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 \
	   fs-transport.scm http-transport.scm \
           client.scm gutils.scm synchash.scm daemon.scm mt.scm dcommon.scm \
	   tree.scm rmt.scm api.scm tdb.scm  \
	   ezsteps.scm lock-queue.scm

GUISRCF  = dashboard-tests.scm dashboard-guimonitor.scm 

OFILES   = $(SRCFILES:%.scm=%.o)
GOFILES  = $(GUISRCF:%.scm=%.o)

ADTLSCR=mt_laststep mt_runstep mt_ezstep

	 (db-path       (conc *toppath* "/megatest.db"))
	 (db-mod-time   0) ;; (file-modification-time db-path))
	 (last-update   0) ;; (current-seconds))
	 (request-update #t)
	 (db             #f))
    (if (not testdat)
	(begin
	  (debug:print 2 "ERROR: No test data found for test " test-id ", exiting")
	  (exit 1))
	(let* ((run-id        (if testdat (db:test-get-run_id testdat) #f))
	       (keydat        (if testdat (rmt:get-key-val-pairs run-id) #f))
	       (rundat        (if testdat (rmt:get-run-info run-id) #f))
	       (runname       (if testdat (db:get-value-by-header (db:get-row rundat)
								  (db:get-header rundat)
								  "runname") #f))

(define (cdb:tests-update-cpuload-diskfree serverdat test-id cpuload diskfree)
  (cdb:client-call serverdat 'update-cpuload-diskfree #t *default-numtries* cpuload diskfree test-id))

(define (cdb:tests-update-run-duration serverdat test-id minutes)
  (cdb:client-call serverdat 'update-run-duration #t *default-numtries* minutes test-id))

(define (cdb:tests-update-uname-host serverdat test-id uname hostname)
  (cdb:client-call serverdat 'update-uname-host #t *default-numtries* uname hostname test-id))

(define (db:process-triggers test-id newstate newstatus)
  #t)

;; speed up for common cases with a little logic
;; NB// Ultimately this will be deprecated in deference to mt:test-set-state-status-by-id
;;
(define (db:test-set-state-status-by-id db test-id newstate newstatus newcomment)
  (cond
   ((and newstate newstatus newcomment)
    (sqlite3:execute db "UPDATE tests SET state=?,status=?,comment=? WHERE id=?;" newstate newstatus newcomment test-id))
   ((and newstate newstatus)
    (sqlite3:execute db "UPDATE tests SET state=?,status=? WHERE id=?;" newstate newstatus test-id))
   (else
    (if newstate   (sqlite3:execute db "UPDATE tests SET state=?   WHERE id=?;" newstate   test-id))
    (if newstatus  (sqlite3:execute db "UPDATE tests SET status=?  WHERE id=?;" newstatus  test-id))
    (if newcomment (sqlite3:execute db "UPDATE tests SET comment=? WHERE id=?;" newcomment test-id))))

  (db:process-triggers test-id newstate newstatus)
  #t) ;; retrun something to keep the remote calls happy

;; Never used
;; (define (db:test-set-state-status-by-run-id-testname db run-id test-name item-path status state)
;;   (sqlite3:execute db "UPDATE tests SET state=?,status=?,event_time=strftime('%s','now') WHERE run_id=? AND testname=? AND item_path=?;" 
;; 		   state status run-id test-name item-path))

(define (db:get-count-tests-running db)
  (let ((res 0))
    (sqlite3:for-each-row
     (lambda (count)
       (set! res count))
     db

;; MUST BE CALLED local!
(define (db:test-get-paths-matching db keynames target fnamepatt #!key (res '()))
  ;; BUG: Move the values derived from args to parameters and push to megatest.scm
  (let* ((testpatt   (if (args:get-arg "-testpatt")(args:get-arg "-testpatt") "%"))
	 (statepatt  (if (args:get-arg ":state")   (args:get-arg ":state")    "%"))
	 (statuspatt (if (args:get-arg ":status")  (args:get-arg ":status")   "%"))
	 (runname    (if (args:get-arg ":runname") (args:get-arg ":runname")  "%"))
	 (paths-from-db (cdb:remote-run db:test-get-paths-matching-keynames-target-new db keynames target res
					testpatt:   testpatt
					statepatt:  statepatt
					statuspatt: statuspatt
					runname:    runname)))
    (if fnamepatt
	(apply append 
	       (map (lambda (p)

