Megatest

;;======================================================================
;; 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.
;;======================================================================

(use srfi-1 posix regex-case base64 format dot-locking csv-xml z3 sql-de-lite hostinfo md5 message-digest typed-records directory-utils stack
     matchable regex posix srfi-18 extras)
     matchable regex posix srfi-18 extras
     pkts (prefix dbi dbi:))

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

(declare (unit common))

(include "common_records.scm")

	      (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val)
	    (setenv key val))
	  (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val))))

(define home (getenv "HOME"))
(define user (getenv "USER"))

;; GLOBAL GLETCHES
;; GLOBALS

;; CONTEXTS
(defstruct cxt
  (taskdb #f)
  (cmutex (make-mutex)))
;; (define *contexts* (make-hash-table))
;; (define *context-mutex* (make-mutex))

(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))
(define *db-transaction-mutex* (make-mutex))
(define *db-cache-path*       #f)
(define *db-with-db-mutex*    (make-mutex))
(define *db-api-call-time*    (make-hash-table)) ;; hash of command => (list of times)
;; no sync db
(define *no-sync-db*          #f)

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg
(define *runremote*         #f)                ;; if set up for server communication this will hold <host port>
;; (define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
;; (define *server-id*         #f)
(define *server-info*       #f)  ;; good candidate for easily convert to non-global
(define *time-to-exit*      #f)
(define *server-run*        #t)
(define *run-id*            #f)
(define *server-kind-run*   (make-hash-table))
(define *home-host*         #f)
;; (define *total-non-write-delay* 0)
(define *heartbeat-mutex*   (make-mutex))
(define *api-process-request-count* 0)
(define *max-api-process-requests* 0)
(define *server-overloaded*  #f)

;; client
(define *rmt-mutex*         (make-mutex))     ;; remote access calls mutex 

;; RPC transport
(define *rpc:listener*      #f)

;; KEY info
(define *target*            (make-hash-table)) ;; cache the target here; target is keyval1/keyval2/.../keyvalN
(define *keys*              (make-hash-table)) ;; cache the keys here
(define *keyvals*           (make-hash-table))
(define *toptest-paths*     (make-hash-table)) ;; cache toptest path settings here
(define *test-paths*        (make-hash-table)) ;; cache test-id to test run paths here
(define *test-ids*          (make-hash-table)) ;; cache run-id, testname, and item-path => test-id
(define *test-info*         (make-hash-table)) ;; cache the test info records, update the state, status, run_duration etc. from testdat.db

(define *run-info-cache*     (make-hash-table)) ;; run info is stable, no need to reget
(define *launch-setup-mutex* (make-mutex))     ;; need to be able to call launch:setup often so mutex it and re-call the real deal only if *toppath* not set
(define *homehost-mutex*     (make-mutex))

;; Miscellaneous
(define *triggers-mutex*     (make-mutex))     ;; block overlapping processing of triggers

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        (if *toppath* (server:check-if-running *toppath*))) ;; (server:check-if-running *toppath*) #f))
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (conndat           #f)
  (transport         *transport-type*)
  (server-timeout    (server:get-timeout)) ;; default from server:get-timeout

(define (common:get-last-run-version-number)
  (string->number 
   (substring (common:get-last-run-version) 0 6)))

(define (common:set-last-run-version)
  (rmt:set-var "MEGATEST_VERSION" (common:version-signature)))

;; postive number if megatest version > db version
;; negative number if megatest version < db version
(define (common:version-db-delta)
         (- megatest-version (common:get-last-run-version-number)))

(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
	       (common:version-signature))))
               (common:version-signature))))

(define (common:api-changed?)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (common:get-last-run-version)) 0 4))))
  
;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct)
  (db:multi-db-sync 
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'dejunk
   'adj-target
   ;; 'old2new
   'new2old
   (if full
       '(dejunk)
   )
  (if (common:version-changed?)
       '()))
  (if (common:api-changed?)
      (common:set-last-run-version)))

;; Rotate logs, logic: 
;;                 if > 500k and older than 1 week:
;;                     remove previous compressed log and compress this log
;; WARNING: This proc operates assuming that it is in the directory above the
;;          logs directory you wish to log-rotate.

