all-connections stuff above?
			      (close-all-connections!)
			      (mutex-unlock! *http-mutex*)
			      ))
	      (time-out     (lambda ()
			      (thread-sleep! 45)
			      #f))
	      (th1 (make-thread send-recieve "with-input-from-request"))
	      (th2 (make-thread time-out     "time out")))
	 (thread-start! th1)
	 (thread-start! th2)
	 (thread-join! th1)
	 (thread-terminate! th2)
	 (debug:print-info 11 "got res=" res)
	 (if (vector? res)
	     (if (vector-ref res 0)
		 res
		 (begin ;; note: this code also called in nmsg-transport - consider consolidating it
		   (debug:print 0 "ERROR: error occured at server, info=" (vector-ref res 2))
		   (debug:print 0 " client call chain:")
		   (print-call-chain (current-error-port))
		   (debug:print 0 " server call chain:")
		   (pp (vector-ref res 1) (current-error-port))
		   (signal (vector-ref result 0))))
	     (signal (make-composite-condition
		      (make-property-condition 
		       'timeout
		       'message "nmsg-transport:client-api-send-receive-raw timed out talking to server")))))))

;; careful closing of connections stored in *runremote*
;;
(define (http-transport:close-connections run-id)
  (let* ((server-dat (hash-table-ref/default *runremote* run-id #f)))
    (if (vector? server-dat)
	(let ((api-dat (http-transport:server-dat-get-api-uri server-dat)))
	  (close-connection! api-dat)
	  #t)
	#f)))


(define (make-http-transport:server-dat)(make-vector 6))
(define (http-transport:server-dat-get-iface         vec)    (vector-ref  vec 0))
(define (http-transport:server-dat-get-port          vec)    (vector-ref  vec 1))
(define (http-transport:server-dat-get-api-uri       vec)    (vector-ref  vec 2))
(define (http-transport:server-dat-get-api-url       vec)    (vector-ref  vec 3))
(define (http-transport:server-dat-get-api-req       vec)    (vector-ref  vec 4))
(define (http-transport:server-dat-get-last-access   vec)    (vector-ref  vec 5))
(define (http-transport:server-dat-get-socket        vec)    (vector-ref  vec 6))

(define (http-transport:server-dat-make-url vec)
  (if (and (http-transport:server-dat-get-iface vec)
	   (http-transport:server-dat-get-port  vec))
      (conc "http://" 
	    (http-transport:server-dat-get-iface vec)
	    ":"
	    (http-transport:server-dat-get-port  vec))
      #f))

(define (http-transport:server-dat-update-last-access vec)
  (if (vector? vec)
      (vector-set! vec 5 (current-seconds))
      (begin
	(print-call-chain (current-error-port))
	(debug:print 0 "ERROR: call to http-transport:server-dat-update-last-access with non-vector!!"))))

;;
;; connect
;;
(define (http-transport:client-connect iface port)
  (let* ((api-url      (conc "http://" iface ":" port "/api"))
	 (api-uri      (uri-reference (conc "http://" iface ":" port "/api")))
	 (api-req      (make-request method: 'POST uri: api-uri))
	 (server-dat   (vector iface port api-uri api-url api-req (current-seconds))))
    server-dat))

;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running server-id run-id)
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (debug:print-info 0 "Starting the sync-back, keep alive thread in server for run-id=" run-id)
  (let* ((tdbdat      (tasks:open-db))
	 (server-info (let loop ((start-time (current-seconds))
				 (changed    #t)
				 (last-sdat  "not this"))
                        (let ((sdat #f))
			  (thread-sleep! 0.01)
			  (debug:print-info 0 "Waiting for server alive signature")

				      (exit))
				    (loop start-time
					  (equal? sdat last-sdat)
					  sdat)))))))
         (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (server-timeout (server:get-timeout)))







    (let loop ((count         0)
	       (server-state 'available)
	       (bad-sync-count 0))

      ;; Use this opportunity to sync the inmemdb to db
      (if *inmemdb* 
	  (let ((start-time (current-milliseconds))
		(sync-time  #f)
		(rem-time   #f))
	    ;; inmemdb is a dbstruct
	    (condition-case
	     (db:sync-touched *inmemdb* *run-id* force-sync: #t)
	     ((sync-failed)(cond
			    ((> bad-sync-count 10) ;; time to give up
			     (http-transport:server-shutdown server-id port))
			    (else ;; (> bad-sync-count 0)  ;; we've had a fail or two, delay and loop
			     (thread-sleep! 5)
			     (loop count server-state (+ bad-sync-count 1)))))
	     ((exn)
	      (debug:print 0 "ERROR: error from sync code other than 'sync-failed. Attempting to gracefully shutdown the server")
	      (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id " http-transport:keep-running crashed")
	      (exit)))
	    (set! sync-time  (- (current-milliseconds) start-time))
	    (set! rem-time (quotient (- 4000 sync-time) 1000))
	    (debug:print 2 "SYNC: time= " sync-time ", rem-time=" rem-time)
	    
	    (if (and (<= rem-time 4)
		     (> rem-time 0))
		(thread-sleep! rem-time)
		(thread-sleep! 4))) ;; fallback for if the math is changed ...

	  ;;
	  ;; no *inmemdb* yet, set running after our first pass through and start the db
	  ;;
	  (if (eq? server-state 'available)
	      (let ((new-server-id (tasks:server-am-i-the-server? (db:delay-if-busy tdbdat) run-id))) ;; try to ensure no double registering of servers
		(if (equal? new-server-id server-id)
		    (begin
		      (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "dbprep")
		      (thread-sleep! 0.5) ;; give some margin for queries to complete before switching from file based access to server based access
		      (set! *inmemdb*  (db:setup run-id))
		      ;; force initialization
		      ;; (db:get-db *inmemdb* #t)
		      (db:get-db *inmemdb* run-id)
		      (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "running"))
		    (begin ;; gotta exit nicely




		      (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "collision")
		      (http-transport:server-shutdown server-id port))))))
      
      (if (< count 1) ;; 3x3 = 9 secs aprox
	  (loop (+ count 1) 'running bad-sync-count))
      
      ;; Check that iface and port have not changed (can happen if server port collides)
      (mutex-lock! *heartbeat-mutex*)
      (set! sdat *server-info*)
      (mutex-unlock! *heartbeat-mutex*)
      
      (if (or (not (equal? sdat (list iface port)))

      ;; (debug:print 11 "last-access=" last-access ", server-timeout=" server-timeout)
      ;;
      ;; no_traffic, no running tests, if server 0, no running servers
      ;;
      ;; (let ((wait-on-running (configf:lookup *configdat* "server" "wait-on-running"))) ;; wait on running tasks (if not true then exit on time out)
      ;;
      (if (and *server-run*

	       (> (+ last-access server-timeout)
		  (current-seconds)))





	  (begin
	    (debug:print-info 0 "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
	    ;;
	    ;; Consider implementing some smarts here to re-insert the record or kill self is
	    ;; the db indicates so
	    ;;
	    ;; (if (tasks:server-am-i-the-server? tdb run-id)
	    ;;     (tasks:server-set-state! tdb server-id "running"))
	    ;;
	    (loop 0 server-state bad-sync-count))
	  (http-transport:server-shutdown server-id port)))))

(define (http-transport:server-shutdown server-id port)
  (let ((tdbdat (tasks:open-db)))
    (debug:print-info 0 "Starting to shutdown the server.")
    ;; need to delete only *my* server entry (future use)
    (set! *time-to-exit* #t)

				   "Keep running")))
	    (thread-start! th2)
	    (thread-sleep! 0.25) ;; give the server time to settle before starting the keep-running monitor.
	    (thread-start! th3)
	    (set! *didsomething* #t)
	    (thread-join! th2)
	    (exit))))))

(define (http:ping run-id host-port)
  (let* ((server-dat (http-transport:client-connect (car host-port)(cadr host-port)))
	 (login-res  (rmt:login-no-auto-client-setup server-dat run-id)))
    (if (and (list? login-res)
	     (car login-res))
	(begin
	  (print "LOGIN_OK")
	  (exit 0))
	(begin
	  (print "LOGIN_FAILED")
	  (exit 1)))))

(define (http-transport:server-signal-handler signum)
  (signal-mask! signum)
  (handle-exceptions
   exn
   (debug:print " ... exiting ...")
   (let ((th1 (make-thread (lambda ()

(define (launch:setup-for-run #!key (force #f))
  ;; would set values for KEYS in the environment here for better support of env-override but 
  ;; have chicken/egg scenario. need to read megatest.config then read it again. Going to 
  ;; pass on that idea for now
  ;; special case
  (if (or force (not (hash-table? *configdat*)))  ;; no need to re-open on every call
      (begin
	(set! *configinfo* (or (if (get-environment-variable "MT_CMDINFO") ;; we are inside a test - do not reprocess configs
				   (let ((alistconfig (conc (get-environment-variable "MT_LINKTREE") "/"
							    (get-environment-variable "MT_TARGET")   "/"
							    (get-environment-variable "MT_RUNNAME")  "/"
							    ".megatest.cfg")))
				     (if (file-exists? alistconfig)
					 (list (configf:read-alist alistconfig)
					       (get-environment-variable "MT_RUN_AREA_HOME"))
					 #f))
				   #f) ;; no config cached - give up
			       (find-and-read-config 
				(if (args:get-arg "-config")(args:get-arg "-config") "megatest.config")
				environ-patt: "env-override"
				given-toppath: (get-environment-variable "MT_RUN_AREA_HOME")
				pathenvvar: "MT_RUN_AREA_HOME")))
	(set! *configdat*  (if (car *configinfo*)(car *configinfo*) #f))
	(set! *toppath*    (if (car *configinfo*)(cadr *configinfo*) #f))
	(let* ((tmptransport (configf:lookup *configdat* "server" "transport"))
	       (transport    (if tmptransport (string->symbol tmptransport) 'http)))
	  (if (member transport '(http rpc nmsg))
	      (set! *transport-type* transport)
	      (begin
		(debug:print 0 "ERROR: Unrecognised transport " transport)
		(exit))))
	(let ((linktree (configf:lookup *configdat* "setup" "linktree"))) ;; link tree is critical
	  (if linktree
	      (if (not (file-exists? linktree))
		  (begin
		    (handle-exceptions
		     exn
		     (begin

		(debug:print 0 "ERROR: linktree is required in your megatest.config [setup] section")
		(exit 1)))
	  (if (and *toppath*
		   (directory-exists? *toppath*))
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (begin
		(debug:print 0 "ERROR: failed to find the top path to your Megatest area.")
		(exit 1)))
	  )))
  *toppath*)

(define (launch:cache-config)
  ;; if we have a linktree and -runtests and -target and the directory exists dump the config
  ;; to megatest-(current-seconds).cfg and symlink it to megatest.cfg
  (if (and *configdat* 
	   (args:get-arg "-runtests"))
      (let* ((linktree (get-environment-variable "MT_LINKTREE"))
	     (target   (common:args-get-target))
	     (runname  (or (args:get-arg "-runname")
			   (args:get-arg ":runname")))
	     (fulldir  (conc linktree "/"
			     target "/"
			     runname)))
	(debug:print-info 0 "Have -runtests with target=" target ", runname=" runname ", fulldir=" fulldir)
	(if (file-exists? linktree) ;; can't proceed without linktree
	    (begin
	      (if (not (file-exists? fulldir))
		  (create-directory fulldir #t)) ;; need to protect with exception handler 
	      (if (and target
		       runname
		       (file-exists? fulldir))
		  (let ((tmpfile  (conc fulldir "/.megatest.cfg." (current-seconds)))
			(targfile (conc fulldir "/.megatest.cfg")))
		    (debug:print-info 0 "Caching megatest.config in " fulldir "/.megatest.cfg")
		    (configf:write-alist *configdat* tmpfile)
		    (system (conc "ln -sf " tmpfile " " targfile))
		    )))))))

(define (get-best-disk confdat)
  (let* ((disks    (hash-table-ref/default confdat "disks" #f))


	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))
		     (string->number (or m "10000")))))
    (if disks 





	(let ((res (common:get-disk-with-most-free-space disks minspace))) ;; min size of 1000, seems tad dumb












	  (if res





	      (cdr res)
	      (begin
		(if (common:low-noise-print 20 "no valid disks")
		    (debug:print 0 "ERROR: No valid disks found in megatest.config. Please add some to your [disks] section and ensure the directory exists!"))
		(exit 1)))))))

;; Desired directory structure:
;;
;;  <linkdir> - <target> - <testname> -.
;;                                     |
;;                                     v
;;  <rundir>  -  <target>  -    <testname> -|- <itempath(s)>

    (if (not (file-exists? linktree))
	(begin
	  (debug:print 0 "WARNING: linktree did not exist! Creating it now at " linktree)
	  (create-directory linktree #t))) ;; (system (conc "mkdir -p " linktree))))
    ;; create the directory for the tests dir links, this is needed no matter what...
    (if (and (not (directory-exists? lnkbase))
	     (not (file-exists? lnkbase)))
	(handle-exceptions
	 exn
	 (begin
	   (debug:print "ERROR: Problem creating linktree base at " lnkbase)
	   (print-error-message exn (current-error-port)))
	 (create-directory lnkbase #t)))
    
    ;; update the toptest record with its location rundir, cache the path
    ;; This wass highly inefficient, one db write for every subtest, potentially
    ;; thousands of unnecessary updates, cache the fact it was set and don't set it 
    ;; again. 

    ;; Now create the link from the test path to the link tree, however

				lnkpath)
			    testname "")
	  ;; (rmt:general-call 'test-set-rundir run-id lnkpath testname "") ;; toptest-path)
	  (if (or (not curr-test-path)
		  (not (directory-exists? toptest-path)))
	      (begin
		(debug:print-info 2 "Creating " toptest-path " and link " lnkpath)
		(handle-exceptions
		 exn
		 #f ;; don't care to catch and deal with errors here for now.
		 (create-directory toptest-path #t))
		(hash-table-set! *toptest-paths* testname toptest-path)))))

    ;; The toptest path has been created, the link to the test in the linktree has
    ;; been created. Now, if this is an iterated test the real test dir must be created
    (if (not not-iterated) ;; this is an iterated test
	(begin ;; (let ((lnktarget (conc lnkpath "/" item-path)))
	  (debug:print 2 "Setting up sub test run area")

    (set! mt-bindir-path (pathname-directory remote-megatest))
    (if launcher (set! launcher (string-split launcher)))
    ;; set up the run work area for this test
    (if (and (args:get-arg "-preclean") ;; user has requested to preclean for this run
	     (not (member (db:test-get-rundir testinfo)(list "n/a" "/tmp/badname")))) ;; n/a is a placeholder and thus not a read dir
	(begin
	  (debug:print-info 0 "attempting to preclean directory " (db:test-get-rundir testinfo) " for test " test-name "/" item-path)
	  (runs:remove-test-directory testinfo 'remove-data-only))) ;; remove data only, do not perturb the record

    ;; prevent overlapping actions - set to LAUNCHED as early as possible
    ;;
    (tests:test-set-status! run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
    (rmt:roll-up-pass-fail-counts run-id test-name item-path "LAUNCHED")
    (set! diskpath (get-best-disk *configdat*))
    (if diskpath

;; Always use two or four digit decimal
;; 1.01, 1.02...1.10,1.11,1,1101 ... 1.99,2.00..

(declare (unit megatest-version))

(define megatest-version 1.6007)

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

;; (include "common.scm")
;; (include "megatest-version.scm")

(use sqlite3 srfi-1 posix regex regex-case srfi-69 base64 format readline apropos json http-client directory-utils rpc ;; (srfi 18) extras)
     http-client srfi-18) ;;  zmq extras)

;; Added for csv stuff - will be removed
;;
(use sparse-vectors)

(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))
(import (prefix rpc rpc:))

;; (use zmq)

(declare (uses common))
(declare (uses megatest-version))
(declare (uses margs))
(declare (uses runs))
(declare (uses launch))
(declare (uses server))
(declare (uses client))
(declare (uses tests))
(declare (uses genexample))
(declare (uses daemon))
(declare (uses db))


(declare (uses tdb))
(declare (uses mt))
(declare (uses api))
(declare (uses tasks)) ;; only used for debugging.

(define *db* #f) ;; this is only for the repl, do not use in general!!!!