		       (string-split target "/"))
		  " AND "))
	 (testqry (tests:match->sqlqry testpatt))
	 (qrystr (conc "SELECT t.rundir FROM tests AS t INNER JOIN runs AS r ON t.run_id=r.id WHERE "
		       keystr " AND r.runname LIKE '" runname "' AND " testqry
		       " AND t.state LIKE '" statepatt "' AND t.status LIKE '" statuspatt 
		       "' ORDER BY t.event_time ASC;")))

    (sqlite3:for-each-row 
     (lambda (p)
       (set! res (cons p res)))
     db 
     qrystr)
    res))

(define (db:test-get-paths-matching-keynames-target-new db keynames target res 
							#!key
							(testpatt   "%")
							(statepatt  "%")
							(statuspatt "%")
							(runname    "%"))
  (let* ((row-ids '())
	 (keystr (string-intersperse 
		  (map (lambda (key val)
			 (conc key " like '" val "'"))
		       keynames 
		       (string-split target "/"))
		  " AND "))
	 (testqry (tests:match->sqlqry testpatt))
	 (runsqry (sqlite3:prepare db (conc "SELECT id FROM runs WHERE " keystr " AND runname LIKE '" runname "';")))
	 (tstsqry (sqlite3:prepare db (conc "SELECT rundir FROM tests WHERE run_id=? AND " testqry " AND state LIKE '" statepatt "' AND status LIKE '" statuspatt "' ORDER BY event_time ASC;"))))
    (sqlite3:for-each-row
     (lambda (rid)
       (set! row-ids (cons rid row-ids)))
     runsqry)
    (for-each (lambda (rid)
		(sqlite3:for-each-row 
		 (lambda (p)
		   (set! res (cons p res)))
		 tstsqry rid))
	      row-ids)
    (sqlite3:finalize! tstsqry)
    (sqlite3:finalize! runsqry)
    res))

;; look through tests from matching runs for a file
(define (db:test-get-first-path-matching db keynames target fname)
  ;; [refpaths] is the section where references to other megatest databases are stored
  (let ((mt-paths (configf:get-section "refpaths"))
	(res       (db:test-get-paths-matching db keynames target fname)))
    (let loop ((pathdat (if (null? paths) #f (car mt-paths)))

(define (cdb:logout serverdat keyval signature)
  (cdb:client-call serverdat 'logout #t *default-numtries* keyval signature))

(define (cdb:num-clients serverdat)
  (cdb:client-call serverdat 'numclients #t *default-numtries*))

;; I think this would be more efficient if executed on client side FIXME???
(define (cdb:test-set-status-state serverdat test-id status state msg)
  (if (member state '("LAUNCHED" "REMOTEHOSTSTART"))
      (cdb:client-call serverdat 'set-test-start-time #t *default-numtries* test-id))
  (if msg
      (cdb:client-call serverdat 'state-status-msg #t *default-numtries* state status msg test-id)
      (cdb:client-call serverdat 'state-status #t *default-numtries* state status test-id))) ;; run-id test-name item-path minutes cpuload diskfree tmpfree) 

;; Set the test event_time to current time. Call this when setting a test to LAUNCHED or REMOTEHOSTSTART
;; (define (cdb:set-test-start-time! serverdat test-id)
;;   (cdb:client-call serverdat 'set-test-start-time #t *default-numtries* test-id))

(define (cdb:test-rollup-test_data-pass-fail serverdat test-id)
  (cdb:client-call serverdat 'test_data-pf-rollup #t *default-numtries* test-id test-id test-id test-id))

(define (cdb:pass-fail-counts serverdat test-id fail-count pass-count)
  (cdb:client-call serverdat 'pass-fail-counts #t *default-numtries* fail-count pass-count test-id))

(define (cdb:tests-register-test serverdat run-id test-name item-path)
  (cdb:client-call serverdat 'register-test #t *default-numtries* run-id test-name item-path))