             (file-age (- (current-seconds)(file-modification-time fullname))))
        (if (or (and (string-match "^.*.log" file)
                     (> (file-size fullname) 200000))
                (and (string-match "^server-.*.log" file)
                     (> (- (current-seconds) (file-modification-time fullname))
                        (* 8 60 60))))
            (let ((gzfile (conc fullname ".gz")))
              (if (file-exists? gzfile)
              (if (common:file-exists? gzfile)
                  (begin
                    (debug:print-info 0 *default-log-port* "removing " gzfile)
                    (delete-file gzfile)))
              (debug:print-info 0 *default-log-port* "compressing " file)
              (system (conc "gzip " fullname)))
            (if (> file-age (* (string->number (or (configf:lookup *configdat* "setup" "log-expire-days") "30")) 24 3600))
                (handle-exceptions
                 exn
                 #f
                 (delete-file fullname)))))))
   '()
   "logs"))

;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:version-changed?)
      (if (common:api-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (file-exists? mtconf) (file-exists? dbfile) (not read-only)
             ((and (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions.")
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (file-exists? mtconf))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (file-exists? dbfile))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))
      (begin
	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
	(exit 1))))
              (exit 1)))))))
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

;;======================================================================
;; S P A R S E   A R R A Y S
;;======================================================================

(define (make-sparse-array)
  (let ((a (make-sparse-vector)))

;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (handle-exceptions
      exn
      #f ;; don't really care what went wrong right now. NOTE: I have not seen this one actually fail.
    (if (file-exists? fname)
    (if (common:file-exists? fname)
	(if (> (- (current-seconds)(file-modification-time fname)) expire-time)
	    (begin
	      (delete-file* fname)
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (file-exists? fname)
	  (if (common:file-exists? fname)
	      (with-input-from-file fname
		(lambda ()
		  (equal? key-string (read-line))))
	      #f)))))

(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))

      #f ;; I don't really care why this failed (at least for now)
    (delete-file* fname)))

;;======================================================================
;; S T A T E S   A N D   S T A T U S E S
;;======================================================================

;; BBnote: *common:std-states* - dashboard filter control and test control state buttons defined here; used in set-fields-panel and dboard:make-controls
(define *common:std-states*   
(define *common:std-states*   ;; for toggle buttons in dashboard
  '((0 "ARCHIVED")
    (1 "STUCK")
    (2 "KILLREQ")
    (3 "KILLED")
    (4 "NOT_STARTED")
    (5 "COMPLETED")
    (6 "LAUNCHED")
    (7 "REMOTEHOSTSTART")
    (8 "RUNNING")
    ))

;; BBnote: *common:std-statuses* dashboard filter control and test control status buttons defined here; used in set-fields-panel and dboard:make-controls
(define *common:std-statuses*
  '(;; (0 "DELETED")
    (1 "n/a")
    (2 "PASS")
    (3 "SKIP")
    (4 "WARN")
    (5 "WAIVED")
    (6 "CHECK")
    (7 "STUCK/DEAD")
    (8 "FAIL")
    (9 "ABORT")))

(define *common:ended-states*       ;; states which indicate the test is stopped and will not proceed
  '("COMPLETED" "ARCHIVED" "KILLED" "KILLREQ" "STUCK" "INCOMPLETE"))

(define *common:badly-ended-states* ;; these roll up as CHECK, i.e. results need to be checked
  '("KILLED" "KILLREQ" "STUCK" "INCOMPLETE" "DEAD"))

;; BBnote: *common:running-states* used from db:set-state-status-and-roll-up-items
(define *common:running-states*     ;; test is either running or can be run
  '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED"))
  '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED" "STARTED"))

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

(define *common:not-started-ok-statuses* ;; if not one of these statuses when in not_started state treat as dead
  '("n/a" "na" "PASS" "FAIL" "WARN" "CHECK" "WAIVED" "DEAD" "SKIP"))

      (getenv "MT_TESTSUITE_NAME")
      (if (string? *toppath* )
          (pathname-file *toppath*)
          #f))) ;; (pathname-file (current-directory)))))

(define common:get-area-name common:get-testsuite-name)