  -import-megatest.db     : migrate a database from v1.55 series to v1.60 series
  -sync-to-megatest.db    : migrate data back to megatest.db
  -update-meta            : update the tests metadata for all tests
  -setvars VAR1=val1,VAR2=val2 : Add environment variables to a run NB// these are
                                 overwritten by values set in config files.
  -server -|hostname      : start the server (reduces contention on megatest.db), use
                            - to automatically figure out hostname
  -transport http|zmq     : use http or zmq for transport (default is http) 
  -daemonize              : fork into background and disconnect from stdin/out
  -log logfile            : send stdout and stderr to logfile
  -list-servers           : list the servers 
  -stop-server id         : stop server specified by id (see output of -list-servers), use
                            0 to kill all
  -repl                   : start a repl (useful for extending megatest)
  -load file.scm          : load and run file.scm
  -mark-incompletes       : find and mark incomplete tests
  -ping run-id|host:port  : ping server, exit with 0 if found

Utilities
  -env2file fname         : write the environment to fname.csh and fname.sh
  -refdb2dat refdb        : convert refdb to sexp or to format specified by -dumpmode
                            formats: perl, ruby, sqlite3, csv (for csv the -o param
                            will substitute %s for the sheet name in generating 
                            multiple sheets)
  -o                      : output file for refdb2dat (defaults to stdout)
  -archive cmd            : archive runs specified by selectors to one of disks specified
                            in the [archive-disks] section.
                            cmd: keep-html, restore, save, save-remove

Spreadsheet generation
  -extract-ods fname.ods  : extract an open document spreadsheet from the database
  -pathmod path           : insert path, i.e. path/runame/itempath/logfile.html
                            will clear the field if no rundir/testname/itempath/logfile
                            if it contains forward slashes the path will be converted
                            to windows style

			":expected"
			":tol"
			":units"
			;; misc
			"-start-dir"
			"-server"
			"-stop-server"
			"-transport"
			"-kill-server"
			"-port"
			"-extract-ods"
			"-pathmod"
			"-env2file"
			"-setvars"
			"-set-state-status"
			"-set-run-status"
			"-debug" ;; for *verbosity* > 2
			"-gen-megatest-test"
			"-override-timeout"
			"-test-files"  ;; -test-paths is for listing all
			"-load"        ;; load and exectute a scheme file
			"-section"
			"-var"
			"-dumpmode"
			"-run-id"
			"-ping"
			"-refdb2dat"
			"-o"
			"-log"
			"-archive"
			) 
		 (list  "-h" "-help" "--help"
			"-version"
		        "-force"
		        "-xterm"
		        "-showkeys"
		        "-show-keys"
		        "-test-status"
			"-set-values"
			"-load-test-data"
			"-summarize-items"
		        "-gui"
			"-daemonize"
			"-preclean"
			;; misc

			"-repl"
			"-lock"
			"-unlock"
			"-list-servers"
                        "-run-wait"      ;; wait on a run to complete (i.e. no RUNNING)

			;; misc queries

			"-q" ;; quiet 0, errors/warnings only
		       )
		 args:arg-hash
		 0))

;; The watchdog is to keep an eye on things like db sync etc.
;;
(define *time-zero* (current-seconds))
(define *watchdog*
  (make-thread 
   (lambda ()
     (thread-sleep! 0.05) ;; delay for startup
     (let ((legacy-sync (configf:lookup *configdat* "setup" "megatest-db"))
	   (debug-mode  (debug:debug-mode 1))
	   (last-time   (current-seconds)))
       (let loop ()
	 ;; sync for filesystem local db writes
	 ;;
	 (let ((start-time      (current-seconds))
	       (servers-started (make-hash-table)))
	   (for-each 
	    (lambda (run-id)

			  (debug:print-info 0 "Sync of newdb to olddb for run-id " run-id " completed in " sync-time " seconds")))
		    ;; (if (> sync-time 10) ;; took more than ten seconds, start a server for this run
		    ;;     (begin
		    ;;       (debug:print-info 0 "Sync is taking a long time, start up a server to assist for run " run-id)
		    ;;       (server:kind-run run-id)))))
		    (hash-table-delete! *db-local-sync* run-id)))
	      (mutex-unlock! *db-multi-sync-mutex*))
	    (hash-table-keys *db-local-sync*))
	   (if (and debug-mode
		    (> (- start-time last-time) 60))
	       (begin
		 (set! last-time start-time)
		 (debug:print-info 1 "timestamp -> " (seconds->time-string (current-seconds)) ", time since start -> " (seconds->hr-min-sec (- (current-seconds) *time-zero*))))))
	 
	 ;; keep going unless time to exit
	 ;;
	 (if (not *time-to-exit*)
	     (let delay-loop ((count 0))
	       (if (and (not *time-to-exit*)
			(< count 11)) ;; aprox 5-6 seconds
		   (begin
		     (thread-sleep! 1)
		     (delay-loop (+ count 1))))
	       (loop))))))
   "Watchdog thread"))

(thread-start! *watchdog*)

(if (args:get-arg "-log")
    (let ((oup (open-output-file (args:get-arg "-log"))))
      (debug:print-info 0 "Sending log output to " (args:get-arg "-log"))
      (current-error-port oup)

	(map (lambda (x)
	       (string-intersperse 
		x
		" => "))
	     (common:get-disks *configdat*))
	"\n"))
      (set! *didsomething* #t)))

(define (make-sparse-array)
  (let ((a (make-sparse-vector)))
    (sparse-vector-set! a 0 (make-sparse-vector))
    a))

(define (sparse-array? a)
  (and (sparse-vector? a)
       (sparse-vector? (sparse-vector-ref a 0))))

(define (sparse-array-ref a x y)
  (let ((row (sparse-vector-ref a x)))
    (if row
	(sparse-vector-ref row y)
	#f)))

(define (sparse-array-set! a x y val)
  (let ((row (sparse-vector-ref a x)))
    (if row
	(sparse-vector-set! row y val)
	(let ((new-row (make-sparse-vector)))
	  (sparse-vector-set! a x new-row)
	  (sparse-vector-set! new-row y val)))))

;; csv processing record
(define (make-refdb:csv)
  (vector 
   (make-sparse-array)
   (make-hash-table)
   (make-hash-table)
   0
   0))
(define-inline (refdb:csv-get-svec     vec)    (vector-ref  vec 0))
(define-inline (refdb:csv-get-rows     vec)    (vector-ref  vec 1))
(define-inline (refdb:csv-get-cols     vec)    (vector-ref  vec 2))
(define-inline (refdb:csv-get-maxrow   vec)    (vector-ref  vec 3))
(define-inline (refdb:csv-get-maxcol   vec)    (vector-ref  vec 4))
(define-inline (refdb:csv-set-svec!    vec val)(vector-set! vec 0 val))
(define-inline (refdb:csv-set-rows!    vec val)(vector-set! vec 1 val))
(define-inline (refdb:csv-set-cols!    vec val)(vector-set! vec 2 val))
(define-inline (refdb:csv-set-maxrow!  vec val)(vector-set! vec 3 val))
(define-inline (refdb:csv-set-maxcol!  vec val)(vector-set! vec 4 val))

(define (get-dat results sheetname)
  (or (hash-table-ref/default results sheetname #f)
      (let ((tmp-vec  (make-refdb:csv)))
	(hash-table-set! results sheetname tmp-vec)
	tmp-vec)))

(if (args:get-arg "-refdb2dat")
    (let* ((input-db (args:get-arg "-refdb2dat"))
	   (out-file (args:get-arg "-o"))
	   (out-fmt  (or (args:get-arg "-dumpmode") "scheme"))
	   (out-port (if (and out-file 
			      (not (member out-fmt '("sqlite3" "csv"))))
			 (open-output-file out-file)
			 (current-output-port)))
	   (res-data (configf:read-refdb input-db))
	   (data     (car res-data))
	   (msg      (cadr res-data)))
      (if (not data)
	  (debug:print 0 "Bad input? data=" data) ;; some error occurred
	  (with-output-to-port out-port
	    (lambda ()
	      (case (string->symbol out-fmt)
		((scheme)(pp data))
		((perl)
		 ;; (print "%hash = (")
		 ;;        key1 => 'value1',

		    (print "data[\"" sheetname "\"][\"" sectionname "\"][\"" varname "\"] = \"" val "\""))
		  initproc1:
		  (lambda (sheetname)
		    (print "data[\"" sheetname "\"] = {}"))
		  initproc2:
		  (lambda (sheetname sectionname)
		    (print "data[\"" sheetname "\"][\"" sectionname "\"] = {}"))))
		((csv)
		 (let* ((results  (make-hash-table)) ;; (make-sparse-array)))
			(row-cols (make-hash-table))) ;; hash of hashes where section => ht { row-<name> => num or col-<name> => num
		   ;; (print "data=")
		   ;; (pp data)
		   (configf:map-all-hier-alist
		    data
		    (lambda (sheetname sectionname varname val)
		      ;; (print "sheetname: " sheetname ", sectionname: " sectionname ", varname: " varname ", val: " val)
		      (let* ((dat      (get-dat results sheetname))
			     (vec      (refdb:csv-get-svec dat))
			     (rownames (refdb:csv-get-rows dat))
			     (colnames (refdb:csv-get-cols dat))
			     (currrown (hash-table-ref/default rownames varname #f))
			     (currcoln (hash-table-ref/default colnames sectionname #f))
			     (rown     (or currrown 
					   (let* ((lastn   (refdb:csv-get-maxrow dat))
						  (newrown (+ lastn 1)))
					     (refdb:csv-set-maxrow! dat newrown)
					     newrown)))
			     (coln     (or currcoln 
					   (let* ((lastn   (refdb:csv-get-maxcol dat))
						  (newcoln (+ lastn 1)))
					     (refdb:csv-set-maxcol! dat newcoln)
					     newcoln))))
			(if (not (sparse-array-ref vec 0 coln)) ;; (eq? rown 0)
			    (begin
			      (sparse-array-set! vec 0 coln sectionname)
			      ;; (print "sparse-array-ref " 0 "," coln "=" (sparse-array-ref vec 0 coln))
			      ))
			(if (not (sparse-array-ref vec rown 0)) ;; (eq? coln 0)
			    (begin
			      (sparse-array-set! vec rown 0 varname)
			      ;; (print "sparse-array-ref " rown "," 0 "=" (sparse-array-ref vec rown 0))
			      ))
			(if (not currrown)(hash-table-set! rownames varname rown))
			(if (not currcoln)(hash-table-set! colnames sectionname coln))
			;; (print "dat=" dat ", rown=" rown ", coln=" coln)
			(sparse-array-set! vec rown coln val)
			;; (print "sparse-array-ref " rown "," coln "=" (sparse-array-ref vec rown coln))
			)))
		   (for-each
		    (lambda (sheetname)
		      (let* ((sheetdat (get-dat results sheetname))
			     (svec     (refdb:csv-get-svec sheetdat))
			     (maxrow   (refdb:csv-get-maxrow sheetdat))
			     (maxcol   (refdb:csv-get-maxcol sheetdat))
			     (fname    (if out-file 
					   (string-substitute "%s" sheetname out-file) ;; "/foo/bar/%s.csv")
					   (conc sheetname ".csv"))))
			(with-output-to-file fname
			  (lambda ()
			    ;; (print "Sheetname: " sheetname)
			    (let loop ((row       0)
				       (col       0)
				       (curr-row '())
				       (result   '()))
			      (let* ((val (sparse-array-ref svec row col))
				     (disp-val (if val
						   (conc "\"" val "\"")
						   "")))
				(if (> col 0)(display ","))
				(display disp-val)
				(cond
				 ((> row maxrow)(display "\n") result)
				 ((>= col maxcol)
				  (display "\n")
				  (loop (+ row 1) 0 '() (append result (list curr-row))))
				 (else
				  (loop row (+ col 1) (append curr-row (list val)) result)))))))))
		    (hash-table-keys results))))
		((sqlite3)
		 (let* ((db-file   (or out-file (pathname-file input-db)))
			(db-exists (file-exists? db-file))
			(db        (sqlite3:open-database db-file)))
		   (if (not db-exists)(sqlite3:execute db "CREATE TABLE data (sheet,section,var,val);"))
		   (configf:map-all-hier-alist
		    data

      ))

(if (args:get-arg "-ping")
    (let* ((run-id        (string->number (args:get-arg "-run-id")))
	   (host:port     (args:get-arg "-ping")))
      (server:ping run-id host:port)))

;;       (set! *did-something* #t)
;; 	      (begin
;; 		(print ((rpc:procedure 'testing (car host-port)(cadr host-port))))
;; 		(case (server:get-transport)
;; 		  ((http)(http:ping run-id host-port))
;; 		  ((rpc) (rpc:procedure 'server:login (car host-port)(cadr host-port));;  *toppath*)) ;; (rpc-transport:ping  run-id (car host-port)(cadr host-port)))
;; 		  (else  (debug:print 0 "ERROR: No transport set")(exit)))))

;;======================================================================
;; Start the server - can be done in conjunction with -runall or -runtests (one day...)
;;   we start the server if not running else start the client thread
;;======================================================================

(if (args:get-arg "-server")

    ;;  Setup client for all expect listed here
    (if (null? (lset-intersection 
		     equal?
		     (hash-table-keys args:arg-hash)
		     '("-list-servers"
		       "-stop-server"
		       "-show-cmdinfo"
		       "-list-runs"
		       "-ping")))
	(if (launch:setup-for-run)
	    (let ((run-id    (and (args:get-arg "-run-id")
				  (string->number (args:get-arg "-run-id")))))
	      ;; (set! *fdb*   (filedb:open-db (conc *toppath* "/db/paths.db")))
	      ;; if not list or kill then start a client (if appropriate)
	      (if (or (args-defined? "-h" "-version" "-gen-megatest-area" "-gen-megatest-test")
		      (eq? (length (hash-table-keys args:arg-hash)) 0))

;;======================================================================
;; Get test paths matching target, runname, and testpatt
(if (or (args:get-arg "-test-files")(args:get-arg "-test-paths"))
    ;; if we are in a test use the MT_CMDINFO data
    (if (getenv "MT_CMDINFO")
	(let* ((startingdir (current-directory))
	       (cmdinfo   (common:read-encoded-string (getenv "MT_CMDINFO")))
	       (transport (assoc/default 'transport cmdinfo))
	       (testpath  (assoc/default 'testpath  cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (state     (args:get-arg ":state"))

		       paths))))))

;;======================================================================
;; Archive tests
;;======================================================================
;; Archive tests matching target, runname, and testpatt
(if (args:get-arg "-archive")






























    ;; else do a general-run-call
    (general-run-call 
     "-archive"

     "Archive"
     (lambda (target runname keys keyvals)



       (operate-on 'archive))))

;;======================================================================
;; Extract a spreadsheet from the runs database
;;======================================================================

(if (args:get-arg "-extract-ods")
    (general-run-call

(define (megatest:step step state status logfile msg)
  (if (not (getenv "MT_CMDINFO"))
      (begin
	(debug:print 0 "ERROR: MT_CMDINFO env var not set, -step must be called *inside* a megatest invoked environment!")
	(exit 5))
      (let* ((cmdinfo   (common:read-encoded-string (getenv "MT_CMDINFO")))
	     (transport (assoc/default 'transport cmdinfo))
	     (testpath  (assoc/default 'testpath  cmdinfo))
	     (test-name (assoc/default 'test-name cmdinfo))
	     (runscript (assoc/default 'runscript cmdinfo))
	     (db-host   (assoc/default 'db-host   cmdinfo))
	     (run-id    (assoc/default 'run-id    cmdinfo))
	     (test-id   (assoc/default 'test-id   cmdinfo))
	     (itemdat   (assoc/default 'itemdat   cmdinfo))

	      (debug:print 0 "ERROR: You must specify :state and :status with every call to -step")
	      (exit 6))))))

(if (args:get-arg "-step")
    (begin
      (megatest:step 
       (args:get-arg "-step")
       (or (args:get-arg "-state")(args:get-arg ":state"))
       (or (args:get-arg "-status")(args:get-arg ":status"))
       (args:get-arg "-setlog")
       (args:get-arg "-m"))
      ;; (if db (sqlite3:finalize! db))
      (set! *didsomething* #t)))
    
(if (or (args:get-arg "-setlog")       ;; since setting up is so costly lets piggyback on -test-status
	;;     (not (args:get-arg "-step")))  ;; -setlog may have been processed already in the "-step" previous
	;;     NEW POLICY - -setlog sets test overall log on every call.
	(args:get-arg "-set-toplog")
	(args:get-arg "-test-status")
	(args:get-arg "-set-values")
	(args:get-arg "-load-test-data")
	(args:get-arg "-runstep")
	(args:get-arg "-summarize-items"))
    (if (not (getenv "MT_CMDINFO"))
	(begin
	  (debug:print 0 "ERROR: MT_CMDINFO env var not set, commands -test-status, -runstep and -setlog must be called *inside* a megatest environment!")
	  (exit 5))
	(let* ((startingdir (current-directory))
	       (cmdinfo   (common:read-encoded-string (getenv "MT_CMDINFO")))
	       (transport (assoc/default 'transport cmdinfo))
	       (testpath  (assoc/default 'testpath  cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))

;; Copyright 2006-2014, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')

;;======================================================================
;; MLAUNCH
;;
;;   take jobs from the given queue and keep launching them keeping
;;   the cpu load at the targeted level
;;
;;======================================================================

(use sqlite3 srfi-1 posix regex regex-case srfi-69 dot-locking format)
(import (prefix sqlite3 sqlite3:))

(declare (unit mlaunch))
(declare (uses db))
(declare (uses common))

;; (if (args:get-arg "-host")
;;     (begin
;;       (set! *runremote* (string-split (args:get-arg "-host" ":")))
;;       (client:launch))
;;     (client:launch))

;; ease debugging by loading ~/.dashboardrc
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.dashboardrc")))
  (if (file-exists? debugcontrolf)
      (load debugcontrolf)))

(define *dbdir* (conc (configf:lookup *configdat* "setup" "linktree") "/.db"))
(define *dbstruct-local*  (make-dbr:dbstruct path:  *dbdir*
					     local: #t))
(define *db-file-path* (db:dbfile-path 0))

;; HACK ALERT: this is a hack, please fix.