;; more transactioned calls, these for roll-up-pass-fail stuff
(define (cdb:update-pass-fail-counts serverdat run-id test-name)
  (cdb:client-call serverdat 'update-fail-pass-counts #t *default-numtries* run-id test-name run-id test-name run-id test-name))

(define (cdb:top-test-set-running serverdat run-id test-name)
  (cdb:client-call serverdat 'top-test-set-running #t *default-numtries* run-id test-name))

(define (cdb:top-test-set-per-pf-counts serverdat run-id test-name)
  (cdb:client-call serverdat 'top-test-set-per-pf-counts #t *default-numtries* run-id test-name run-id test-name run-id test-name))

;;=

(define (cdb:flush-queue serverdat)
  (cdb:client-call serverdat 'flush #f *default-numtries*))

(define (cdb:kill-server serverdat pid)
  (cdb:client-call serverdat 'killserver #t *default-numtries* pid))

;;======================================================================
;; A G R E G A T E D   T R A N S A C T I O N   D B   W R I T E S 
;;======================================================================

(define db:queries 
  (list '(register-test          "INSERT OR IGNORE INTO tests (run_id,testname,event_time,item_path,state,status) VALUES (?,?,strftime('%s','now'),?,'NOT_STARTED','n/a');")
	;; Test state and status
	'(set-test-state         "UPDATE tests SET state=?   WHERE id=?;")
	'(set-test-status        "UPDATE tests SET state=?   WHERE id=?;")
	'(state-status           "UPDATE tests SET state=?,status=? WHERE id=?;")

	'(state-status-msg       "UPDATE tests SET state=?,status=?,comment=? WHERE id=?;")
	;; Test comment
	'(set-test-comment       "UPDATE tests SET comment=? WHERE id=?;")
	'(set-test-start-time    "UPDATE tests SET event_time=strftime('%s','now') WHERE id=?;")
	'(pass-fail-counts       "UPDATE tests SET fail_count=?,pass_count=? WHERE id=?;")
	;; test_data-pf-rollup is used to set a tests PASS/FAIL based on the pass/fail info from the steps
	'(test_data-pf-rollup    "UPDATE tests
                                    SET status=CASE WHEN (SELECT fail_count FROM tests WHERE id=?) > 0 
                                      THEN 'FAIL'
                                    WHEN (SELECT pass_count FROM tests WHERE id=?) > 0 AND 
                                      (SELECT status FROM tests WHERE id=?) NOT IN ('WARN','FAIL')
                                    THEN 'PASS'
                                    ELSE status
                                    END WHERE id=?;")
	'(test-set-log            "UPDATE tests SET final_logf=? WHERE id=?;")
	'(test-set-rundir-by-test-id "UPDATE tests SET rundir=? WHERE id=?")
	'(test-set-rundir         "UPDATE tests SET rundir=? WHERE run_id=? AND testname=? AND item_path=?;")
	'(delete-tests-in-state   "DELETE FROM tests WHERE state=? AND run_id=?;")
	'(tests:test-set-toplog   "UPDATE tests SET final_logf=? WHERE run_id=? AND testname=? AND item_path='';")
	'(update-cpuload-diskfree "UPDATE tests SET cpuload=?,diskfree=? WHERE id=?;")
	'(update-run-duration     "UPDATE tests SET run_duration=? WHERE id=?;")
	'(update-uname-host       "UPDATE tests SET uname=?,host=? WHERE id=?;")
	'(update-test-state       "UPDATE tests SET state=? WHERE state=? AND run_id=? AND testname=? AND NOT (item_path='' AND testname IN (SELECT DISTINCT testname FROM tests WHERE testname=? AND item_path != ''));")
	'(update-test-status      "UPDATE tests SET status=? WHERE status like ? AND run_id=? AND testname=? AND NOT (item_path='' AND testname IN (SELECT DISTINCT testname FROM tests WHERE testname=? AND item_path != ''));")
	;; stuff for roll-up-pass-fail-counts
	'(update-fail-pass-counts "UPDATE tests 
             SET fail_count=(SELECT count(id) FROM tests WHERE run_id=? AND testname=? AND item_path != '' AND status IN ('FAIL','CHECK')),
                 pass_count=(SELECT count(id) FROM tests WHERE run_id=? AND testname=? AND item_path != '' AND status IN ('PASS','WARN','WAIVED'))
             WHERE run_id=? AND testname=? AND item_path='';")
	'(top-test-set-running  "UPDATE tests SET state='RUNNING' WHERE run_id=? AND testname=? AND item_path='';")
	'(top-test-set-per-pf-counts "UPDATE tests
                       SET state=CASE 
                                   WHEN (SELECT count(id) FROM tests 
                                                WHERE run_id=? AND testname=?
                                                     AND item_path != '' 
                                                     AND state in ('RUNNING','NOT_STARTED')) > 0 THEN 'RUNNING'
                                   ELSE 'COMPLETED' END,
                            status=CASE 
                                  WHEN fail_count > 0 THEN 'FAIL' 
                                  WHEN pass_count > 0 AND fail_count=0 THEN 'PASS' 
                                  WHEN (SELECT count(id) FROM tests
                                         WHERE run_id=? AND testname=?
                                              AND item_path != ''
                                              AND status = 'SKIP') > 0 THEN 'SKIP'
                                  ELSE 'UNKNOWN' END
                       WHERE run_id=? AND testname=? AND item_path='';")
    ))