(define (common:get-db-tmp-area)
(define (common:get-db-tmp-area . junk)
  (if *db-cache-path*
      *db-cache-path*
      (if *toppath*
	  (let ((dbpath (create-directory (conc "/tmp/" (current-user-name)
						"/megatest_localdb/"
						(common:get-testsuite-name) "/"
						(string-translate *toppath* "/" ".")) #t)))
	    (set! *db-cache-path* dbpath)
	    dbpath)
      (if *toppath* ;; common:get-create-writeable-dir
	  (handle-exceptions
	      exn
	      (begin
		(debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
		(exit 1))
	    (let ((dbpath (common:get-create-writeable-dir
			   (list (conc "/tmp/" (current-user-name)
				       "/megatest_localdb/"
				       (common:get-testsuite-name) "/"
				       (string-translate *toppath* "/" ".")))))) ;;  #t))))
	      (set! *db-cache-path* dbpath)
	      dbpath))
	  #f)))

(define (common:get-area-path-signature)
  (message-digest-string (md5-primitive) *toppath*))

(define (common:get-signature str)
  (message-digest-string (md5-primitive) str))

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

(define (common:run-sync?)
    (and (common:on-homehost?)
	 (args:get-arg "-server")))

;;   (let ((ohh (common:on-homehost?))
;; 	(srv (args:get-arg "-server")))
;;     (and ohh srv)))
    ;; (debug:print-info 0 *default-log-port* "common:run-sync? ohh=" ohh ", srv=" srv)

;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
(define (common:sync-to-megatest.db dbstruct) 
  (let ((start-time         (current-seconds))
	(res                (db:multi-db-sync dbstruct 'new2old)))
    (let ((sync-time (- (current-seconds) start-time)))
      (debug:print-info 3 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
      (if (common:low-noise-print 30 "sync new to old")
	  (debug:print-info 0 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))))
    res))




(define *wdnum* 0)
(define *wdnum*mutex (make-mutex))
;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)

        (debug:print-info 13 *default-log-port* "duration-since-last-sync="duration-since-last-sync)
        (if (and (not *time-to-exit*)
                 (< duration-since-last-sync sync-cool-off-duration))
            (thread-sleep! (- sync-cool-off-duration duration-since-last-sync)))
        (if (not *time-to-exit*)
            (let ((golden-mtdb-mtime (file-modification-time golden-mtpath))
                  (tmp-mtdb-mtime    (file-modification-time tmp-mtpath)))
              (if (> golden-mtdb-mtime tmp-mtdb-mtime)
                  (let ((res (db:multi-db-sync dbstruct 'old2new)))
                    (debug:print-info 13 *default-log-port* "rosync called, " res " records transferred.")))
	      (if (> golden-mtdb-mtime tmp-mtdb-mtime)
		  (if (< golden-mtdb-mtime (- (current-seconds) 3)) ;; file has NOT been touched in past three seconds, this way multiple servers won't fight to sync back
		      (let ((res (db:multi-db-sync dbstruct 'old2new)))
			(debug:print-info 13 *default-log-port* "rosync called, " res " records transferred."))))
              (loop (current-seconds)))
            #t)))
    (debug:print-info 0 *default-log-port* "Exiting readonly-watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" mtpath="golden-mtpath)))