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

(require-extension (srfi 18) extras tcp s11n)

(use sqlite3 srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest)
(import (prefix sqlite3 sqlite3:))

;; (use nanomsg)

(declare (unit nmsg-transport))

(declare (uses common))
(declare (uses db))
(declare (uses tests))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
(declare (uses server))

(include "common_records.scm")
(include "db_records.scm")

;; Transition to pub --> sub with pull <-- push
;;
;;   1. client sends request to server via push to the pull port
;;   2. server puts request in queue or processes immediately as appropriate
;;   3. server puts responses from completed requests into pub port 
;;
;; TODO
;;
;; Done Tested
;; [x]  [ ]    1. Add columns pullport pubport to servers table
;; [x]  [ ]    2. Add rm of monitor.db if older than 11/12/2012 
;; [x]  [ ]    3. Add create of pullport and pubport with finding of available ports
;; [x]  [ ]    4. Add client compose of request
;; [x]  [ ]        - name of client: testname/itempath-test_id-hostname 
;; [x]  [ ]        - name of request: callname, params
;; [x]  [ ]        - request key: f(clientname, callname, params)
;; [x]  [ ]    5. Add processing of subscription hits
;; [x]  [ ]        - done when get key 
;; [x]  [ ]        - return results
;; [x]  [ ]    6. Add timeout processing
;; [x]  [ ]        - after 60 seconds
;; [ ]  [ ]            i. check server alive, connect to new if necessary
;; [ ]  [ ]           ii. resend request
;; [ ]  [ ]    7. Turn self ping back on

(define (nmsg-transport:make-server-url hostport #!key (bindall #f))
  (if (not hostport)
      #f
      (conc "tcp://" (if bindall "*" (car hostport)) ":" (cadr hostport))))

(define *server-loop-heart-beat* (current-seconds))
(define *heartbeat-mutex* (make-mutex))

;;======================================================================
;; S E R V E R
;;======================================================================

(define (nmsg-transport:run dbstruct hostn run-id server-id #!key (retrynum 1000))
  (debug:print 2 "Attempting to start the server ...")
  (let* ((start-port      (portlogger:open-run-close portlogger:find-port))
	 (server-thread   (make-thread (lambda ()
					 (nmsg-transport:try-start-server dbstruct run-id start-port server-id))
				       "server thread"))
	 (tdbdat          (tasks:open-db)))
    (thread-start! server-thread)
    (thread-sleep! 0.1)
    (if (nmsg-transport:ping hostn start-port timeout: 2 expected-key: (current-process-id))
	(let ((interface (if (equal? hostn "-")(get-host-name) hostn)))
	  (tasks:server-set-interface-port (db:delay-if-busy tdbdat) server-id interface start-port)
	  (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "dbprep")
	  (set! *server-info* (list hostn start-port)) ;; probably not needed anymore? currently used by keep-running
	  (thread-sleep! 3) ;; give some margin for queries to complete before switching from file based access to server based access
	  ;; (set! *inmemdb*  dbstruct)
	  (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "running")
	  (thread-start! (make-thread
			  (lambda ()(nmsg-transport:keep-running server-id run-id))
			  "keep running"))
	  (thread-join! server-thread))
	(if (> retrynum 0)
	    (begin
	      (debug:print 0 "WARNING: Failed to connect to server (self) on host " hostn ":" start-port ", trying again.")
	      (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id "failed to start, never received server alive signature")
	      (portlogger:open-run-close portlogger:set-failed start-port)
	      (nmsg-transport:run dbstruct hostn run-id server-id))
	    (begin
	      (debug:print 0 "ERROR: could not find an open port to start server on. Giving up")
	      (exit 1))))))

(define (nmsg-transport:try-start-server dbstruct run-id portnum server-id)
  (let ((repsoc (nn-socket 'rep)))
    (nn-bind repsoc (conc "tcp://*:" portnum))
    (let loop ((msg-in (nn-recv repsoc)))
      (let* ((dat    (db:string->obj msg-in transport: 'nmsg)))
	(debug:print 0 "server, received: " dat)
	(let ((result (api:execute-requests dbstruct dat)))
	  (debug:print 0 "server, sending: " result)
	  (nn-send repsoc (db:obj->string result  transport: 'nmsg)))
	(loop (nn-recv repsoc))))))

;; all routes though here end in exit ...
;;
(define (nmsg-transport:launch run-id)
  (let* ((tdbdat   (tasks:open-db))
	 (dbstruct (db:setup run-id))
	 (hostn    (or (args:get-arg "-server") "-")))
    (set! *run-id*   run-id)
    (set! *inmemdb* dbstruct)
    ;; with nbfake daemonize isn't really needed
    ;;
    ;; (if (args:get-arg "-daemonize")
    ;;     (begin
    ;;       (daemon:ize)
    ;;       (if *alt-log-file* ;; we should re-connect to this port, I think daemon:ize disrupts it
    ;;           (begin
    ;;     	(current-error-port *alt-log-file*)
    ;;     	(current-output-port *alt-log-file*)))))
    (if (server:check-if-running run-id)
	(begin
	  (debug:print-info 0 "Server for run-id " run-id " already running")
	  (exit 0)))
    (let loop ((server-id (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id))
	       (remtries  4))
      (if (not server-id)
	  (if (> remtries 0)
	      (begin
		(thread-sleep! 2)
		(if (not (server:check-if-running run-id))
		    (loop (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id)
			  (- remtries 1))
		    (begin
		      (debug:print-info 0 "Another server took the slot, exiting")
		      (exit 0))))
	      (begin
		;; since we didn't get the server lock we are going to clean up and bail out
		(debug:print-info 2 "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
		(tasks:server-delete-records-for-this-pid (db:delay-if-busy tdbdat) " http-transport:launch")
		))
	  ;; locked in a server id, try to start up
	  (nmsg-transport:run dbstruct hostn run-id server-id))
      (set! *didsomething* #t)
      (exit))))

;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

(define (nmsg-transport:mk-signature)
  (message-digest-string (md5-primitive) 
			 (with-output-to-string
			   (lambda ()
			     (write (list (current-directory)
					  (argv)))))))

;;======================================================================
;; C L I E N T S
;;======================================================================

;; ping the server at host:port
;;   return the open socket if successful (return-socket == #t)
;;   expect the key expected-key returned in payload
;;   send our-key or #f as payload
;;
(define (nmsg-transport:ping hostn port #!key (timeout 3)(return-socket #t)(expected-key #f)(our-key #f)(socket #f))
  ;; send a random number along with pid and check that we get it back
  (let* ((host    (if (or (not hostn)
			  (equal? hostn "-")) ;; use localhost
		      (get-host-name)
		      hostn))
	 (req     (or socket
		      (let ((soc (nn-socket 'req)))
			(nn-connect soc (conc "tcp://" host ":" port))
			soc)))
	 (success #t)
	 (dat     (vector "ping" our-key))
	 (result  (condition-case 
		   (nmsg-transport:client-api-send-receive-raw req dat timeout: timeout)
		   ((timeout)(set! success #f) #f)))
	 (key     (if success 
		      (vector-ref result 1)
		      #f)))
    (debug:print 0 "success=" success ", key=" key ", expected-key=" expected-key ", equal? " (equal? key expected-key))
    (if (and success
	     (or (not expected-key) ;; just getting a reply is good enough then
		 (equal? key expected-key)))
	(if return-socket
	    req
	    (begin
	      (if (not socket)(nn-close req)) ;; don't want a side effect of closing socket if handed it
	      #t))
	(begin
	  (if (not socket)(nn-close req)) ;; failed to ping, close socket as side effect
	  #f))))

;; send data to server, wait max of timeout seconds for a response.
;; return #( success/fail result )
;;
;; for effiency it is easier to do the obj->string and string->obj here.
;;
(define (nmsg-transport:client-api-send-receive-raw socreq indat #!key (enable-send #t)(timeout 25))
  (let* ((success     #f)
	 (result      #f)
	 (keepwaiting #t)
	 (dat         (db:obj->string indat transport: 'nmsg))
	 (send-recv   (make-thread
		       (lambda ()
			 (nn-send socreq dat)
			 (let* ((res (nn-recv socreq)))
			   (set! success #t)
			   (set! result (db:string->obj res transport: 'nmsg))))
		       "send-recv"))
	 (timeout     (make-thread
		       (lambda ()
			 (let loop ((count 0))
			   (thread-sleep! 1)
			   (debug:print-info 1 "send-receive-raw, still waiting after " count " seconds...")
			   (if (and keepwaiting (< count timeout)) ;; yes, this is very aproximate
			       (loop (+ count 1))))
			 (if keepwaiting
			     (begin
			       (print "timeout waiting for ping")
			       (thread-terminate! send-recv))))
		       "timeout")))
    ;; replace with condition-case?
    (handle-exceptions
     exn
     (set! result "timeout")
     (thread-start! timeout)
     (thread-start! send-recv)
     (thread-join! send-recv)
     (if success (thread-terminate! timeout)))
    ;; raise timeout error if timed out
    (if success
	(if (and (vector? result)
		 (vector-ref result 0)) ;; did it fail at the server?
	    result                ;; nope, all good
	    (begin
	      (debug:print 0 "ERROR: error occured at server, info=" (vector-ref result 2))
	      (debug:print 0 " client call chain:")
	      (print-call-chain (current-error-port))
	      (debug:print 0 " server call chain:")
	      (pp (vector-ref result 1) (current-error-port))
	      (signal (vector-ref result 0))))
	(signal (make-composite-condition
		 (make-property-condition 'timeout 'message "nmsg-transport:client-api-send-receive-raw timed out talking to server"))))))

;; run nmsg-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (nmsg-transport:keep-running server-id run-id)
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (let* ((server-info (let loop ()
                        (let ((sdat #f))
                          (mutex-lock! *heartbeat-mutex*)
                          (set! sdat *server-info*)
                          (mutex-unlock! *heartbeat-mutex*)
                          (if sdat 
			      (begin
				(debug:print-info 0 "keep-running got sdat=" sdat)
				sdat)
                              (begin
                                (thread-sleep! 0.5)
                                (loop))))))
         (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (tdbdat      (tasks:open-db))
	 (server-timeout (let ((tmo (configf:lookup  *configdat* "server" "timeout")))
			   (if (and (string? tmo)
				    (string->number tmo))
			       (* 60 60 (string->number tmo))
			       ;; (* 3 24 60 60) ;; default to three days
			       (* 60 1)         ;; default to one minute
			       ;; (* 60 60 25)      ;; default to 25 hours
			       ))))
    (print "Keep-running got server pid " server-id ", using iface " iface " and port " port)
    (let loop ((count 0))
      (thread-sleep! 4) ;; no need to do this very often
      ;; NB// sync currently does NOT return queue-length
      (let () ;; (queue-len (cdb:client-call server-info 'sync #t 1)))
      ;; (print "Server running, count is " count)
        (if (< count 1) ;; 3x3 = 9 secs aprox
            (loop (+ count 1)))
        
        (mutex-lock! *heartbeat-mutex*)
        (set! last-access *last-db-access*)
        (mutex-unlock! *heartbeat-mutex*)
	(db:sync-touched *inmemdb* run-id force-sync: #t)
        (if (and *server-run*
	       (> (+ last-access server-timeout)
		  (current-seconds)))
            (begin
              (debug:print-info 0 "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
              (loop 0))
            (begin
              (debug:print-info 0 "Starting to shutdown the server.")
              (set! *time-to-exit* #t)
	      (db:sync-touched *inmemdb* run-id force-sync: #t)
              (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id " http-transport:keep-running")
              (debug:print-info 0 "Server shutdown complete. Exiting")
              (exit)
	      ))))))

;;======================================================================
;; C L I E N T S
;;======================================================================

(define (nmsg-transport:client-connect iface portnum)
  (let* ((reqsoc      (nmsg-transport:ping iface portnum return-socket: #t)))
    (vector iface portnum #f #f #f (current-seconds) reqsoc)))

;; returns result, there is no sucess/fail flag - handled via excpections
;;
(define (nmsg-transport:client-api-send-receive run-id connection-info cmd param #!key (remtries 5))
  ;; NB// In the html version of this routine there is a call to 
  ;;      tasks:kill-server-run-id when there is an exception
  (mutex-lock! *http-mutex*)
  (let* ((packet  (vector cmd param))
	 (reqsoc  (http-transport:server-dat-get-socket connection-info))
	 (res     (nmsg-transport:client-api-send-receive-raw reqsoc packet)))
;;	 (status  (vector-ref rawres 0))
;;	 (result  (vector-ref rawres 1)))
    (mutex-unlock! *http-mutex*)
    res)) ;; (vector status (if status (db:string->obj result transport: 'nmsg) result))))
	
;;======================================================================
;; J U N K 
;;======================================================================

;; DO NOT USE
;;
(define (nmsg-transport:client-signal-handler signum)
  (handle-exceptions
   exn
   (debug:print " ... exiting ...")
   (let ((th1 (make-thread (lambda ()
			     (if (not *received-response*)
				 (receive-message* *runremote*))) ;; flush out last call if applicable
			   "eat response"))
	 (th2 (make-thread (lambda ()
			     (debug:print 0 "ERROR: Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
			     (thread-sleep! 3) ;; give the flush three seconds to do it's stuff
			     (debug:print 0 "       Done.")
			     (exit 4))
			   "exit on ^C timer")))
     (thread-start! th2)
     (thread-start! th1)
     (thread-join! th2))))

            (loop (let ((l (read-line fh)))
                    (if (eof-object? l) l (proc l)))
                  (append result (list curr)))
            result))))

;; here is an example line where the shell is sh or bash
;; "find / -print 2&>1 > findall.log"
(define (run-n-wait cmdline #!key (params #f)(print-cmd #f))
  (if print-cmd 
      (debug:print 0 
		   (if (string? print-cmd)
		       print-cmd
		       "")
		   cmdline
		   (if params
		       (string-intersperse params " ")
		       "")))
  (let ((pid (if params
		 (process-run cmdline params)
		 (process-run cmdline))))
    (let loop ((i 0))
      (let-values (((pid-val exit-status exit-code) (process-wait pid #t)))
         (if (eq? pid-val 0)
	     (begin
	       (thread-sleep! 2)
	       (loop (+ i 1)))
	     (values pid-val exit-status exit-code))))))

(use json format)

(declare (unit rmt))
(declare (uses api))
(declare (uses tdb))
(declare (uses http-transport))
(declare (uses nmsg-transport))

;;
;; THESE ARE ALL CALLED ON THE CLIENT SIDE!!!
;;

;; ;; For debugging add the following to ~/.megatestrc
;;a
;; (require-library trace)
;; (import trace)
;; (trace
;; rmt:send-receive
;; api:execute-requests
;; )

;; generate entries for ~/.megatestrc with the following
;;
;;  grep define ../rmt.scm | grep rmt: |perl -pi -e 's/\(define\s+\((\S+)\W.*$/\1/'|sort -u


;;======================================================================
;;  S U P P O R T   F U N C T I O N S
;;======================================================================

;; NOT USED?
;;
;; (define (rmt:call-transport run-id connection-info cmd jparams)
;;   (case (server:get-transport)
;;     ((rpc)  ( rpc-transport:client-api-send-receive run-id connection-info cmd jparams))
;;     ((http) (http-transport:client-api-send-receive run-id connection-info cmd jparams))
;;     ((fs)   ( fs-transport:client-api-send-receive run-id connection-info cmd jparams))
;;     ((zmq)  (zmq-transport:client-api-send-receive run-id connection-info cmd jparams))
;;     (else   ( rpc-transport:client-api-send-receive run-id connection-info cmd jparams))))

;;
(define (rmt:write-frequency-over-limit? cmd run-id)
  (and (not (member cmd api:read-only-queries))
       (let* ((tmprec (hash-table-ref/default *write-frequency* run-id #f))
	      (record (if tmprec tmprec 
			  (let ((v (vector (current-seconds) 0)))
			    (hash-table-set! *write-frequency* run-id v)
			    v)))
	      (count  (+ 1 (vector-ref record 1)))
	      (start  (vector-ref record 0))
	      (queries-per-second (/ (* count 1.0)
				     (max (- (current-seconds) start) 1))))
	 (vector-set! record 1 count)
	 (if (and (> count 10)
		  (> queries-per-second 10))
	     (begin
	       (debug:print-info 1 "db write rate too high, starting a server, count=" count " start=" start " run-id=" run-id " queries-per-second=" queries-per-second)
	       #t)
	     #f))))

;; if a server is either running or in the process of starting call client:setup
;; else return #f to let the calling proc know that there is no server available
;;
(define (rmt:get-connection-info run-id)
  (let ((cinfo (hash-table-ref/default *runremote* run-id #f)))
    (if cinfo
	cinfo
	;; NB// can cache the answer for server running for 10 seconds ...
	;;  ;; (and (not (rmt:write-frequency-over-limit? cmd run-id))
	(if (tasks:server-running-or-starting? (db:delay-if-busy (tasks:open-db)) run-id)
	    (client:setup run-id)
	    #f))))

(define *send-receive-mutex* (make-mutex)) ;; should have separate mutex per run-id


(define (rmt:send-receive cmd rid params #!key (attemptnum 1)) ;; start attemptnum at 1 so the modulo below works as expected
  ;; clean out old connections
  (mutex-lock! *db-multi-sync-mutex*)
  (let ((expire-time (- (current-seconds) (server:get-timeout) 10))) ;; don't forget the 10 second margin
    (for-each 
     (lambda (run-id)
       (let ((connection (hash-table-ref/default *runremote* run-id #f)))
         (if (and (vector? connection)
        	  (< (http-transport:server-dat-get-last-access connection) expire-time))
             (begin
               (debug:print-info 0 "Discarding connection to server for run-id " run-id ", too long between accesses")
               ;; SHOULD CLOSE THE CONNECTION HERE
	       (case *transport-type*
		 ((nmsg)(nn-close (http-transport:server-dat-get-socket 
				   (hash-table-ref *runremote* run-id)))))
               (hash-table-delete! *runremote* run-id)))))
     (hash-table-keys *runremote*)))
  (mutex-unlock! *db-multi-sync-mutex*)
  ;; (mutex-lock! *send-receive-mutex*)
  (let* ((run-id          (if rid rid 0))
	 (connection-info (rmt:get-connection-info run-id)))

    ;; the nmsg method does the encoding under the hood (the http method should be changed to do this also)
    (if connection-info
	;; use the server if have connection info
	(let* ((dat     (case *transport-type*
			  ((http)(condition-case
				  (http-transport:client-api-send-receive run-id connection-info cmd params)
				  ((commfail)(vector #f "communications fail"))
				  ((exn)(vector #f "other fail"))))
			  ((nmsg)(condition-case
				  (nmsg-transport:client-api-send-receive run-id connection-info cmd params)
				  ((timeout)(vector #f "timeout talking to server"))))
			  (else  (exit))))
	       (success (if (vector? dat) (vector-ref dat 0) #f))
	       (res     (if (vector? dat) (vector-ref dat 1) #f)))
	  (if (vector? connection-info)(http-transport:server-dat-update-last-access connection-info))
	  (if success


	      (begin


		;; (mutex-unlock! *send-receive-mutex*)
		(case *transport-type* 
		  ((http) res) ;; (db:string->obj res))


		  ((nmsg) res))) ;; (vector-ref res 1)))
	      (begin ;; let ((new-connection-info (client:setup run-id)))
		(debug:print 0 "WARNING: Communication failed, trying call to rmt:send-receive again.")
		;; (case *transport-type*
		;;   ((nmsg)(nn-close (http-transport:server-dat-get-socket connection-info))))
		(hash-table-delete! *runremote* run-id) ;; don't keep using the same connection
		;; NOTE: killing server causes this process to block forever. No idea why. Dec 2. 
		;; (if (eq? (modulo attemptnum 5) 0)
		;;     (tasks:kill-server-run-id run-id tag: "api-send-receive-failed"))
		;; (mutex-unlock! *send-receive-mutex*) ;; close the mutex here to allow other threads access to communications
		(tasks:start-and-wait-for-server (tasks:open-db) run-id 15)
		;; (nmsg-transport:client-api-send-receive run-id connection-info cmd param remtries: (- remtries 1))))))