;; do not run these as part of the transaction
(define db:special-queries   '(rollup-tests-pass-fail
			       ;; db:roll-up-pass-fail-counts  ;; WHY NOT!?
                               login
                               immediate
			       flush
			       sync
			       set-verbosity
			       killserver
			       ))

     (lambda (id itempath state status run_duration logf comment)
       (set! res (cons (vector id itempath state status run_duration logf comment) res)))
     db
     "SELECT id,item_path,state,status,run_duration,final_logf,comment FROM tests WHERE run_id=? AND testname=? AND item_path != '';"
     run-id test-name)
    res))








































;;======================================================================
;; Tests meta data
;;======================================================================

;; read the record given a testname
(define (db:testmeta-get-record db testname)
  (let ((res #f))

How To Do Things
================

Tricks
------

This section is a compendium of a various useful tricks for debugging,
configuring and generally getting the most out of Megatest.

Debugging Tricks
----------------

Examining The Environment
~~~~~~~~~~~~~~~~~~~~~~~~~

During Config File Processing
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Organising Your Tests and Tasks
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
/nfs/ch/disks/ch_unienv_disk005/qa_mrwellan/interim/src/megatest/tests/fdktestqa/testqa
----------------------------
[tests-paths]
1 #{get misc parent}/simplerun/tests
----------------------------

-------------------
[setup]
-------------------

The runscript method is a brute force way to run scripts where the
user is responsible for setting STATE and STATUS

-------------------
runscript main.csh
-------------------


ww30.2
cellname/LVS/cellname.LAYOUT_ERRORS

Error: text open

ww31.3
cellname/LVS/cellname.LAYOUT_ERRORS

Error: text open

which can launch jobs on local and remote Linux hosts. Currently
megatest uses the network filesystem to call home to your master
sqlite3 database.

include::../plan.txt[]
include::getting_started.txt[]
include::writing_tests.txt[]
include::howto.txt[]
include::reference.txt[]

[appendix]
Example Appendix
================
One or more optional appendixes go here at section level zero.

					  new-status
					  (args:get-arg "-m") #f)
		  ;; need to update the top test record if PASS or FAIL and this is a subtest
		  (if (not (equal? item-path ""))
		      (cdb:roll-up-pass-fail-counts *runremote* run-id test-name item-path new-status))))
	    ;; for automated creation of the rollup html file this is a good place...
	    (if (not (equal? item-path ""))
		(tests:summarize-items #f run-id test-id test-name #f)) ;; don't force - just update if no
	    )))
    (pop-directory)
    rollup-status))

				 (mutex-lock! *http-mutex*)
				 (set! res (with-input-from-request ;; was dat
					   fullurl 
					   (list (cons 'key "thekey")
						 (cons 'cmd cmd)
						 (cons 'params params))
					   read-string))
				 ;; Shouldn't this be a call to the managed call-all-connections stuff above?
				(close-all-connections!)
				(mutex-unlock! *http-mutex*)
				))
	                          ;; (if cleanup
				  ;;   ;; mutex already set
				  ;;   (begin
				  ;;     (set! res dat)
				  ;;     (http-transport:dec-requests-count-and-close-all-connections))

;; Copyright 2006-2013, 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.

	  (set-run-config-vars run-id keyvals target) ;; (db:get-target db run-id))
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
	  (set-megatest-env-vars run-id)
	  (set-item-env-vars itemdat)