        
(define (common:writable-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (let ((legacy-sync (common:run-sync?))
	(debug-mode  (debug:debug-mode 1))
	(last-time   (current-seconds))
        (this-wd-num     (begin (mutex-lock! *wdnum*mutex) (let ((x *wdnum*)) (set! *wdnum* (add1 *wdnum*)) (mutex-unlock! *wdnum*mutex) x))))
    (debug:print-info 2 *default-log-port* "Periodic sync thread started.")
    (debug:print-info 3 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id)" this-wd-num="this-wd-num)
    (if (and legacy-sync (not *time-to-exit*))
	(let* (;;(dbstruct (db:setup))
	       (mtdb     (dbr:dbstruct-mtdb dbstruct))
	       (mtpath   (db:dbdat-get-path mtdb)))
	  (debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
	  (let loop ()
	    ;; sync for filesystem local db writes
	    ;;
	    (mutex-lock! *db-multi-sync-mutex*)
	    (let* ((need-sync        (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
		   (sync-in-progress *db-sync-in-progress*)
		   (should-sync      (and (not *time-to-exit*)
                                          (> (- (current-seconds) *db-last-sync*) 5))) ;; sync every five seconds minimum
		   (start-time       (current-seconds))
		   (mt-mod-time      (file-modification-time mtpath))
		   (recently-synced  (< (- start-time mt-mod-time) 4))
		   (will-sync        (and (or need-sync should-sync)
					  (not sync-in-progress)
					  (not recently-synced))))
              (debug:print-info 13 *default-log-port* "WD writable-watchdog top of loop.  need-sync="need-sync" sync-in-progress="sync-in-progress" should-sync="should-sync" start-time="start-time" mt-mod-time="mt-mod-time" recently-synced="recently-synced" will-sync="will-sync)
	      ;; (if recently-synced (debug:print-info 0 *default-log-port* "Skipping sync due to recently-synced flag=" recently-synced))
	      ;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
	      (if will-sync (set! *db-sync-in-progress* #t))
	      (mutex-unlock! *db-multi-sync-mutex*)
	      (if will-sync
		  (let ((res (common:sync-to-megatest.db dbstruct))) ;; did we sync any data? If so need to set the db touched flag to keep the server alive
		    (if (> res 0) ;; some records were transferred, keep the db alive
			(begin
			  (mutex-lock! *heartbeat-mutex*)
			  (set! *db-last-access* (current-seconds))
			  (mutex-unlock! *heartbeat-mutex*)
			  (debug:print-info 0 *default-log-port* "sync called, " res " records transferred."))
			(debug:print-info 2 *default-log-port* "sync called but zero records transferred"))))
	      (if will-sync
		  (begin
		    (mutex-lock! *db-multi-sync-mutex*)
		    (set! *db-sync-in-progress* #f)
		    (set! *db-last-sync* start-time)
		    (mutex-unlock! *db-multi-sync-mutex*)))
	      (if (and debug-mode
		       (> (- start-time last-time) 60))
		  (begin
		    (set! last-time start-time)
		    (debug:print-info 4 *default-log-port* "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))
                  ;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
                                                            
		  (if (and (not *time-to-exit*)
			   (< count 4)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (if (not *time-to-exit*) (loop))))
	    (if (common:low-noise-print 30)
		(debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" this-wd-num="this-wd-num)))))))

;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define (common:watchdog)
  (debug:print-info 13 *default-log-port* "common:watchdog entered.")
  (if (launch:setup)
  (if (common:on-homehost?)
      (let ((dbstruct (db:setup #t)))
	(debug:print-info 13 *default-log-port* "after db:setup with dbstruct="dbstruct)
	(cond
	 ((dbr:dbstruct-read-only dbstruct)
	  (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	  (common:readonly-watchdog dbstruct))
	 (else
	  (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
	  (common:writable-watchdog dbstruct)))
	(debug:print-info 13 *default-log-port* "watchdog done."))
      (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost")))
      (if (common:on-homehost?)
	  (let ((dbstruct (db:setup #t)))
	    (debug:print-info 13 *default-log-port* "after db:setup with dbstruct=" dbstruct)
	    (cond
	     ((dbr:dbstruct-read-only dbstruct)
	      (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	      (common:readonly-watchdog dbstruct))
	     (else
	      (debug:print-info 13 *default-log-port* "loading writable-watchdog.")
	      (server:writable-watchdog dbstruct)))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))


(define (std-exit-procedure)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f

				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (http-client#close-all-connections!)
                              (if (and *runremote*
                                       (remote-conndat *runremote*))
                                  (begin
                                    (http-client#close-all-connections!))) ;; for http-client
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin

(define (common:which cmds)
  (if (null? cmds)
      #f
      (let loop ((hed (car cmds))
		 (tal (cdr cmds)))
	(let ((res (with-input-from-pipe (conc "which " hed) read-line)))
	  (if (and (string? res)
		   (file-exists? res))
		   (common:file-exists? res))
	      res
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))
  
(define (common:get-install-area)
  (let ((exe-path (car (argv))))
    (if (file-exists? exe-path)
    (if (common:file-exists? exe-path)
	(handle-exceptions
	 exn
	 #f
	 (pathname-directory
	  (pathname-directory 
	   (pathname-directory exe-path))))
	#f)))

				   (thread-sleep! delay-time)
				   (common:get-homehost trynum: (- trynum 1)))
				 (begin
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: Failed to read .homehost file after trying five times. Giving up and exiting, message: "  ((condition-property-accessor 'exn 'message) exn))
				   (exit 1)))
			   (let ((hhf (conc *toppath* "/.homehost")))
			     (if (file-exists? hhf)
			     (if (common:file-exists? hhf)
				 (with-input-from-file hhf read-line)
				 (if (file-write-access? *toppath*)
				     (begin
				       (with-output-to-file hhf
					 (lambda ()
					   (print bestadrs)))
				       (begin