		;; no longer killing the server in http-transport:client-api-send-receive
		;; may kill it here but what are the criteria?
		;; start with three calls then kill server
		;; (if (eq? attemptnum 3)(tasks:kill-server-run-id run-id))
		;; (thread-sleep! 2)
		(rmt:send-receive cmd run-id params attemptnum: (+ attemptnum 1)))))
	;; no connection info? try to start a server, or access locally if no
	;; server and the query is read-only
	;;
	;; Note: The tasks db was checked for a server in starting mode in the rmt:get-connection-info call
	;;
	(if (and (< attemptnum 15)

		 (member cmd api:write-queries))
	    (begin
	      (hash-table-delete! *runremote* run-id)
	      ;; (mutex-unlock! *send-receive-mutex*)
	      (tasks:start-and-wait-for-server (db:delay-if-busy (tasks:open-db)) run-id 10)
	      ;; (client:setup run-id) ;; client setup happens in rmt:get-connection-info
	      (thread-sleep! (random 5)) ;; give some time to settle and minimize collison?
	      (rmt:send-receive cmd rid params attemptnum: (+ attemptnum 1)))
	    (begin
	      ;; (debug:print 0 "ERROR: Communication failed!")
	      ;; (mutex-unlock! *send-receive-mutex*)
	      ;; (exit)
	      (rmt:open-qry-close-locally cmd run-id params)
	      )))))

(define (rmt:update-db-stats run-id rawcmd params duration)
  (mutex-lock! *db-stats-mutex*)
  (handle-exceptions
   exn
   (begin
     (debug:print 0 "WARNING: stats collection failed in update-db-stats")
     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
     (print "exn=" (condition->list exn))
     #f) ;; if this fails we don't care, it is just stats
   (let* ((cmd      (conc "run-id=" run-id " " (if (eq? rawcmd 'general-call) (car params) rawcmd)))
	  (stat-vec (hash-table-ref/default *db-stats* cmd #f)))
     (if (not (vector? stat-vec))
	 (let ((newvec (vector 0 0)))
	   (hash-table-set! *db-stats* cmd newvec)
	   (set! stat-vec newvec)))
     (vector-set! stat-vec 0 (+ (vector-ref stat-vec 0) 1))
     (vector-set! stat-vec 1 (+ (vector-ref stat-vec 1) duration))))
  (mutex-unlock! *db-stats-mutex*))

			     (if (> tot 10)
				 (cons newmax-cmd currmax)
				 (cons 'none 0))
			     (loop (car tal)(cdr tal) newmax-cmd currmax)))))))
    (mutex-unlock! *db-stats-mutex*)
    res))
	  
(define (rmt:open-qry-close-locally cmd run-id params #!key (remretries 5))
  (let* ((dbstruct-local (if *dbstruct-db*
			     *dbstruct-db*
			     (let* ((dbdir (conc    (configf:lookup *configdat* "setup" "linktree") "/.db"))
				    (db (make-dbr:dbstruct path:  dbdir local: #t)))
			       (set! *dbstruct-db* db)
			       db)))
	 (db-file-path   (db:dbfile-path 0))
	 ;; (read-only      (not (file-read-access? db-file-path)))
	 (start          (current-milliseconds))
	 (resdat         (api:execute-requests dbstruct-local (vector (symbol->string cmd) params)))
	 (success        (vector-ref resdat 0))
	 (res            (vector-ref resdat 1))
	 (duration       (- (current-milliseconds) start)))
    (if (not success)
	(if (> remretries 0)
	    (begin
	      (debug:print 0 "ERROR: local query failed. Trying again.")
	      (thread-sleep! (/ (random 5000) 1000)) ;; some random delay 
	      (rmt:open-qry-close-locally cmd run-id params remretries: (- remretries 1)))
	    (begin
	      (debug:print 0 "ERROR: too many retries in rmt:open-qry-close-locally, giving up")
	      #f))
	(begin
	  (rmt:update-db-stats run-id cmd params duration)
	  ;; mark this run as dirty if this was a write
	  (if (not (member cmd api:read-only-queries))
	      (let ((start-time (current-seconds)))
		(mutex-lock! *db-multi-sync-mutex*)
		;; (if (not (hash-table-ref/default *db-local-sync* run-id #f))
		;; just set it every time. Is a write more expensive than a read and does it matter?
		(hash-table-set! *db-local-sync* (or run-id 0) start-time) ;; the oldest "write"
		(mutex-unlock! *db-multi-sync-mutex*)))
	  res))))

(define (rmt:send-receive-no-auto-client-setup connection-info cmd run-id params)
  (let* ((run-id   (if run-id run-id 0))
	 ;; (jparams  (db:obj->string params)) ;; (rmt:dat->json-str params))
	 (res  	   (handle-exceptions
		    exn
		    #f
		    (http-transport:client-api-send-receive run-id connection-info cmd params))))
;;		    ((commfail) (vector #f "communications fail")))))
    (if (and res (vector-ref res 0))
	(vector-ref res 1) ;;; YES!! THIS IS CORRECT!! CHANGE IT HERE, THEN CHANGE rmt:send-receive ALSO!!!
	#f)))
;; 	(db:string->obj (vector-ref dat 1))
;; 	(begin
;; 	  (debug:print 0 "ERROR: rmt:send-receive-no-auto-client-setup failed, attempting to continue. Got " dat)
;; 	  dat))))

;; Wrap json library for strings (why the ports crap in the first place?)
(define (rmt:dat->json-str dat)
  (with-output-to-string 
    (lambda ()
      (json-write dat))))

;;  M I S C
;;======================================================================

(define (rmt:login run-id)
  (rmt:send-receive 'login run-id (list *toppath* megatest-version run-id *my-client-signature*)))

;; This login does no retries under the hood - it acts a bit like a ping.
;; Deprecated for nmsg-transport.
;;
(define (rmt:login-no-auto-client-setup connection-info run-id)
  (case *transport-type*
    ((http)(rmt:send-receive-no-auto-client-setup connection-info 'login run-id (list *toppath* megatest-version run-id *my-client-signature*)))
    ((nmsg)(nmsg-transport:client-api-send-receive run-id connection-info 'login (list *toppath* megatest-version run-id *my-client-signature*)))))

;; hand off a call to one of the db:queries statements
;; added run-id to make looking up the correct db possible 
;;
(define (rmt:general-call stmtname run-id . params)
  (rmt:send-receive 'general-call run-id (append (list stmtname run-id) params)))

(define (rmt:sync-inmem->db run-id)

  (if (number? run-id)
      (rmt:send-receive 'get-tests-for-run run-id (list run-id testpatt states statuses offset limit not-in sort-by sort-order qryvals))
      (begin
	(debug:print "ERROR: rmt:get-tests-for-run called with bad run-id=" run-id)
	(print-call-chain (current-error-port))
	'())))

;; get stuff via synchash 
(define (rmt:synchash-get run-id proc synckey keynum params)
  (rmt:send-receive 'synchash-get run-id (list run-id proc synckey keynum params)))

;; IDEA: Threadify these - they spend a lot of time waiting ...
;;
(define (rmt:get-tests-for-runs-mindata run-ids testpatt states status not-in)
  (let ((multi-run-mutex (make-mutex))
	(run-id-list (if run-ids
			 run-ids
			 (rmt:get-all-run-ids)))
	(result      '()))
    (if (null? run-id-list)
	'()




	(let loop ((hed     (car run-id-list))
		   (tal     (cdr run-id-list))
		   (threads '()))
	  (if (> (length threads) 5)
	      (loop hed tal (filter (lambda (th)(not (member (thread-state th) '(terminated dead)))) threads))
	      (let* ((newthread (make-thread
				 (lambda ()
				   (let ((res (rmt:send-receive 'get-tests-for-run-mindata hed (list hed testpatt states status not-in))))
				     (if (list? res)
					 (begin
					   (mutex-lock! multi-run-mutex)
					   (set! result (append result res))
					   (mutex-unlock! multi-run-mutex))
					 (debug:print 0 "ERROR: get-tests-for-run-mindata failed for run-id " hed ", testpatt " testpatt ", states " states ", status " status ", not-in " not-in))))
				 (conc "multi-run-thread for run-id " hed)))
		     (newthreads (cons newthread threads)))
		(thread-start! newthread)
		(thread-sleep! 0.5) ;; give that thread some time to start
		(if (null? tal)
		    newthreads
		    (loop (car tal)(cdr tal) newthreads))))))
    result))

;; ;; IDEA: Threadify these - they spend a lot of time waiting ...
;; ;;
;; (define (rmt:get-tests-for-runs-mindata run-ids testpatt states status not-in)
;;   (let ((run-id-list (if run-ids
;; 			 run-ids

(define (rmt:update-run-event_time run-id)
  (rmt:send-receive 'update-run-event_time #f (list run-id)))

(define (rmt:get-runs-by-patt  keys runnamepatt targpatt offset limit)
  (rmt:send-receive 'get-runs-by-patt #f (list keys runnamepatt targpatt offset limit)))

(define (rmt:find-and-mark-incomplete run-id ovr-deadtime)
  (if (rmt:send-receive 'have-incompletes? run-id (list run-id ovr-deadtime))
      (rmt:send-receive 'mark-incomplete run-id (list run-id ovr-deadtime))))

;;======================================================================
;; M U L T I R U N   Q U E R I E S
;;======================================================================

;; Need to move this to multi-run section and make associated changes
(define (rmt:find-and-mark-incomplete-all-runs #!key (ovr-deadtime #f))

  (rmt:send-receive 'testmeta-update-field #f (list test-name fld val)))

(define (rmt:test-data-rollup run-id test-id status)
  (rmt:send-receive 'test-data-rollup run-id (list run-id test-id status)))

(define (rmt:csv->test-data run-id test-id csvdata)
  (rmt:send-receive 'csv->test-data run-id (list run-id test-id csvdata)))

;;======================================================================
;;  T A S K S
;;======================================================================

(define (rmt:tasks-find-task-queue-records target run-name test-patt state-patt action-patt)
  (rmt:send-receive 'find-task-queue-records #f (list target run-name test-patt state-patt action-patt)))

(define (rmt:tasks-add action owner target runname testpatt params)
  (rmt:send-receive 'tasks-add #f (list action owner target runname testpatt params)))

(define (rmt:tasks-set-state-given-param-key param-key new-state)
  (rmt:send-receive 'tasks-set-state-given-param-key #f (list  param-key new-state)))

;;======================================================================
;; A R C H I V E S
;;======================================================================

(define (rmt:archive-get-allocations  testname itempath dneeded)
  (rmt:send-receive 'archive-get-allocations #f (list testname itempath dneeded)))

(define (rmt:archive-register-block-name bdisk-id archive-path)
  (rmt:send-receive 'archive-register-block-name #f (list bdisk-id archive-path)))

(define (rmt:archive-allocate-testsuite/area-to-block block-id testsuite-name areakey)
  (rmt:send-receive 'archive-allocate-test-to-block #f (list  block-id testsuite-name areakey)))

(define (rmt:archive-register-disk bdisk-name bdisk-path df)
  (rmt:send-receive 'archive-register-disk #f (list bdisk-name bdisk-path df)))

(define (rmt:test-set-archive-block-id run-id test-id archive-block-id)
  (rmt:send-receive 'test-set-archive-block-id run-id (list run-id test-id archive-block-id)))

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

(require-extension (srfi 18) extras tcp s11n rpc)
(import (prefix rpc rpc:))

(use sqlite3 srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest)
(import (prefix sqlite3 sqlite3:))

(declare (unit rpc-transport))

(declare (uses common))
(declare (uses db))
(declare (uses tests))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.

(include "common_records.scm")
(include "db_records.scm")

;; procstr is the name of the procedure to be called as a string
(define (rpc-transport:autoremote procstr params)
  (handle-exceptions
   exn
   (begin
     (debug:print 1 "Remote failed for " proc " " params)
     (apply (eval (string->symbol procstr)) params))
   ;; (if *runremote*
   ;;    (apply (eval (string->symbol (conc "remote:" procstr))) params)
   (apply (eval (string->symbol procstr)) params)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (rpc-transport:launch run-id)
  (set! *run-id*   run-id)
  (if (args:get-arg "-daemonize")
      (daemon:ize))
  (if (server:check-if-running run-id)
      (begin
	(debug:print 0 "INFO: Server for run-id " run-id " already running")
	(exit 0)))
  (let loop ((server-id (open-run-close tasks:server-lock-slot tasks:open-db run-id))
	     (remtries  4))
    (if (not server-id)
	(if (> remtries 0)
	    (begin
	      (thread-sleep! 2)
	      (loop (open-run-close tasks:server-lock-slot tasks:open-db run-id)
		    (- remtries 1)))
	    (begin
	      ;; since we didn't get the server lock we are going to clean up and bail out
	      (debug:print-info 2 "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
	      (open-run-close tasks:server-delete-records-for-this-pid tasks:open-db " rpc-transport:launch")))
	(begin
	  (rpc-transport:run (if (args:get-arg "-server")(args:get-arg "-server") "-") run-id server-id)
	  (exit)))))

(define (rpc-transport:run hostn run-id server-id)
  (debug:print 2 "Attempting to start the rpc server ...")
   ;; (trace rpc:publish-procedure!)

  (rpc:publish-procedure! 'server:login server:login)
  (rpc:publish-procedure! 'testing (lambda () "Just testing"))

  (let* ((db              #f)
	 (hostname        (get-host-name))
	 (ipaddrstr       (let ((ipstr (if (string=? "-" hostn)
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f)))
			    (if ipstr ipstr hostn))) ;; hostname))) 
	 (start-port      (open-run-close tasks:server-get-next-port tasks:open-db))
	 (link-tree-path  (configf:lookup *configdat* "setup" "linktree"))
	 (rpc:listener    (rpc-transport:find-free-port-and-open (rpc:default-server-port)))
	 (th1             (make-thread
			   (lambda ()
			     ((rpc:make-server rpc:listener) #t))
			   "rpc:server"))
			   ;; (cute (rpc:make-server rpc:listener) "rpc:server")
			   ;; 'rpc:server))
	 (hostname        (if (string=? "-" hostn)
			      (get-host-name) 
			      hostn))
	 (ipaddrstr       (if (string=? "-" hostn)
			      (server:get-best-guess-address hostname) ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
			      #f))
	 (portnum         (rpc:default-server-port))
	 (host:port       (conc (if ipaddrstr ipaddrstr hostname) ":" portnum))
	 (tdb             (tasks:open-db)))
    (thread-start! th1)
    (set! db *inmemdb*)
    (open-run-close tasks:server-set-interface-port 
		    tasks:open-db 
		    server-id 
		    ipaddrstr portnum)
    (debug:print 0 "Server started on " host:port)
    
    ;; (trace rpc:publish-procedure!)
    ;; (rpc:publish-procedure! 'server:login server:login)
    ;; (rpc:publish-procedure! 'testing (lambda () "Just testing"))

    ;;======================================================================
    ;;	  ;; end of publish-procedure section
    ;;======================================================================
    ;;
    (on-exit (lambda ()
	       (open-run-close tasks:server-set-state! tasks:open-db server-id "stopped")))

    (set! *rpc:listener* rpc:listener)
    (tasks:server-set-state! tdb server-id "running")
    (set! *inmemdb*  (db:setup run-id))
    ;; if none running or if > 20 seconds since 
    ;; server last used then start shutdown
    (let loop ((count 0))
      (thread-sleep! 5) ;; no need to do this very often
      (let ((numrunning -1)) ;; (db:get-count-tests-running db)))
	(if (or (> numrunning 0)
		(> (+ *last-db-access* 60)(current-seconds)))
	    (begin
	      (debug:print-info 0 "Server continuing, tests running: " numrunning ", seconds since last db access: " (- (current-seconds) *last-db-access*))
	      (loop (+ 1 count)))
	    (begin
	      (debug:print-info 0 "Starting to shutdown the server side")
	      (open-run-close tasks:server-delete-record tasks:open-db server-id " rpc-transport:try-start-server stop")
	      (thread-sleep! 10)
	      (debug:print-info 0 "Max cached queries was " *max-cache-size*)
	      (debug:print-info 0 "Server shutdown complete. Exiting")
	      ))))))

(define (rpc-transport:find-free-port-and-open port)
  (handle-exceptions
   exn
	  (begin
     (print "Failed to bind to port " (rpc:default-server-port) ", trying next port")
     (rpc-transport:find-free-port-and-open (+ port 1)))
   (rpc:default-server-port port)
   (tcp-read-timeout 240000)
   (tcp-listen (rpc:default-server-port) 10000)))

(define (rpc-transport:ping run-id host port)
  (handle-exceptions
   exn
   (begin
     (print "SERVER_NOT_FOUND")
     (exit 1))
   (let ((login-res ((rpc:procedure 'server:login host port) *toppath*)))
     (if (and (list? login-res)
	      (car login-res))
	 (begin
	   (print "LOGIN_OK")
	   (exit 0))
	 (begin
	   (print "LOGIN_FAILED")
	   (exit 1))))))