	  (save-environment-as-files "megatest")
	  ;; open-run-close not needed for test-set-meta-info
	  (tests:set-full-meta-info #f test-id run-id 0 work-area)
	  (tests:test-set-status! test-id "REMOTEHOSTSTART" "n/a" (args:get-arg "-m") #f)
	  ;; (cdb:set-test-start-time!  *runremote* test-id)
	  (if (args:get-arg "-xterm")
	      (set! fullrunscript "xterm")
	      (if (and fullrunscript (not (file-execute-access? fullrunscript)))
		  (system (conc "chmod ug+x " fullrunscript))))
	  ;; We are about to actually kick off the test
	  ;; so this is a good place to remove the records for 
	  ;; any previous runs
	  ;; (db:test-remove-steps db run-id testname itemdat)
	  
	  (let* ((m            (make-mutex))
		 (kill-job?    #f)
		 (exit-info    (vector #t #t #t))
		 (job-thread   #f)
		 (keep-going   #t)
		 (runit        (lambda ()
				 ;; (let-values
				 ;;  (((pid exit-status exit-code)
				 ;;    (run-n-wait fullrunscript)))
				 (tests:test-set-status! test-id "RUNNING" "n/a" #f #f)
				 ;; if there is a runscript do it first
				 (if fullrunscript
				     (let ((pid (process-run fullrunscript)))
				       (let loop ((i 0))
					 (let-values
					  (((pid-val exit-status exit-code) (process-wait pid #t)))
					  (mutex-lock! m)
					  (vector-set! exit-info 0 pid)
					  (vector-set! exit-info 1 exit-status)
					  (vector-set! exit-info 2 exit-code)
					  (set! rollup-status exit-code) 
					  (mutex-unlock! m)
					  (if (eq? pid-val 0)
					      (begin
						(thread-sleep! 1)
						(loop (+ i 1)))
					      )))))
				 ;; then, if runscript ran ok (or did not get called)
				 ;; do all the ezsteps (if any)
				 (if ezsteps
				     (let* ((testconfig (read-config (conc work-area "/testconfig") #f #t environ-patt: "pre-launch-env-vars")) ;; FIXME??? is allow-system ok here?
					    (ezstepslst (hash-table-ref/default testconfig "ezsteps" '())))

								   (mutex-lock! m)
								   (vector-set! exit-info 0 pid)
								   (vector-set! exit-info 1 exit-status)
								   (vector-set! exit-info 2 exit-code)
								   (mutex-unlock! m)
								   (if (eq? pid-val 0)
								       (begin
									 (thread-sleep! 1)
									 (processloop (+ i 1))))
								   ))
                                                     (let ((exinfo (vector-ref exit-info 2))
                                                           (logfna (if logpro-used (conc stepname ".html") "")))
						       ;; testing if procedures called in a remote call cause problems (ans: no or so I suspect)
						       (db:teststep-set-status! #f test-id stepname "end" exinfo #f logfna work-area: work-area))
						     (if logpro-used

					(calc-minutes  (lambda ()
							 (inexact->exact 
							  (round 
							   (- 
							    (current-seconds) 
							    start-seconds)))))
					(kill-tries 0))
				   (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
				   (let loop ((minutes   (calc-minutes)))
				     (begin
				       (set! kill-job? (or (test-get-kill-request test-id) ;; run-id test-name itemdat))
							   (and runtlim (let* ((run-seconds   (- (current-seconds) start-seconds))
									       (time-exceeded (> run-seconds runtlim)))
									  (if time-exceeded
									      (begin
										(debug:print-info 0 "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim)
										#t)
									      #f)))))
				       ;; open-run-close not needed for test-set-meta-info
				       (tests:set-partial-meta-info #f test-id run-id minutes work-area)
				       (if kill-job? 
					   (begin
					     (mutex-lock! m)
					     ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
					     ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