	(cdr hh)
	#f)))

;; do we honor the caches of the config files?
;;
(define (common:use-cache?)
  (let ((res #t)) ;; priority by order of evaluation
    (if *configdat*
    (if *configdat* ;; sillyness here. can't use setup/use-cache to know if we can use the cached files!
	(if (equal? (configf:lookup *configdat* "setup" "use-cache") "no")
	    (set! res #f)
	    (if (equal? (configf:lookup *configdat* "setup" "use-cache") "yes")
		(set! res #t))))
    (if (args:get-arg "-no-cache")(set! res #f)) ;; overrides setting in "setup"
    (if (getenv "MT_USE_CACHE")
	(if (equal? (getenv "MT_USE_CACHE") "yes")

          ((< (+ load (/ (random 250) 1000))         ;; add a random factor to keep from getting in a rut
              (+ best-load (/ (random 250) 1000))  )
           (set! best-load load)
           (set! best-host hostname)))))
     hosts)
    best-host))




(define (common:wait-for-cpuload maxload numcpus waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload numcpus))
	 (loadjmp (- first next)))
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load " first " exceeding max of " adjload (if msg msg ""))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load " first " exceeding max of " adjload " " (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1)))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1))))))

(define (common:wait-for-homehost-load maxload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f))
         (numcpus (common:get-num-cpus hh)))
    (common:wait-for-normalized-load maxload msg: msg remote-host: hh)))

(define (common:get-num-cpus remote-host)
  (let ((proc (lambda ()
		(let loop ((numcpu 0)
			   (inl    (read-line)))
		  (if (eof-object? inl)
		      numcpu
		      (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
				(+ numcpu 1)
				numcpu)
			    (read-line)))))))
    (if remote-host
	(with-input-from-pipe 
	 (conc "ssh " remote-host " cat /proc/cpuinfo")
	 proc)
	(with-input-from-file "/proc/cpuinfo" proc))))

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload #!key (msg #f)(remote-host #f))
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg remote-host: remote-host)))

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

       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key   (car keyval))
			     (val   (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (member key ignorevars)
			(print (if (or (member key ignorevars)
				       (string-search whitesp key))
				   "# setenv "
				   "setenv ")
			       key " " delim (mungeval val) delim)))
		    envvars)))
     (with-output-to-file (conc fname ".sh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key (car keyval))
			     (val (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (or (member key ignorevars)
				       (string-search whitesp key)
				       (string-search ":" key)) ;; internal only values to be skipped.
				   "# export "
				   "export ")
			       key "=" delim (mungeval val) delim)))
                    envvars)))))

;; set some env vars from an alist, return an alist with original values
;; (("VAR" "value") ...)
(define (alist->env-vars lst)
  (if (list? lst)
      (let ((res '()))
	(for-each (lambda (p)
		    (let* ((var (car  p))
			   (val (cadr p))
			   (prv (get-environment-variable var)))
		      (set! res (cons (list var prv) res))
		      (if val 
			  (setenv var (->string val))
			  (safe-setenv var (->string val))
			  (unsetenv var))))
		  lst)
	res)
      '()))

;; clear vars matching pattern, run proc, set vars back
;; if proc is a string run that string as a command with

  (string-intersperse 
   (map (lambda (x)
          (number->string x 16))
        (map string->number
             (string-split instr)))
   "/"))

(define (common:faux-lock keyname)
  (if (rmt:get-var keyname)
      #f
(define (common:faux-lock keyname #!key (wait-time 8))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:set-var keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:get-var keyname))))))
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:get-var keyname))))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:get-var keyname) (rmt:del-var keyname))
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

  
(define (common:in-running-test?)
  (and (args:get-arg "-execute") (get-environment-variable "MT_CMDINFO")))

;; first read ~/views.config if it exists, then read $MTRAH/views.config if it exists
;;
(define (common:load-views-config)
  (let* ((view-cfgdat    (make-hash-table))
	 (home-cfgfile   (conc (get-environment-variable "HOME") "/.mtviews.config"))
	 (mthome-cfgfile (conc *toppath* "/.mtviews.config")))
    (if (file-exists? mthome-cfgfile)
    (if (common:file-exists? mthome-cfgfile)
	(read-config mthome-cfgfile view-cfgdat #t))
    ;; we load the home dir file AFTER the MTRAH file so the user can clobber settings when running the dashboard in read-only areas
    (if (file-exists? home-cfgfile)
    (if (common:file-exists? home-cfgfile)
	(read-config home-cfgfile view-cfgdat #t))
    view-cfgdat))