(define (rpc-transport:client-setup run-id #!key (remtries 10))
  (if *runremote*
      (begin
	(debug:print 0 "ERROR: Attempt to connect to server but already connected")
	#f)
      (let* ((host-info (hash-table-ref/default *runremote* run-id #f))) ;; (open-run-close db:get-var #f "SERVER"))
	(if host-info
	    (let ((iface    (car host-info))
		  (port     (cadr host-info))
		  (ping-res ((rpc:procedure 'server:login host port) *toppath*)))
	      (if ping-res
		  (let ((server-dat (list iface port #f #f #f)))
		    (hash-table-set! *runremote* run-id server-dat)
		    server-dat)
		  (begin
		    (server:try-running run-id)
		    (thread-sleep! 2)
		    (rpc-transport:client-setup run-id (- remtries 1)))))
 	    (let* ((server-db-info (open-run-close tasks:get-server tasks:open-db run-id)))
 	      (debug:print-info 0 "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	      (if server-db-info
 		  (let* ((iface     (tasks:hostinfo-get-interface server-db-info))
 			 (port      (tasks:hostinfo-get-port      server-db-info))
			 (server-dat (list iface port #f #f #f))
 			 (ping-res  ((rpc:procedure 'server:login host port) *toppath*)))
 		    (if start-res
 			(begin
 			  (hash-table-set! *runremote* run-id server-dat)
			  server-dat)
			(begin
			  (server:try-running run-id)
			  (thread-sleep! 2)
			  (rpc-transport:client-setup run-id (- remtries 1)))))
		  (begin
		    (server:try-running run-id)
		    (thread-sleep! 2)
		    (rpc-transport:client-setup run-id (- remtries 1)))))))))
;; 
;; 	     (port     (if (and hostinfo (> (length hostdat) 1))(cadr hostdat) #f)))
;; 	(if (and port
;; 		 (string->number port))
;; 	    (let ((portn (string->number port)))
;; 	      (debug:print-info 2 "Setting up to connect to host " host ":" port)
;; 	      (handle-exceptions
;; 	       exn
;; 	       (begin
;; 		 (debug:print 0 "ERROR: Failed to open a connection to the server at host: " host " port: " port)
;; 		 (debug:print 0 "   EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
;; 		 ;; (open-run-close 
;; 		 ;;  (lambda (db . param) 
;; 		 ;;    (sqlite3:execute db "DELETE FROM metadat WHERE var='SERVER'"))
;; 		 ;;  #f)
;; 		 (set! *runremote* #f))
;; 	       (if (and (not (args:get-arg "-server")) ;; no point in the server using the server using the server
;; 			((rpc:procedure 'server:login host portn) *toppath*))
;; 		   (begin
;; 		     (debug:print-info 2 "Logged in and connected to " host ":" port)
;; 		     (set! *runremote* (vector host portn)))
;; 		   (begin
;; 		     (debug:print-info 2 "Failed to login or connect to " host ":" port)
;; 		     (set! *runremote* #f)))))
;; 	    (debug:print-info 2 "no server available")))))

(declare (uses db))
(declare (uses common))
(declare (uses items))
(declare (uses runconfig))
(declare (uses tests))
(declare (uses server))
(declare (uses mt))
(declare (uses archive))
;; (declare (uses filedb))

(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")
(include "run_records.scm")
(include "test_records.scm")

    (if (> (- currtime lasttime) waitval)
	(begin
	  (hash-table-set! *runs:denoise* key currtime)
	  #t)
	#f)))

(define (runs:can-run-more-tests run-id jobgroup max-concurrent-jobs)
  (thread-sleep! (cond
        	  ((> *runs:can-run-more-tests-count* 20) 2);; obviously haven't had any work to do for a while
        	  (else 0)))
  (let* ((num-running             (rmt:get-count-tests-running run-id))
	 (num-running-in-jobgroup (rmt:get-count-tests-running-in-jobgroup run-id jobgroup))
	 (job-group-limit         (let ((jobg-count (config-lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))
    (if (> (+ num-running num-running-in-jobgroup) 0)

    (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))

    (set-signal-handler! signal/int
			 (lambda (signum)
			   (signal-mask! signum)
			   (print "Received signal " signum ", cleaning up before exit. Please wait...")
			   (let ((tdbdat (tasks:open-db)))
			     (rmt:tasks-set-state-given-param-key task-key "killed"))
			   (print "Killed by signal " signum ". Exiting")
			   (exit)))

    ;; register this run in monitor.db
    (rmt:tasks-add "run-tests" user target runname test-patts task-key) ;; params)
    (rmt:tasks-set-state-given-param-key task-key "running")
    (runs:set-megatest-env-vars run-id inkeys: keys inrunname: runname) ;; these may be needed by the launching process
    (if (file-exists? runconfigf)
	(setup-env-defaults runconfigf run-id *already-seen-runconfig-info* keyvals target)
	(debug:print 0 "WARNING: You do not have a run config file: " runconfigf))

    ;; Now generate all the tests lists
    (set! all-tests-registry (tests:get-all))

    (if (not (null? required-tests))
	(debug:print-info 1 "Adding " required-tests " to the run queue"))
    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print-info 4 "test-records=" (hash-table->alist test-records))
    (let ((reglen (configf:lookup *configdat* "setup" "runqueue")))
      (if (> (length (hash-table-keys test-records)) 0)
	  (let* ((keep-going        #t)
		 (run-queue-retries 5)
		 (th1        (make-thread (lambda ()
					    (handle-exceptions
					     exn
					     (begin
					       (print-call-chain (current-error-port))
					       (debug:print 0 "ERROR: failure in runs:run-tests-queue thread, error: " ((condition-property-accessor 'exn 'message) exn))
					       (if (> run-queue-retries 0)
						   (begin
						     (set! run-queue-retries (- run-queue-retries 1))
						     (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests (any->number reglen) all-tests-registry))))
					     (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests (any->number reglen) all-tests-registry)))
					  "runs:run-tests-queue"))
		 (th2        (make-thread (lambda ()				    
					    ;; (rmt:find-and-mark-incomplete-all-runs))))) CAN'T INTERRUPT IT ...
					    (let ((run-ids (rmt:get-all-run-ids)))
					      (for-each (lambda (run-id)
							  (if keep-going
							      (handle-exceptions
							       exn
							       (debug:print 0 "error in calling find-and-mark-incomplete for run-id " run-id)
							       (rmt:find-and-mark-incomplete run-id #f)))) ;; ovr-deadtime)))
							run-ids)))
					  "runs: mark-incompletes")))
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th1)
	    (set! keep-going #f)
	    (thread-join! th2)
	    ;; if run-count > 0 call, set -preclean and -rerun STUCK/DEAD
	    (if (> run-count 0)
		(begin
		  (if (not (hash-table-ref/default flags "-preclean" #f))
		      (hash-table-set! flags "-preclean" #t))
		  (if (not (hash-table-ref/default flags "-rerun" #f))
		      (hash-table-set! flags "-rerun" "STUCK/DEAD,n/a,ZERO_ITEMS"))
		  (runs:run-tests target runname test-patts user flags run-count: (- run-count 1)))))
	  (debug:print-info 0 "No tests to run")))
    (debug:print-info 4 "All done by here")
    (rmt:tasks-set-state-given-param-key task-key "done")
    ;; (sqlite3:finalize! tasks-db)
    ))


;; loop logic. These are used in runs:run-tests-queue to make it a bit more readable.
;;
;; If reg not full and have items in tal then loop with (car tal)(cdr tal) reg reruns

	     (newtal      (append tal (list hed)))
	     (regfull     (>= (length reg) reglen))
	     (num-running (rmt:get-count-tests-running-for-run-id run-id)))

	;; every couple minutes verify the server is there for this run
	(if (and (common:low-noise-print 60 "try start server"  run-id)
		 (tasks:need-server run-id))
	    (tasks:start-and-wait-for-server tdbdat run-id 10)) ;; NOTE: delay and wait is done under the hood
	
	(if (> num-running 0)
	  (set! last-time-some-running (current-seconds)))

      (if (> (current-seconds)(+ last-time-some-running 240))
	  (hash-table-set! *max-tries-hash* tfullname (+ (hash-table-ref/default *max-tries-hash* tfullname 0) 1)))
	;; (debug:print 0 "max-tries-hash: " (hash-table->alist *max-tries-hash*))

;; fields are passing in through 
;; action:
;;    'remove-runs
;;    'set-state-status
;;
;; NB// should pass in keys?
;;
(define (runs:operate-on action target runnamepatt testpatt #!key (state #f)(status #f)(new-state-status #f)(mode 'remove-all)(options '()))
  (common:clear-caches) ;; clear all caches
  (let* ((db           #f)
	 (tdbdat       (tasks:open-db))
	 (keys         (rmt:get-keys))
	 (rundat       (mt:get-runs-by-patt keys runnamepatt target))
	 (header       (vector-ref rundat 0))
	 (runs         (vector-ref rundat 1))

							       (else          'event_time))))))
	 (let* ((run-id    (db:get-value-by-header run header "id"))
		(run-state (db:get-value-by-header run header "state"))
		(run-name  (db:get-value-by-header run header "runname"))
		(tests     (if (not (equal? run-state "locked"))
			       (proc-get-tests run-id)
			       '()))
		(lasttpath "/does/not/exist/I/hope")
		(worker-thread #f))
	   (debug:print-info 4 "runs:operate-on run=" run ", header=" header)
	   (if (not (null? tests))
	       (begin
		 (case action
		   ((remove-runs)
		    (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    ;; seek and kill in flight -runtests with % as testpatt here
		    (if (equal? testpatt "%")
			(tasks:kill-runner target run-name)
			(debug:print 0 "not attempting to kill any run launcher processes as testpatt is " testpatt))
		    (debug:print 1 "Removing tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((set-state-status)
		    (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    (debug:print 1 "Modifying state and staus for tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((print-run)
		    (debug:print 1 "Printing info for run " runkey ", run=" run ", tests=" tests ", header=" header)
		    action)
		   ((run-wait)
		    (debug:print 1 "Waiting for run " runkey ", run=" runnamepatt " to complete"))
		   ((archive)
		    (debug:print 1 "Archiving data for run: " runkey " " (db:get-value-by-header run header "runname"))
		    (set! worker-thread (make-thread (lambda ()
						       (case (string->symbol (args:get-arg "-archive"))
							 ((save save-remove keep-html)(archive:run-bup (args:get-arg "-archive") run-id run-name tests))
							 ((restore)(archive:bup-restore (args:get-arg "-archive") run-id run-name tests))
							 (else (debug:print 0 "ERROR: unrecognised sub command to -archive. Run \"megatest\" to see help"))))
						     "archive-bup-thread"))
		    (thread-start! worker-thread))
		   (else
		    (debug:print-info 0 "action not recognised " action)))
		 
		 ;; actions that operate on one test at a time can be handled below
		 ;;
		 (let ((sorted-tests     (sort tests (lambda (a b)(let ((dira ;; (rmt:sdb-qry 'getstr 
									 (db:test-get-rundir a)) ;; )  ;; (filedb:get-path *fdb* (db:test-get-rundir a)))
									(dirb ;; (rmt:sdb-qry 'getstr 
									 (db:test-get-rundir b))) ;; ) ;; ((filedb:get-path *fdb* (db:test-get-rundir b))))
								    (if (and (string? dira)(string? dirb))
									(> (string-length dira)(string-length dirb))
									#f)))))

					    (mt:test-set-state-status-by-id run-id (db:test-get-id test) "KILLREQ" "n/a" #f)
						  (thread-sleep! 1)))
					    ;; NOTE: This is suboptimal as the testdata will be used later and the state/status may have changed ...
					    (if (null? tal)
						(loop new-test-dat tal)
						(loop (car tal)(append tal (list new-test-dat)))))
					  (begin
					    (runs:remove-test-directory new-test-dat mode) ;; 'remove-all)
					    (if (not (null? tal))
						(loop (car tal)(cdr tal))))))))
			       ((set-state-status)
				(debug:print-info 2 "new state " (car state-status) ", new status " (cadr state-status))
				(mt:test-set-state-status-by-id run-id (db:test-get-id test) (car state-status)(cadr state-status) #f)
				(if (not (null? tal))
				    (loop (car tal)(cdr tal))))
			       ((run-wait)
				(debug:print-info 2 "still waiting, " (length tests) " tests still running")
				(thread-sleep! 10)
				(let ((new-tests (proc-get-tests run-id)))
				  (if (null? new-tests)
				      (debug:print-info 1 "Run completed according to zero tests matching provided criteria.")
				      (loop (car new-tests)(cdr new-tests)))))
			       ((archive)
				(if (not toplevel-with-children)
				    (begin
				      (debug:print-info 0 "Estimating disk space usage for " test-fulln)
				      (debug:print-info 0 "   " (common:get-disk-space-used (conc run-dir "/")))))
				(if (not (null? tal))
				    (loop (car tal)(cdr tal))))
			       )))
		       )
		     (if worker-thread (thread-join! worker-thread))))))
	   ;; remove the run if zero tests remain
	   (if (eq? action 'remove-runs)
	       (let ((remtests (mt:get-tests-for-run (db:get-value-by-header run header "id") #f '("DELETED") '("n/a") not-in: #t)))
		 (if (null? remtests) ;; no more tests remaining
		     (let* ((dparts  (string-split lasttpath "/"))
			    (runpath (conc "/" (string-intersperse 
						(take dparts (- (length dparts) 1))

		       )))))
	 ))
     runs)
    ;; (sqlite3:finalize! (db:delay-if-busy tdbdat))
    )
  #t)

(define (runs:remove-test-directory test mode) ;; remove-data-only)
  (let* ((run-dir       (db:test-get-rundir test))    ;; run dir is from the link tree
	 (real-dir      (if (file-exists? run-dir)
			    (resolve-pathname run-dir)
			    #f)))
    (case mode
      ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "CLEANING" "LOCKED" #f))
      ((remove-all)      (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "REMOVING" "LOCKED" #f))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVE_REMOVING" #f #f)))
    (debug:print-info 1 "Attempting to remove " (if real-dir (conc " dir " real-dir " and ") "") " link " run-dir)
    (if (and real-dir 
	     (> (string-length real-dir) 5)
	     (file-exists? real-dir)) ;; bad heuristic but should prevent /tmp /home etc.
	(begin ;; let* ((realpath (resolve-pathname run-dir)))
	  (debug:print-info 1 "Recursively removing " real-dir)
	  (if (file-exists? real-dir)

		 (delete-directory run-dir)))
	    (if (and run-dir
		     (not (member run-dir (list "n/a" "/tmp/badname"))))
		(debug:print 0 "WARNING: not removing " run-dir " as it either doesn't exist or is not a symlink")
		(debug:print 0 "NOTE: the run dir for this test is undefined. Test may have already been deleted."))
	    ))
    ;; Only delete the records *after* removing the directory. If things fail we have a record 
    (case mode
      ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "NOT_STARTED" "n/a" #f))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVED" #f #f))
      (else (rmt:delete-test-records (db:test-get-run_id test) (db:test-get-id test))))))

;;======================================================================
;; Routines for manipulating runs
;;======================================================================

;; Since many calls to a run require pretty much the same setup 
;; this wrapper is used to reduce the replication of code
(define (general-run-call switchname action-desc proc)
  (let ((runname (or (args:get-arg "-runname")(args:get-arg ":runname")))
	(target  (common:args-get-target)))
    (cond
     ((not target)
      (debug:print 0 "ERROR: Missing required parameter for " switchname ", you must specify the target with -target")
      (exit 3))
     ((not runname)
      (debug:print 0 "ERROR: Missing required parameter for " switchname ", you must specify the run name with -runname runname")
      (exit 3))
     (else
      (let (;; (db   #f)
	    (keys #f))
	(if (launch:setup-for-run)
	    (launch:cache-config)
	    (begin 
	      (debug:print 0 "Failed to setup, exiting")
	      (exit 1)))
	;; (if (args:get-arg "-server")
	;;     (cdb:remote-run server:start db (args:get-arg "-server")))
	(set! keys (keys:config-get-fields *configdat*))
	;; have enough to process -target or -reqtarg here
	(if (args:get-arg "-reqtarg")
	    (let* ((runconfigf (conc  *toppath* "/runconfigs.config")) ;; DO NOT EVALUATE ALL 
		   (runconfig  (read-config runconfigf #f #t environ-patt: #f)))
	      (if (hash-table-ref/default runconfig (args:get-arg "-reqtarg") #f)
		  (keys:target-set-args keys (args:get-arg "-reqtarg") args:arg-hash)
		    
		  (begin
		    (debug:print 0 "ERROR: [" (args:get-arg "-reqtarg") "] not found in " runconfigf)
		    ;; (if db (sqlite3:finalize! db))
		    (exit 1)
		    )))
	    (if (args:get-arg "-target")
		(keys:target-set-args keys (args:get-arg "-target" args:arg-hash) args:arg-hash)))
	(if (not (car *configinfo*))
	    (begin
	      (debug:print 0 "ERROR: Attempted to " action-desc " but run area config file not found")
	      (exit 1))
	    ;; Extract out stuff needed in most or many calls
	    ;; here then call proc
	    (let* ((keyvals    (keys:target->keyval keys target)))
	      (proc target runname keys keyvals)))
	;; (if db (sqlite3:finalize! db))
	(set! *didsomething* #t))))))

;;======================================================================
;; Lock/unlock runs
;;======================================================================

(define (runs:handle-locking target keys runname lock unlock user)

(declare (unit server))

(declare (uses common))
(declare (uses db))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
(declare (uses synchash))
(declare (uses http-transport))
(declare (uses rpc-transport))
(declare (uses nmsg-transport))
(declare (uses launch))
(declare (uses daemon))

(include "common_records.scm")
(include "db_records.scm")

(define (server:make-server-url hostport)
  (if (not hostport)

;;

;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id)
  (case *transport-type*
    ((http)(http-transport:launch run-id))
    ((nmsg)(nmsg-transport:launch run-id))

    ((rpc)  (rpc-transport:launch run-id))

    (else (debug:print 0 "ERROR: unknown server type " *transport-type*))))
;;       (else   (debug:print 0 "ERROR: No known transport set, transport=" transport ", using rpc")
;; 	      (rpc-transport:launch run-id)))))


;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

;; Get the transport
(define (server:get-transport)
  (if *transport-type*
      *transport-type*
      (let ((ttype (string->symbol
		    (or (args:get-arg "-transport")
			(configf:lookup *configdat* "server" "transport")
			"rpc"))))
	(set! *transport-type* ttype)
	ttype)))
	    
;; Generate a unique signature for this server
(define (server:mk-signature)
  (message-digest-string (md5-primitive) 
			 (with-output-to-string
			   (lambda ()
			     (write (list (current-directory)
					  (argv)))))))


;; When using zmq this would send the message back (two step process)
;; with spiffy or rpc this simply returns the return data to be returned
;; 
(define (server:reply return-addr query-sig success/fail result)
  (debug:print-info 11 "server:reply return-addr=" return-addr ", result=" result)
  ;; (send-message pubsock target send-more: #t)
  ;; (send-message pubsock 
  (case (server:get-transport)
    ((rpc)  (db:obj->string (vector success/fail query-sig result)))
    ((http) (db:obj->string (vector success/fail query-sig result)))
    ((zmq)
     (let ((pub-socket (vector-ref *runremote* 1)))
       (send-message pub-socket return-addr send-more: #t)
       (send-message pub-socket (db:obj->string (vector success/fail query-sig result)))))
    ((fs)   result)
    (else 
     (debug:print 0 "ERROR: unrecognised transport type: " *transport-type*)
     result)))

;; Given a run id start a server process    ### NOTE ### > file 2>&1 
;; if the run-id is zero and the target-host is set 
;; try running on that host
;;
(define  (server:run run-id)
  (let* ((curr-host   (get-host-name))

	  (setenv "TARGETHOST" target-host)))
    (setenv "TARGETHOST_LOGF" logfile)
    (system (conc "nbfake " cmdln))
    (unsetenv "TARGETHOST_LOGF")
    (if (get-environment-variable "TARGETHOST")(unsetenv "TARGETHOST"))
    ;; (system cmdln)
    (pop-directory)))

(define (server:get-client-signature)
  (if *my-client-signature* *my-client-signature*
      (let ((sig (server:mk-signature)))
	(set! *my-client-signature* sig)
	*my-client-signature*)))

;; kind start up of servers, wait 40 seconds before allowing another server for a given
;; run-id to be launched
(define (server:kind-run run-id)
  (let ((last-run-time (hash-table-ref/default *server-kind-run* run-id #f)))
    (if (or (not last-run-time)
	    (> (- (current-seconds) last-run-time) 30))

	       (trycount 0))
    (if server
	;; note: client:start will set *runremote*. this needs to be changed
	;;       also, client:start will login to the server, also need to change that.
	;;
	;; client:start returns #t if login was successful.
	;;
	(let ((res (case *transport-type*
		     ((http)(server:ping-server run-id 
						(tasks:hostinfo-get-interface server)
						(tasks:hostinfo-get-port      server)))
		     ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
						 (tasks:hostinfo-get-port      server)
						 timeout: 2)))))
	  ;; if the server didn't respond we must remove the record
	  (if res
	      #t
	      (begin
		(debug:print-info 0 "server at " server " not responding, removing record")
		(tasks:server-force-clean-running-records-for-run-id (db:delay-if-busy tdbdat) run-id 
				" server:check-if-running")

		(res "NOREPLY"))
       (if (eof-object? inl)
	   (case (string->symbol res)
	     ((NOREPLY)  #f)
	     ((LOGIN_OK) #t)
	     (else       #f))
	   (loop (read-line) inl))))))

(define (server:login toppath)
  (lambda (toppath)
    (set! *last-db-access* (current-seconds))
    (if (equal? *toppath* toppath)
	(begin
	  ;; (debug:print-info 2 "login successful")
	  #t)
	(begin
	  ;; (debug:print-info 2 "login failed")
	  #f))))

(define (server:get-timeout)
  (let ((tmo (configf:lookup  *configdat* "server" "timeout")))
    (if (and (string? tmo)
	     (string->number tmo))
	(* 60 60 (string->number tmo))
	;; (* 3 24 60 60) ;; default to three days
	(* 60 1)         ;; default to one minute
	;; (* 60 60 25)      ;; default to 25 hours
	)))

    
;; (cdb:remote-run db:get-keys #f)  
;; (cdb:remote-run db:get-num-runs #f "%")
;; (cdb:remote-run db:get-runs #f runnamepatt numruns *start-run-offset* keypatts)
;;
;; keynum => the field to use as the unique key (usually 0 but can be other field)
;;
(define (synchash:client-get proc synckey keynum synchash run-id . params)

  (let* ((data   (rmt:synchash-get run-id proc synckey keynum params))
	 (newdat (car data))
	 (removs (cadr data))
	 (myhash (hash-table-ref/default synchash synckey #f)))
    (if (not myhash)
	(begin
	  (set! myhash (make-hash-table))
	  (hash-table-set! synchash synckey myhash)))

     removs)
    ;; WHICH ONE!?
    ;; data)) ;; return the changed and deleted list
    (list newdat removs))) ;; synchash))

(define *synchashes* (make-hash-table))

(define (synchash:server-get dbstruct run-id proc synckey keynum params)
  ;; (debug:print-info 2 "synckey: " synckey ", keynum: " keynum ", params: " params)
  (let* ((dbdat     (db:get-db dbstruct run-id))
	 (db        (db:dbdat-get-db dbdat))
	 (synchash  (hash-table-ref/default *synchashes* synckey #f))
	 (newdat    (apply (case proc
			     ((db:get-runs)                   db:get-runs)
			     ((db:get-tests-for-run-mindata)  db:get-tests-for-run-mindata)
			     ((db:get-test-info-by-ids)       db:get-test-info-by-ids)
			     (else
			      (print "ERROR: sync for hash " proc " not setup! Edits needed in synchash.scm")
			      print))
			   db params))
	 (postdat  #f)
	 (make-indexed (lambda (x)

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')

(use sqlite3 srfi-1 posix regex regex-case srfi-69 dot-locking format)
(import (prefix sqlite3 sqlite3:))

(declare (unit tasks))
(declare (uses db))
(declare (uses rmt))
(declare (uses common))

(include "task_records.scm")

;;======================================================================
;; Tasks db
;;======================================================================

	     (set! *db-write-access* write-access)) ;; only unset so other db's also can use this control
	 (sqlite3:set-busy-handler! mdb handler)
	 (db:set-sync mdb) ;; (sqlite3:execute mdb (conc "PRAGMA synchronous = 0;"))
	 ;;  (if (or (and (not exists)
	 ;; 	      (file-write-access? *toppath*))
	 ;; 	 (not (file-read-access? dbpath)))
	 ;;      (begin
	 ;; 
	 ;; TASKS QUEUE MOVED TO main.db
	 ;;
	 ;; (sqlite3:execute mdb "CREATE TABLE IF NOT EXISTS tasks_queue (id INTEGER PRIMARY KEY,
         ;;                        action TEXT DEFAULT '',
         ;;                        owner TEXT,
         ;;                        state TEXT DEFAULT 'new',
         ;;                        target TEXT DEFAULT '',
         ;;                        name TEXT DEFAULT '',
         ;;                        testpatt TEXT DEFAULT '',
         ;;                        keylock TEXT,
         ;;                        params TEXT,
         ;;                        creation_time TIMESTAMP,
         ;;                        execution_time TIMESTAMP);")
	 (sqlite3:execute mdb "CREATE TABLE IF NOT EXISTS monitors (id INTEGER PRIMARY KEY,
                                pid INTEGER,
                                start_time TIMESTAMP,
                                last_update TIMESTAMP,
                                hostname TEXT,
                                username TEXT,
                               CONSTRAINT monitors_constraint UNIQUE (pid,hostname));")

                                  hostname TEXT,
                                  cmdline TEXT,
                                  login_time TIMESTAMP,
                                  logout_time TIMESTAMP DEFAULT -1,
                                CONSTRAINT clients_constraint UNIQUE (pid,hostname));")
	       
	       ;))

	 (set! *task-db* (cons mdb dbpath))
	 *task-db*))))

;;======================================================================
;; Server and client management
;;======================================================================

(define (tasks:hostinfo-get-hostname    vec)    (vector-ref  vec 6))

(define (tasks:server-lock-slot mdb run-id)
  (tasks:server-clean-out-old-records-for-run-id mdb run-id " tasks:server-lock-slot")
  (if (< (tasks:num-in-available-state mdb run-id) 4)
      (begin 
	(tasks:server-set-available mdb run-id)
	(thread-sleep! (/ (random 1500) 1000)) ;; (thread-sleep! 2) ;; Try removing this. It may not be needed.
	(tasks:server-am-i-the-server? mdb run-id))
      #f))
	
;; register that this server may come online (first to register goes though with the process)
(define (tasks:server-set-available mdb run-id)
  (sqlite3:execute 
   mdb 
   "INSERT INTO servers (pid,hostname,port,pubport,start_time,      priority,state,mt_version,heartbeat,   interface,transport,run_id)
                   VALUES(?, ?,       ?,   ?, strftime('%s','now'), ?,       ?,    ?,-1,?,        ?,        ?);"
   (current-process-id)          ;; pid
   (get-host-name)               ;; hostname
   -1                            ;; port
   -1                            ;; pubport
   (random 1000)                 ;; priority (used a tiebreaker on get-available)
   "available"                   ;; state
   (common:version-signature)    ;; mt_version
   -1                            ;; interface
   ;; (conc (server:get-transport)) ;; transport
   (conc *transport-type*)    ;; transport
   run-id
   ))

(define (tasks:num-in-available-state mdb run-id)
  (let ((res 0))
    (sqlite3:for-each-row
     (lambda (num-in-queue)

		(loop (get-rand-port)(- remtries 1))
		(get-rand-port))
	    port))))))

(define (tasks:server-am-i-the-server? mdb run-id)
  (let* ((all    (tasks:server-get-servers-vying-for-run-id mdb run-id))
	 (first  (if (null? all)
		     #f;; (begin (debug:print 0 "ERROR: no servers listed, should be at least one by now.") 
		       ;;      (sqlite3:finalize! mdb)
		       ;;      (exit 1))
		     (car (db:get-rows all)))))
    (if first
	(let* ((header   (db:get-header all))
	       (id       (db:get-value-by-header first header "id"))
	       (hostname (db:get-value-by-header first header "hostname"))
	       (pid      (db:get-value-by-header first header "pid"))
	       (priority (db:get-value-by-header first header "priority")))
	  ;; (debug:print 0 "INFO: am-i-the-server got record " first)
	  ;; for now a basic check. add tiebreaking by priority later
	  (if (and (equal? hostname (get-host-name))
		   (equal? pid      (current-process-id)))
	      id
	      #f))
	#f)))
	     
;; Use: (db:get-value-by-header (car (db:get-rows dat)) (db:get-header dat) "fieldname")
;;  to extract info from the structure returned
;;
(define (tasks:server-get-servers-vying-for-run-id mdb run-id)
   (let* ((header (list "id" "hostname" "pid" "interface" "port" "pubport" "state" "run_id" "priority" "start_time"))

     (lambda (id)
       (set! res id))
     mdb ;; NEEDS dbprep ADDED
     "SELECT id FROM servers WHERE run_id=? AND state = 'running';" run-id)
    res))

(define (tasks:need-server run-id)
  (configf:lookup *configdat* "server" "required"))

;; 	(maxqry (cdr (rmt:get-max-query-average run-id)))
;; 	(threshold   (string->number (or (configf:lookup *configdat* "server" "server-query-threshold") "10"))))
;;     (cond
;;      (forced 
;;       (if (common:low-noise-print 60 run-id "server required is set")
;; 	  (debug:print-info 0 "Server required is set, starting server for run-id " run-id "."))
;;       #t)
;;      ((> maxqry threshold)
;;       (if (common:low-noise-print 60 run-id "Max query time execeeded")
;; 	  (debug:print-info 0 "Max avg query time of " maxqry "ms exceeds limit of " threshold "ms, server needed for run-id " run-id "."))
;;       #t)
;;      (else
;;       #f))))

;; try to start a server and wait for it to be available
;;
(define (tasks:start-and-wait-for-server tdbdat run-id delay-max-tries)
  ;; ensure a server is running for this run
  (let loop ((server-dat (tasks:get-server (db:delay-if-busy tdbdat) run-id))
	     (delay-time 0))
      (if (and (not server-dat)
	       (< delay-time delay-max-tries))
	  (begin
	    (if (common:low-noise-print 60 "tasks:start-and-wait-for-server" run-id)
		(debug:print 0 "Try starting server for run-id " run-id))
	    (thread-sleep! (/ (random 2000) 1000))
	    (server:kind-run run-id)
	    (thread-sleep! (min delay-time 1))
	    (loop (tasks:get-server (db:delay-if-busy tdbdat) run-id)(+ delay-time 1))))))

(define (tasks:get-all-servers mdb)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id)
       ;;                       0  1     2         3      4     5          6        7     8          9          10        11     12
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id) res)))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport,run_id 
        FROM servers WHERE state NOT LIKE 'defunct%' ORDER BY start_time DESC;")
    res))

(define (tasks:get-server-records mdb run-id)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id)
       ;;                       0  1     2         3      4     5          6        7     8          9          10        11     12
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport run-id) res)))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport,run_id 
        FROM servers WHERE run_id=? AND state NOT LIKE 'defunct%' ORDER BY start_time DESC;"
     run-id)
    (reverse res)))

;; no elegance here ...
;;
(define (tasks:kill-server hostname pid)
  (debug:print-info 0 "Attempting to kill server process " pid " on host " hostname)
  (setenv "TARGETHOST" hostname)
  (setenv "TARGETHOST_LOGF" "server-kills.log")

	  (debug:print-info 0 "Killing server " server-id " for run-id " run-id " on host " hostname " with pid " pid)
	  (tasks:kill-server hostname pid)
	  (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id tag) )
	(debug:print-info 0 "No server found for run-id " run-id ", nothing to kill"))
    ;; (sqlite3:finalize! tdb)
    ))
    

;;======================================================================


;; M O N I T O R S

;;======================================================================
















































































































(define (tasks:remove-monitor-record mdb)

  (sqlite3:execute mdb "DELETE FROM monitors WHERE pid=? AND hostname=?;"
















		   (current-process-id)




		   (get-host-name)))














(define (tasks:get-monitors mdb)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (a . rem)
       (set! res (cons (apply vector a rem) res)))
     mdb
     "SELECT id,pid,strftime('%m/%d/%Y %H:%M',datetime(start_time,'unixepoch'),'localtime'),strftime('%m/%d/%Y %H:%M:%S',datetime(last_update,'unixepoch'),'localtime'),hostname,username FROM monitors ORDER BY last_update ASC;")
    (reverse res)
    ))


















(define (tasks:monitors->text-table monitors)
  (let ((fmtstr "~4a~8a~20a~20a~10a~10a"))
    (conc (format #f fmtstr "id" "pid" "start time" "last update" "hostname" "user") "\n"
	  (string-intersperse 
	   (map (lambda (monitor)
		  (format #f fmtstr
			  (tasks:monitor-get-id          monitor)

     (lambda (id pid host last-update delta)
       (print "Going to delete stale record for monitor with pid " pid " on host " host " last updated " delta " seconds ago")
       (set! deadlist (cons id deadlist)))
     mdb 
     "SELECT id,pid,hostname,last_update,strftime('%s','now')-last_update AS delta FROM monitors WHERE delta > 700;")
    (sqlite3:execute mdb (conc "DELETE FROM monitors WHERE id IN ('" (string-intersperse (map conc deadlist) "','") "');")))
  )
(define (tasks:register-monitor db mdb)
  (let* ((pid (current-process-id))
	 (hostname (get-host-name))
	 (userinfo (user-information (current-user-id)))
	 (username (car userinfo)))
    (print "Register monitor, pid: " pid ", hostname: " hostname ", username: " username)
    (sqlite3:execute mdb "INSERT INTO monitors (pid,start_time,last_update,hostname,username) VALUES (?,strftime('%s','now'),strftime('%s','now'),?,?);"
		     pid hostname username)))

(define (tasks:get-num-alive-monitors mdb)
  (let ((res 0))
    (sqlite3:for-each-row 
     (lambda (count)
       (set! res count))
     mdb
     "SELECT count(id) FROM monitors WHERE last_update < (strftime('%s','now') - 300) AND username=?;"
     (car (user-information (current-user-id))))
    res))

;; 
(define (tasks:start-monitor db mdb)
  (if (> (tasks:get-num-alive-monitors mdb) 2) ;; have two running, no need for more
      (debug:print-info 1 "Not starting monitor, already have more than two running")
      (let* ((megatestdb     (conc *toppath* "/megatest.db"))
	     (monitordbf     (conc (configf:lookup *configdat* "setup" "linktree") "/.db/monitor.db"))
	     (last-db-update 0)) ;; (file-modification-time megatestdb)))
	(task:register-monitor mdb)
	(let loop ((count      0)
		   (next-touch 0)) ;; next-touch is the time where we need to update last_update
	  ;; if the db has been modified we'd best look at the task queue
	  (let ((modtime (file-modification-time megatestdbpath )))
	    (if (> modtime last-db-update)
		(tasks:process-queue db mdb last-db-update megatestdb next-touch))
	    ;; WARNING: Possible race conditon here!!
	    ;; should this update be immediately after the task-get-action call above?
	    (if (> (current-seconds) next-touch)
		(begin
		  (tasks:monitors-update mdb)
		  (loop (+ count 1)(+ (current-seconds) 240)))
		(loop (+ count 1) next-touch)))))))
      
;;======================================================================
;; T A S K S   Q U E U E
;;
;;   NOTE:: These operate on task_queue which is in main.db
;;
;;======================================================================

;; NOTE: It might be good to add one more layer of checking to ensure
;;       that no task gets run in parallel.