;;======================================================================
;; Manage pkts, used in servers, tests and likely other contexts so put
;; in common
;;======================================================================

(define common:pkt-spec
  '((server . ((action    . a)
	       (pid       . d)
	       (ipaddr    . i)
	       (port      . p)))
    			  
    (test   . ((cpuuse    . c)
	       (diskuse   . d)
	       (item-path . i)
	       (runname   . r)
	       (state     . s)
	       (target    . t)
	       (status    . u)))))

(define (common:get-pkts-dirs mtconf use-lt)
  (let* ((pktsdirs-str (or (configf:lookup mtconf "setup"  "pktsdirs")
			   (and use-lt
				(conc *toppath* "/lt/.pkts"))))
	 (pktsdirs  (if pktsdirs-str
			(string-split pktsdirs-str " ")
			#f)))
    pktsdirs))

;; use-lt is use linktree "lt" link to find pkts dir
(define (common:with-queue-db mtconf proc #!key (use-lt #f)(toppath-in #f))
  (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
	 (pktsdir  (if pktsdirs (car pktsdirs) #f))
	 (toppath  (or (configf:lookup mtconf "scratchdat" "toppath")
		       toppath-in))
	 (pdbpath  (or (configf:lookup mtconf "setup"  "pdbpath") pktsdir)))
    (cond
     ((not (and  pktsdir toppath pdbpath))
      (debug:print 0 *default-log-port* "ERROR: settings are missing in your megatest.config for area management.")
      (debug:print  0 *default-log-port* "  you need to have pktsdir in the [setup] section."))
     ((not (common:file-exists? pktsdir))
      (debug:print 0 *default-log-port* "ERROR: pkts directory not found " pktsdir))
     ((not (equal? (file-owner pktsdir)(current-effective-user-id)))
      (debug:print 0 *default-log-port* "ERROR: directory " pktsdir " is not owned by " (current-effective-user-name)))
     (else
	(let* ((pdb  (open-queue-db pdbpath "pkts.db"
				    schema: '("CREATE TABLE groups (id INTEGER PRIMARY KEY,groupname TEXT, CONSTRAINT group_constraint UNIQUE (groupname));"))))
	  (proc pktsdirs pktsdir pdb)
	  (dbi:close pdb))))))

(define (common:load-pkts-to-db mtconf)
  (common:with-queue-db
   mtconf
   (lambda (pktsdirs pktsdir pdb)
     (for-each
      (lambda (pktsdir) ;; look at all
	(cond
	 ((not (common:file-exists? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory " pktsdir " does not exist."))
	 ((not (directory? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not a directory."))
	 ((not (file-read-access? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not readable."))
	 (else
	  (debug:print-info 0 *default-log-port* "Loading packets found in " pktsdir)
	  (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	    (for-each
	     (lambda (pkt)
	       (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))
		      (exists  (lookup-by-uuid pdb uuid #f)))
		 (if (not exists)
		     (let* ((pktdat (string-intersperse
				     (with-input-from-file pkt read-lines)
				     "\n"))
			    (apkt   (pkt->alist pktdat))
			    (ptype  (alist-ref 'T apkt)))
		       (add-to-queue pdb pktdat uuid (or ptype 'cmd) #f 0)
		       (debug:print 4 *default-log-port* "Added " uuid " of type " ptype " to queue"))
		     (debug:print 4 *default-log-port* "pkt: " uuid " exists, skipping...")
		     )))
	     pkts)))))
      pktsdirs))))

(define (common:get-pkt-alists pkts)
  (map (lambda (x)
	 (alist-ref 'apkt x)) ;; 'pkta pulls out the alist from the read pkt
       pkts))

;; given list of pkts (alist mode) return list of D cards as Unix epoch, sorted descending
;; also delete duplicates by target i.e. (car pkt)
;;
(define (common:get-pkt-times pkts)
  (delete-duplicates
   (sort 
    (map (lambda (x)
	   `(,(alist-ref 't x) . ,(string->number (alist-ref 'D x))))
	 pkts)
    (lambda (a b)(> (cdr a)(cdr b))))      ;; sort descending
   (lambda (a b)(equal? (car a)(car b))))) ;; remove duplicates by target