;; register a task
(define (tasks:add dbstruct action owner target runname testpatt params)
  (db:with-db 
   dbstruct #f #t
   (lambda (db)
     (sqlite3:execute db "INSERT INTO tasks_queue (action,owner,state,target,name,testpatt,params,creation_time,execution_time)
                             VALUES (?,?,'new',?,?,?,?,strftime('%s','now'),0);" 
		      action
		      owner
		      target
		      runname
		      testpatt
		      (if params params "")))))

(define (keys:key-vals-hash->target keys key-params)
  (let ((tmp (hash-table-ref/default key-params (vector-ref (car keys) 0) "")))
    (if (> (length keys) 1)
	(for-each (lambda (key)
		    (set! tmp (conc tmp "/" (hash-table-ref/default key-params (vector-ref key 0) ""))))
		  (cdr keys)))
    tmp))
								
;; for use from the gui, not ported
;;
;; (define (tasks:add-from-params mdb action keys key-params var-params)
;;   (let ((target    (keys:key-vals-hash->target keys key-params))
;; 	(owner     (car (user-information (current-user-id))))
;; 	(runname   (hash-table-ref/default var-params "runname" #f))
;; 	(testpatts (hash-table-ref/default var-params "testpatts" "%"))
;; 	(params    (hash-table-ref/default var-params "params"    "")))
;;     (tasks:add mdb action owner target runname testpatts params)))

;; return one task from those who are 'new' OR 'waiting' AND more than 10sec old
;;
(define (tasks:snag-a-task dbstruct)
  (let ((res    #f)
	(keytxt (conc (current-process-id) "-" (get-host-name) "-" (car (user-information (current-user-id))))))
    (db:with-db
     dbstruct #f #t
     (lambda (db)
       ;; first randomly set a new to pid-hostname-hostname
       (sqlite3:execute
	db 
	"UPDATE tasks_queue SET keylock=? WHERE id IN
           (SELECT id FROM tasks_queue 
              WHERE state='new' OR 
                    (state='waiting' AND (strftime('%s','now')-execution_time) > 10) OR
                    state='reset'
              ORDER BY RANDOM() LIMIT 1);" keytxt)

       (sqlite3:for-each-row
	(lambda (id . rem)
	  (set! res (apply vector id rem)))
	db
	"SELECT id,action,owner,state,target,name,test,item,params,creation_time,execution_time FROM tasks_queue WHERE keylock=? ORDER BY execution_time ASC LIMIT 1;" keytxt)
       (if res ;; yep, have work to be done
	   (begin
	     (sqlite3:execute db "UPDATE tasks_queue SET state='inprogress',execution_time=strftime('%s','now') WHERE id=?;"
			      (tasks:task-get-id res))
	     res)
	   #f)))))

(define (tasks:reset-stuck-tasks dbstruct)
  (let ((res '()))
    (db:with-db
     dbstruct #f #t
     (lambda (db)
       (sqlite3:for-each-row
	(lambda (id delta)
	  (set! res (cons id res)))
	db
	"SELECT id,strftime('%s','now')-execution_time AS delta FROM tasks_queue WHERE state='inprogress' AND delta>700 ORDER BY delta DESC LIMIT 2;")
       (sqlite3:execute 
	db 
	(conc "UPDATE tasks_queue SET state='reset' WHERE id IN ('" (string-intersperse (map conc res) "','") "');")
	)))))

;; return all tasks in the tasks_queue table
;;
(define (tasks:get-tasks dbstruct types states)
  (let ((res '()))
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (sqlite3:for-each-row
	(lambda (id . rem)
	  (set! res (cons (apply vector id rem) res)))
	db
	(conc "SELECT id,action,owner,state,target,name,test,item,params,creation_time,execution_time 
                  FROM tasks_queue "
	      ;; WHERE  
	      ;;   state IN " statesstr " AND 
	      ;;   action IN " actionsstr 
	      " ORDER BY creation_time DESC;"))
       res))))

;; remove tasks given by a string of numbers comma separated
(define (tasks:remove-queue-entries dbstruct task-ids)
  (db:with-db
   dbstruct #f #t
   (lambda (db)
     (sqlite3:execute db (conc "DELETE FROM tasks_queue WHERE id IN (" task-ids ");")))))

(define (tasks:process-queue dbstruct)
  (let* ((task   (tasks:snag-a-task dbstruct))
	 (action (if task (tasks:task-get-action task) #f)))
    (if action (print "tasks:process-queue task: " task))
    (if action
	(case (string->symbol action)
	  ((run)       (tasks:start-run     dbstruct task))
	  ((remove)    (tasks:remove-runs   dbstruct task))
	  ((lock)      (tasks:lock-runs     dbstruct task))
	  ;; ((monitor)   (tasks:start-monitor db task))
	  ((rollup)    (tasks:rollup-runs   dbstruct task))
	  ((updatemeta)(tasks:update-meta   dbstruct task))
	  ((kill)      (tasks:kill-monitors dbstruct task))))))

(define (tasks:tasks->text tasks)
  (let ((fmtstr "~10a~10a~10a~12a~20a~12a~12a~10a"))
    (conc (format #f fmtstr "id" "action" "owner" "state" "target" "runname" "testpatts" "params") "\n"
	  (string-intersperse 
	   (map (lambda (task)
		  (format #f fmtstr
			  (tasks:task-get-id     task)
			  (tasks:task-get-action task)
			  (tasks:task-get-owner  task)
			  (tasks:task-get-state  task)
			  (tasks:task-get-target task)
			  (tasks:task-get-name   task)
			  (tasks:task-get-test   task)
			  ;; (tasks:task-get-item   task)
			  (tasks:task-get-params task)))
		tasks) "\n"))))
   
(define (tasks:set-state dbstruct task-id state)
  (db:with-db 
   dbstruct #f #t
   (lambda (db)
     (sqlite3:execute db "UPDATE tasks_queue SET state=? WHERE id=?;" 
		      state 
		      task-id))))

;;======================================================================
;; Access using task key (stored in params; (hash-table->alist flags) hostname pid
;;======================================================================

(define (tasks:param-key->id dbstruct task-params)
  (db:with-db
   dbstruct #f #f
   (lambda (db)
     (handle-exceptions
      exn
      #f
      (sqlite3:first-result db "SELECT id FROM tasks_queue WHERE params LIKE ?;"
			    task-params)))))

(define (tasks:set-state-given-param-key dbstruct param-key new-state)
  (db:with-db
   dbstruct #f #t
   (lambda (db)
     (sqlite3:execute db "UPDATE tasks_queue SET state=? WHERE params LIKE ?;" new-state param-key))))

(define (tasks:get-records-given-param-key dbstruct param-key state-patt action-patt test-patt)
  (db:with-db
   dbstruct #f #f
   (lambda (db)
     (handle-exceptions
      exn
      '()
      (sqlite3:first-row db "SELECT id,action,owner,state,target,name,testpatt,keylock,params WHERE
                               params LIKE ? AND state LIKE ? AND action LIKE ? AND testpatt LIKE ?;"
			 param-key state-patt action-patt test-patt)))))






(define (tasks:find-task-queue-records dbstruct target run-name test-patt state-patt action-patt)
  ;; (handle-exceptions
  ;;  exn
  ;;  '()
  ;;  (sqlite3:first-row
  (let ((db (db:delay-if-busy (db:get-db dbstruct #f)))
	(res '()))
    (sqlite3:for-each-row 
     (lambda (a . b)
       (set! res (cons (cons a b) res)))
     db "SELECT id,action,owner,state,target,name,testpatt,keylock,params FROM tasks_queue 
           WHERE
              target = ? AND name = ? AND state LIKE ? AND action LIKE ? AND testpatt LIKE ?;"
     target run-name state-patt action-patt test-patt)
    res)) ;; )

;; kill any runner processes (i.e. processes handling -runtests) that match target/runname
;; 
;; do a remote call to get the task queue info but do the killing as self here.
;;
(define (tasks:kill-runner target run-name)
  (let ((records    (rmt:tasks-find-task-queue-records target run-name "%" "running" "run-tests"))
	(hostpid-rx (regexp "\\s+(\\w+)\\s+(\\d+)$"))) ;; host pid is at end of param string
    (if (null? records)
	(debug:print 0 "No run launching processes found for " target " / " run-name)
	(debug:print 0 "Found " (length records) " run(s) to kill."))
    (for-each 
     (lambda (record)
       (let* ((param-key (list-ref record 8))

		   (let ((old-targethost (getenv "TARGETHOST")))
		     (setenv "TARGETHOST" hostname)
		     (setenv "TARGETHOST_LOGF" "server-kills.log")
		     (system (conc "nbfake kill " pid))
		     (if old-targethost (setenv "TARGETHOST" old-targethost))
		     (unsetenv "TARGETHOST")
		     (unsetenv "TARGETHOST_LOGF"))))
	     (debug:print 0 "ERROR: no record or improper record for " target "/" run-name " in tasks_queue in main.db"))))
     records)))









;; (define (tasks:start-run dbstruct mdb task)
;;   (let ((flags (make-hash-table)))
;;     (hash-table-set! flags "-rerun" "NOT_STARTED")
;;     (if (not (string=? (tasks:task-get-params task) ""))
;; 	(hash-table-set! flags "-setvars" (tasks:task-get-params task)))
;;     (print "Starting run " task)
;;     ;; sillyness, just call the damn routine with the task vector and be done with it. FIXME SOMEDAY
;;     (runs:run-tests db
;; 		    (tasks:task-get-target task)
;; 		    (tasks:task-get-name   task)
;; 		    (tasks:task-get-test   task)
;; 		    (tasks:task-get-item   task)
;; 		    (tasks:task-get-owner  task)
;; 		    flags)
;;     (tasks:set-state mdb (tasks:task-get-id task) "waiting")))
;; 
;; (define (tasks:rollup-runs db mdb task)
;;   (let* ((flags (make-hash-table)) 
;; 	 (keys  (db:get-keys db))
;; 	 (keyvals (keys:target-keyval keys (tasks:task-get-target task))))
;;     ;; (hash-table-set! flags "-rerun" "NOT_STARTED")
;;     (print "Starting rollup " task)
;;     ;; sillyness, just call the damn routine with the task vector and be done with it. FIXME SOMEDAY
;;     (runs:rollup-run db
;; 		     keys 
;; 		     keyvals
;; 		     (tasks:task-get-name  task)
;; 		     (tasks:task-get-owner  task))
;;     (tasks:set-state mdb (tasks:task-get-id task) "waiting")))

(use nanomsg srfi-18 sqlite3 numbers)

(define resp (nn-socket 'rep))

(use zmq posix numbers)

(define cname "Bob")
(define runtime 10)
(let ((args (argv)))
  (if (< (length args) 3)
      (begin
	(print "Usage: mockupclient clientname runtime")
	(exit))
      (begin
	(set! cname (cadr args))
	(set! runtime (string->number (caddr args))))))
      
;; (define start-delay (/ (random 100) 9))
;; (define runtime     (+ 1 (/ (random 200) 2)))

(print "Starting client " cname " with runtime " runtime)

(include "mockupclientlib.scm")

(set! endtime (+ (current-seconds) runtime))

;; first ping the server to ensure we have a connection
(if (server-ping cname 5)
    (print "SUCCESS: Client " cname " connected to server")
    (begin
      (print "ERROR: Client " cname " failed ping of server, exiting")
      (exit)))

(let loop ()
  (let ((x (random 15))
	(varname (list-ref (list "hello" "goodbye" "saluton" "kiaorana")(random 4))))
    (case x
      ;; ((1)(dbaccess cname 'sync "nodat"    #f))
      ((2 3 4 5)(dbaccess cname 'set varname (random 999)))
      ((6 7 8 9 10)(print cname ": Get \"" varname "\" " (dbaccess cname 'get varname #f)))
      (else
       (thread-sleep! 0.011)))
    (if (< (current-seconds) endtime)
	(loop))))

(print "Client " cname " all done!!")

(define reqs (nn-socket 'req))

(connect-socket reqs "tcp://localhost:6563")

(thread-sleep! 0.2)

(define (server-ping cname timeout)
  (let ((msg     (conc cname ":ping:" timeout))
	(maxtime (+ (current-seconds) timeout)))
    (print "pinging server from " cname " with timeout " timeout)
    (let loop ((res   #f))
      (if (< maxtime (current-seconds))
	  #f ;; failed to ping
	  (if (equal? res "Got ping")
	      #t
	      (begin
		(print "Ping received from server " res)
		(send-message push msg)
		(thread-sleep! 0.1)
		(loop (receive-message sub non-blocking: #t))))))))
  
(define (dbaccess cname cmd var val #!key (numtries 20))
  (let* ((msg (conc cname ":" cmd ":" (if val (conc var " " val) var)))
	 (res #f)
	 (mtx1 (make-mutex))
	 (do-access (lambda ()
		      (let ((tmpres #f))
			(print "Sending msg: " msg)
			(send-message push msg)
			(print "Message " msg " sent")
			(print "Client " cname " waiting for response to " msg)
			(print "Client " cname " received address " (receive-message* sub))
			(set! tmpres (receive-message* sub))
			(mutex-lock! mtx1)
			(set! res tmpres)
			(mutex-unlock! mtx1))))
	 (th1 (make-thread do-access "do access"))
	 (th2 (make-thread (lambda ()
			     (let ((result #f))
			       (mutex-lock! mtx1)
			       (set! result res)
			       (mutex-unlock! mtx1)
			       (thread-sleep! 5)
			       (if (not result)
				   (if (> numtries 0)
				       (begin
					 (print "WARNING: access timed out for " cname ", trying again. Trys remaining=" numtries)
					 (dbaccess cname cmd var val numtries: (- numtries 1)))
				       (begin
					 (print "ERROR: dbaccess timed out. Exiting")
					 (exit)))))
			     "timeout thread"))))
    (thread-start! th1)
    (thread-start! th2)
    (thread-join! th1)
    (if res (print "SUCCESS: received " res " with " numtries " remaining possible attempts"))
    res))

;; pub/sub with envelope address
;; Note that if you don't insert a sleep, the server will crash with SIGPIPE as soon
;; as a client disconnects.  Also a remaining client may receive tons of
;; messages afterward.

(use nanomsg srfi-18 sqlite3 numbers)

(define resp (nn-socket 'rep))
(define cname "server")
(define total-db-accesses 0)
(define start-time (current-seconds))

(nn-bind resp  "tcp://*:6563")

(thread-sleep! 0.2)

(define (open-db)
  (let* ((dbpath    "mockup.db")
	 (dbexists  (file-exists? dbpath))
	 (db        (open-database dbpath)) ;; (never-give-up-open-db dbpath))
	 (handler   (make-busy-timeout 10)))
    (set-busy-handler! db handler)
    (if (not dbexists)
	(for-each
	 (lambda (stmt)
	   (execute db stmt))
	 (list
	  "PRAGMA SYNCHRONOUS=0;"
	  "CREATE TABLE clients (id INTEGER PRIMARY KEY,name TEXT,num_accesses INTEGER DEFAULT 0);"
	  "CREATE TABLE vars    (var TEXT,val TEXT,CONSTRAINT vars_constraint UNIQUE (var));")))
    db))

(define cid-cache (make-hash-table))

(define (get-client-id db cname)
  (let ((cid (hash-table-ref/default cid-cache cname #f)))
    (if cid 
	cid
	(begin
	  (execute db "INSERT OR REPLACE INTO clients (name) VALUES(?);" cname)
	  (for-each-row 
	   (lambda (id)
	     (set! cid id))
	   db
	   "SELECT id FROM clients WHERE name=?;" cname)
	  (hash-table-set! cid-cache cname cid)
	  (set! total-db-accesses (+ total-db-accesses 2))
	  cid))))

(define (count-client db cname)
  (let ((cid (get-client-id db cname)))
    (execute db "UPDATE clients SET num_accesses=num_accesses+1 WHERE id=?;" cid)
    (set! total-db-accesses (+ total-db-accesses 1))
    ))

(define db (open-db))
;; (define queuelst '())
;; (define mx1 (make-mutex))

(define max-queue-len 0)

(define (process-queue queuelst)
  (let ((queuelen (length queuelst)))
    (if (> queuelen max-queue-len)
	(set! max-queue-len queuelen))
    (for-each
     (lambda (item)
       (let ((cname (vector-ref item 1))
	     (clcmd (vector-ref item 2))
	     (cdata (vector-ref item 3)))
	 (send-message pub cname send-more: #t)
	 (send-message pub (case clcmd
			     ((sync)
			      (conc queuelen))
			     ((set)
			      (set! total-db-accesses (+ total-db-accesses 1))
			      (apply execute db "INSERT OR REPLACE INTO vars (var,val) VALUES (?,?);" (string-split cdata))
			      "ok")
			     ((get)
			      (set! total-db-accesses (+ total-db-accesses 1))
			      (let ((res "noval"))
				(for-each-row
				 (lambda (val)
				   (set! res val))
				 db 
				 "SELECT val FROM vars WHERE var=?;" cdata)
				res))
			     (else (conc "unk cmd: " clcmd))))))
     queuelst)))

;; SERVER THREAD
(define th1 (make-thread 
	     (lambda ()
	       (let ((last-run 0)) ;; current-seconds when run last
		 (let loop ((queuelst '()))
		   (let* ((indat (receive-message* pull))
			  (parts (string-split indat ":"))
			  (cname (car parts))                   ;; client name
			  (clcmd (string->symbol (cadr parts))) ;; client cmd
			  (cdata (caddr parts))                 ;; client data
			  (svect (vector (current-seconds) cname clcmd cdata))) ;; record for the queue
		     ;; (print "Server received message: " indat)
		     (count-client db cname)
		     (case clcmd
		       ((ping)
			(print "Got ping from " cname)
			(send-message pub cname send-more: #t)
			(send-message pub "Got ping")
			(loop queuelst))
		       ((sync) ;; just process the queue
			(print "Got sync from " cname)
			(process-queue (cons svect queuelst))
			(loop '()))
		       ((get)
			(process-queue (cons svect queuelst))
			(loop '()))
		       (else
			(loop (cons svect queuelst))))))))
	     "server thread"))

(include "mockupclientlib.scm")

;; SYNC THREAD
;; send a sync to the pull port
(define th2 (make-thread
	     (lambda ()
	       (let ((last-action-time (current-seconds)))
		 (let loop ()
		   (thread-sleep! 5)
		   (let ((queuelen (string->number (dbaccess "server" 'sync "nada" #f)))
			 (last-action-delta #f))
		     (if (> queuelen 1)(set! last-action-time (current-seconds)))
		     (set! last-action-delta (- (current-seconds) last-action-time))
		     (print "Server: Got queuelen=" queuelen ", last-action-delta=" last-action-delta)
		     (if (< last-action-delta 60)
			 (loop)
			 (print "Server exiting, 25 seconds since last access"))))))
	     "sync thread"))

(thread-start! th1)
(thread-start! th2)
(thread-join! th2)

(let* ((run-time       (- (current-seconds) start-time))
       (queries/second (/  total-db-accesses run-time)))
  (print "Server exited! Total db accesses=" total-db-accesses " in " run-time " seconds for " queries/second " queries/second with max queue length of: " max-queue-len))

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

(define push (nn-socket 'push))
(define pull1 (nn-socket 'pull))
(define pull2 (nn-socket 'pull))

(nn-bind    push  "inproc://test")
(nn-connect pull1 "inproc://test")
(nn-connect pull2 "inproc://test")

(nn-send push "a")
(nn-send push "b")
(nn-send push "c")
(nn-send push "d")

(define ((th sock))
  (print (current-thread) ": " (nn-recv sock))
  (print (current-thread) ": " (nn-recv sock))
  (print (current-thread) " is done"))

(thread-start! (th pull1))
(thread-start! (th pull2))

(thread-sleep! 1)

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

(define req   (nn-socket 'req))

(nn-connect req  "tcp://localhost:22022")

;; (with-output-to-string (lambda ()(serialize obj)))
(define (client-send-receive soc msg)
  (nn-send soc msg)
  (nn-recv soc))

(define ((talk-to-server soc))
  (let loop ((cnt 20))
    (let ((name (list-ref '("Matt" "Tom" "Bob" "Jill" "James" "Jane")(random 6))))
      (print "Sending " name)
      (print (client-send-receive req name))
      (if (> cnt 0)(loop (- cnt 1)))))
  (print (client-send-receive req "quit"))
  (nn-close req)
  (exit))

;; (thread-start! (lambda ()
;; 		 (thread-sleep! 20)
;; 		 (print "Give up on waiting for the server")
;; 		 (nn-close req)
;; 		 (exit)))

(thread-join! (thread-start! (talk-to-server req)))

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

;; (use trace)
;; (trace nn-bind nn-socket nn-assert nn-recv nn-send thread-terminate! nn-close )

(define port  22022)
(define host  "127.0.0.1")

(define rep   (nn-socket 'rep))

(print "connecting, got: " (nn-bind    rep  (conc "tcp://" "*" ":" port)))

(define (server soc)
  (print "server starting")
  (let loop ((msg-in (nn-recv soc)))
    (print "server received: " msg-in)
    (cond
     ((equal? msg-in "quit")
      (nn-send soc "Ok, quitting"))
     ((and (>= (string-length msg-in) 4)
	   (equal? (substring msg-in 0 4) "ping"))
      (nn-send soc (conc (current-process-id)))
      (loop (nn-recv soc)))
     ;;((and (>= (string-length msg-in)
     (else
      (let ((this-task (random 15)))
	(thread-sleep! this-task)
	(nn-send soc (conc "hello " msg-in " this task took " this-task " seconds to complete"))
	(loop (nn-recv soc)))))))

(define (ping-self host port #!key (return-socket #t))
  ;; send a random number along with pid and check that we get it back
  (let* ((req     (nn-socket 'req))
	 (key     "ping")
	 (success #f)
	 (keepwaiting #t)
	 (ping    (make-thread
		   (lambda ()
		     (print "ping: sending string \"" key "\", expecting " (current-process-id))
		     (nn-send req key)
		     (let ((result  (nn-recv req)))
		       (if (equal? (conc (current-process-id)) result)
			   (begin
			     (print "ping, success: received \"" result "\"")
			     (set! success #t))
			   (begin
			     (print "ping, failed: received key \"" result "\"")
			     (set! keepwaiting #f)
			     (set! success #f)))))
		   "ping"))
	 (timeout (make-thread (lambda ()
				 (let loop ((count 0))
				   (thread-sleep! 1)
				   (print "still waiting after count seconds...")
				   (if (and keepwaiting (< count 10))
				       (loop (+ count 1))))
				 (if keepwaiting
				     (begin
				       (print "timeout waiting for ping")
				       (thread-terminate! ping))))
			       "timeout")))
    (nn-connect req (conc "tcp://" host ":" port))
    (handle-exceptions
     exn
     (begin
       (print-call-chain)
       (print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
       (print "exn=" (condition->list exn))
       (print "ping failed to connect to " host ":" port))
     (thread-start! timeout)
     (thread-start! ping)
     (thread-join! ping)
     (if success (thread-terminate! timeout)))
    (if return-socket
	(if success req #f)
	(begin
	  (nn-close req)
	  success))))

(let ((server-thread (make-thread (lambda ()(server rep)) "server")))
  (thread-start! server-thread)
  ;; (thread-sleep! 1)
  (if (ping-self host port)
      (begin
	(thread-join! server-thread)
	(nn-close rep))
      (print "ping failed")))

(exit)

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

(define req   (nn-socket 'req))
(define rep   (nn-socket 'rep))

(nn-bind    rep  "inproc://test")
(nn-connect req  "inproc://test")

(define (client-send-receive soc msg)
  (nn-send soc msg)
  (nn-recv soc))

(define ((server soc))
  (let loop ((msg-in (nn-recv soc)))
    (if (not (equal? msg-in "quit"))
	(begin
	  (nn-send soc (conc "hello " msg-in))
	  (loop (nn-recv soc))))))

(thread-start! (server rep))

(print (client-send-receive req "Matt"))
(print (client-send-receive req "Tom"))

;; (client-send-receive req "quit")

(nn-close req)
(nn-close rep)
(exit)

;;;; client.scm
(use rpc posix)

(define call (rpc:procedure 'foo "localhost"))

(do ((i 10 (sub1 i)))
    ((zero? i))
  (print "-> " (call (random 100))))

;;;; server.scm
(use rpc)

(rpc:publish-procedure!
 'foo
 (lambda (x)
   (print "foo: " x)
   #f))

(rpc:publish-procedure!
 'fini
 (lambda () (print "fini") (thread-start! (lambda () (thread-sleep! 3) (print "terminate") (exit))) #f))

((rpc:make-server (tcp-listen (rpc:default-server-port))) #t)

;; Call this one to do all the work and get a standardized list of tests
(define (tests:get-all)
  (let* ((test-search-path   (tests:get-tests-search-path *configdat*)))
    (tests:get-valid-tests (make-hash-table) test-search-path)))

(define (tests:get-tests-search-path cfgdat)
  (let ((paths (map cadr (configf:get-section cfgdat "tests-paths"))))
    (append paths (list (conc *toppath* "/tests")))))

(define (tests:get-valid-tests test-registry tests-paths)
  (if (null? tests-paths) 
      test-registry
      (let loop ((hed (car tests-paths))
		 (tal (cdr tests-paths)))
	(if (file-exists? hed)

stopserver :
	cd ..;make -j && make install
	cd fullrun;$(MEGATEST) -stop-server 0

repl :
	cd ..;make -j && make install
	cd fullrun;$(MEGATEST) -:b -repl

test0 : cleanprep
	cd simplerun ; $(MEGATEST) -server - -debug $(DEBUG)

test1 : cleanprep

test2 : fullprep

	cd fullrun;time $(MEGATEST) -debug $(DEBUG) -run-wait -runtests % -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_b -m "This is a comment specific to a run" -v $(LOGGING)

test4a : cleanprep
	cd fullrun;time $(MEGATEST) -debug $(DEBUG) -preclean -runtests all_toplevel -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_b -m "This is a comment specific to a run" -v $(LOGGING)

# NOTE: Only one instance can be a server
test5 : cleanprep
	rm -f fullrun/a*.log fullrun/logs/*
	@echo "WARNING: No longer running fullprep, test converage may be lessened"
	cd fullrun;sleep 0;$(MEGATEST) -preclean -runtests % -target $(TARGET) :runname $(RUNNAME)_aa -debug $(DEBUG) $(LOGGING) > aa.log 2> aa.log &
	cd fullrun;sleep 0;$(MEGATEST) -preclean -runtests % -target $(TARGET) :runname $(RUNNAME)_ab -debug $(DEBUG) $(LOGGING) > ab.log 2> ab.log &
	cd fullrun;sleep 5;$(MEGATEST) -preclean -runtests % -target $(TARGET) :runname $(RUNNAME)_ac -debug $(DEBUG) $(LOGGING) > ac.log 2> ac.log &
	cd fullrun;sleep 8;$(MEGATEST) -preclean -runtests % -target $(TARGET) :runname $(RUNNAME)_ad -debug $(DEBUG) $(LOGGING) > ad.log 2> ad.log &	
#	cd fullrun;sleep 0;$(MEGATEST) -preclean -runtests % -target $(TARGET) :runname $(RUNNAME)_ae -debug $(DEBUG) $(LOGGING) > ae.log 2> ae.log &	
#	cd fullrun;sleep 0;$(MEGATEST) -preclean -runtests % -target $(TARGET) :runname $(RUNNAME)_af -debug $(DEBUG) $(LOGGING) > af.log 2> af.log &

# Some simple checks for bootstrapping and run loop logic 

test9 : minsetup test9a test9b test9c test9d test9e

test9a :
	@echo Run super-simple mintest e, no waitons.
	cd mintest;$(DASHBOARD)&
	cd mintest;$(MEGATEST) -preclean -runtests e -target $(VER) -runname $(shell date +%H.%M.%S) -debug $(DEBUG)

test9b :
	@echo Run simple mintest d with one waiton c
	cd mintest;$(MEGATEST) -preclean -runtests d -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test9c :
	@echo Run mintest a with full waiton chain a -> b -> c -> d -> e
	cd mintest;$(MEGATEST) -preclean -runtests a -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test9d :
	@echo Run an itemized test with no items
	cd mintest;$(MEGATEST) -preclean -runtests g -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test9e :
	@echo Run mintest a1 with full waiton chain with d1fail: a1 -> b1 -> c1 -> d1fail -> e1
	cd mintest;$(MEGATEST) -preclean -runtests a1 -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test10 :
	@echo Run a bunch of different targets simultaneously
	(cd fullrun;$(MEGATEST) -server - ;sleep 2)&
	for targ in mint/btrfs/mintdir sunos/sshfs/loc; do \
	   (cd fullrun;$(MEGATEST) -preclean -runtests priority_10_waiton_1 -target $$targ :runname $(RUNNAME) &); done
	for sys in ubuntu suse redhat debian;do \

	mkdir -p mintest/runs mintest/links
	cd mintest;$(MEGATEST) -stop-server 0
	cd mintest;$(MEGATEST) -server - -debug $(DEBUG) > server.log 2> server.log & 
	sleep 3
	cd mintest;$(DASHBOARD) -rows 18 &

cleanprep : ../*.scm Makefile */*.config
	mkdir -p fullrun/tmp/mt_runs fullrun/tmp/mt_links /tmp/$(USER)/adisk1
	cd ..;make -j;make install
	rm -f */logging.db
	touch cleanprep

fullprep : cleanprep
	cd fullrun;$(MEGATEST) -remove-runs :runname $(RUNNAME)% -target %/%/% -testpatt %/%
	cd fullrun;$(BINPATH)/dashboard -rows 15 &

waivercommentpatt ^WW\d+ [a-z].*
incomplete-timeout 1

# yes, anything else is no
run-wait yes






# If set to "default" the old code is used. Otherwise defaults to 200 or uses
# numeric value given.
#
runqueue 20

# Default runtimelim 1d 1h 1m 10s

blah

MAX_ALLOWED_LOAD 200
# XTERM   [system xterm]
# RUNDEAD [system exit 56]

[server]

# Use http instead of direct filesystem access
transport http
# transport fs
# transport nmsg

synchronous 0

# If the server can't be started on this port it will try the next port until
# it succeeds
port 9080

# This server will keep running this number of hours after last access. 
# Three minutes is 0.05 hours
# timeout 0.025
timeout 0.061

# Server is required - slower but more resistant to Sqlite issues.
required yes

# Start server when average query takes longer than this
# server-query-threshold 55500
server-query-threshold 100
timeout 0.01

# daemonize yes
# hostname #{scheme (get-host-name)}

## disks are:
## name host:/path/to/area
## -or-

#       to your jobgroups where N is the number of parallel runs you are likely to see
#      
sqlite3 6
blockz  10
#       to your jobgroups where N is the number of parallel runs you are likely to see
#       

#======================================================================
# Machine flavors
#
#   These specify lists of hosts or scripts to use or call for various
#   flavors of task.
#
#======================================================================

[flavors]

plain hosts: xena, phoebe
strong command: NBFAKE_HOST=zeus nbfake
arm hosts: cubian

[archive]

# use machines of these flavor
useflavors plain
targsize 2G

# minimum space required on an archive disk before allowing archiving to start (MB)
minspace 10

[archive-disks]

# Archives will be organised under these paths like this:
#  <testsuite>/<creationdate>
# Within the archive the data is structured like this:
#  <target>/<runname>/<test>/
disk0 /tmp/#{getenv USER}/adisk1

[fields]
X TEXT

[setup]
max_concurrent_jobs 50
linktree #{getenv MT_RUN_AREA_HOME}/linktree
transport http

[server]
port 8090

[jobtools]
useshell yes

(test "setup for run" #t (begin (launch:setup-for-run)
 				(string? (getenv "MT_RUN_AREA_HOME"))))

;; NON Server tests go here

(test #f #f (db:dbdat-get-path *db*))
(test #f #f (db:get-run-name-from-id *db* run-id))
;; (test #f '("SYSTEM" "RELEASE") (rmt:get-keys))

;; (exit)

;; Server tests go here 
(for-each
 (lambda (run-id)
   (test #f #f (tasks:server-running-or-starting? (db:delay-if-busy (tasks:open-db)) run-id))
   (server:kind-run run-id)
   (test "did server start within 20 seconds?"
	 #t
	 (let loop ((remtries 20)
		    (running (tasks:server-running-or-starting? (db:delay-if-busy
								 (tasks:open-db))
								run-id)))
	   (if running 
	       (> running 0)
	       (if (> remtries 0)
		   (begin
		     (thread-sleep! 1)
		     (loop (- remtries 1)
			   (tasks:server-running-or-starting? (db:delay-if-busy
							       (tasks:open-db))
							      run-id)))))))
   
   (test "did server become available" #t
	 (let loop ((remtries 10)
		    (res      (tasks:get-server (db:delay-if-busy (tasks:open-db)) run-id)))
	   (if res
	       (vector? res)
	       (begin
		 (if (> remtries 0)
		     (begin
		       (thread-sleep! 1.1)
		       (loop (- remtries 1)(tasks:get-server (db:delay-if-busy (tasks:open-db)) run-id)))
		     res)))))
   )
 (list 0 1))

(define user    (current-user-name))
(define runname "mytestrun")
(define keys    (rmt:get-keys))
(define runinfo #f)
(define keyvals '(("SYSTEM" "abc")("RELEASE" "def")))
(define header  (list "SYSTEM" "RELEASE" "id" "runname" "state" "status" "owner" "event_time"))

;; Setup
;;
(test #f #f  (not (client:setup run-id)))
(test #f #f  (not (hash-table-ref/default *runremote* run-id #f)))

;; Login
;;
(test #f'(#t "successful login")
      (rmt:login-no-auto-client-setup (hash-table-ref/default *runremote* run-id #f) run-id))
(test #f '(#t "successful login")
      (rmt:login run-id))

;; Keys
;;
(test #f '("SYSTEM" "RELEASE")  (rmt:get-keys))

;; No data in db
;;
(test #f '() (rmt:get-all-run-ids))
(test #f #f  (rmt:get-run-name-from-id run-id))
(test #f 
      (vector
       header
       (vector #f #f #f #f))

      (rmt:get-run-info run-id))

;; Insert data into db
;;
(test #f 1 (rmt:register-run keyvals runname "new" "n/a" user))
;; (test #f #f (rmt:get-runs-by-patt keys runname))
(test #f #t (rmt:general-call 'register-test run-id run-id "test-one" ""))
(define test-one-id #f)
(test #f 1  (let ((test-id (rmt:get-test-id run-id "test-one" "")))
	      (set! test-one-id test-id)
	      test-id))
(define test-one-rec #f)
(test #f "test-one" (let ((test-rec (rmt:get-test-info-by-id run-id test-one-id)))
		      (set! test-one-rec test-rec)
		      (vector-ref test-rec 2)))

;; With data in db
;;
(print "Using runame=" runname)
(test #f '(1)    (rmt:get-all-run-ids))
(test #f runname (rmt:get-run-name-from-id run-id))
(test #f 
      runname
      (let ((run-info (rmt:get-run-info run-id)))
	(db:get-value-by-header (db:get-rows run-info)
				(db:get-header run-info)
				"runname")))


(for-each (lambda (run-id)
;; test killing server
;;
(tasks:kill-server-run-id run-id)

(test #f #f (tasks:server-running-or-starting? (db:delay-if-busy (tasks:open-db)) run-id))
)
(list 0 1))

;; Tests to assess reading/writing while servers are starting/stopping
(define start-time (current-seconds))
(let loop ((test-state 'start))
  (let* ((server-dats (tasks:get-server-records (db:delay-if-busy (tasks:open-db)) run-id))
	 (first-dat   (if (not (null? server-dats))
			  (car server-dats)
			  #f)))
    (map (lambda (dat)
	   (apply print (intersperse (vector->list dat) ", ")))
	 server-dats)
    (test #f test-one-rec (rmt:get-test-info-by-id run-id test-one-id))
    (thread-sleep! 1)
    (case test-state
      ((start)
       (print "Trying to start server")
       (server:kind-run run-id)
       (loop 'server-started))
      ((server-started)
       (case (vector-ref first-dat)
	 ((running)
	  (print "Server appears to be running. Now ask it to shutdown")
	  (rmt:kill-server run-id)
	  (loop 'server-shutdown))
	 ((shutting-down)
	  (loop test-state))
	 (else (print "Don't know what to do if get here"))))
      ((server-shutdown)
       (loop test-state)))))

;;======================================================================
;; END OF TESTS
;;======================================================================


;; (test #f #f (client:setup run-id))

;; (set! *transport-type* 'http)
;; 
;; (test "setup for run" #t (begin (launch:setup-for-run)
;; 				(string? (getenv "MT_RUN_AREA_HOME"))))