					     ;;       between tries?
					     (let* ((pid (vector-ref exit-info 0)))
					       (if (number? pid)
						   (process-signal pid signal/kill)
						   ;; (begin
						   ;;   (debug:print 0 "WARNING: Request received to kill job (attempt # " kill-tries ")")
						   ;;   (let ((processes (cmd-run->list (conc "pgrep -l -P " pid))))
						   ;;     (for-each 

						     (tests:test-set-status! test-id "KILLED"  "FAIL"
								     (args:get-arg "-m") #f)
						     (sqlite3:finalize! tdb)
						     (exit 1))))
					     (set! kill-tries (+ 1 kill-tries))
					     (mutex-unlock! m)))
				       ;; (sqlite3:finalize! db)
				       (if keep-going
					   (begin
				       (thread-sleep! (+ 10 (random 10))) ;; add some jitter to the call home time to spread out the db accesses
					     (if keep-going
						 (loop (calc-minutes)))))))))) ;; NOTE: Checking twice for keep-going is intentional
		 (th1          (make-thread monitorjob "monitor job"))
		 (th2          (make-thread runit "run job")))
	    (set! job-thread th2)
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th2)
	    (set! keep-going #f)
	    (thread-sleep! 1)
	    (thread-terminate! th1) ;; Not sure if this is a good idea
	    (thread-sleep! 0.1) ;; give thread th1 a chance to be done TODO: Verify this is needed. At 0.1 I was getting fail to stop, increased to total of 1.1 sec.
	    (mutex-lock! m)
	    (let* ((item-path (item-list->path itemdat))
		   (testinfo  (cdb:get-test-info-by-id *runremote* test-id))) ;; )) ;; run-id test-name item-path)))
	      ;; Am I completed?
	      (if (equal? (db:test-get-state testinfo) "RUNNING") ;; (not (equal? (db:test-get-state testinfo) "COMPLETED"))
		  (let ((new-state  (if kill-job? "KILLED" "COMPLETED") ;; (if (eq? (vector-ref exit-info 2) 0) ;; exited with "good" status
				                                        ;; "COMPLETED"
							                ;; (db:test-get-state testinfo)))   ;; else preseve the state as set within the test
				    )
			(new-status (cond
				     ((not (vector-ref exit-info 1)) "FAIL") ;; job failed to run
				     ((eq? rollup-status 0)
				      ;; if the current status is AUTO then defer to the calculated value (i.e. leave this AUTO)
				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO" "PASS"))
				     ((eq? rollup-status 1) "FAIL")
				     ((eq? rollup-status 2)
				      ;; if the current status is AUTO the defer to the calculated value but qualify (i.e. make this AUTO-WARN)
				      (if (equal? (db:test-get-status testinfo) "AUTO") "AUTO-WARN" "WARN"))
				     (else "FAIL")))) ;; (db:test-get-status testinfo)))
		    (debug:print-info 1 "Test exited in state=" (db:test-get-state testinfo) ", setting state/status based on exit code of " (vector-ref exit-info 1) " and rollup-status of " rollup-status)
		    (tests:test-set-status! test-id 
					    new-state
					    new-status
					    (args:get-arg "-m") #f)
		    ;; need to update the top test record if PASS or FAIL and this is a subtest
		    ;; NO NEED TO CALL roll-up-pass-fail-counts HERE, THIS IS DONE IN roll-up-pass-fail-counts called by tests:test-set-status!
		    ;; (if (not (equal? item-path ""))
		    ;;     (begin
		    ;;       (thread-sleep! 0.1) ;; give other processes an opportunity to access the db as rollup is lower priority
		    ;;       (cdb:roll-up-pass-fail-counts *runremote* run-id test-name item-path new-status)))

		    ))
	      ;; for automated creation of the rollup html file this is a good place...
	      (if (not (equal? item-path ""))
		  (tests:summarize-items #f run-id test-id test-name #f))) ;; don't force - just update if no

	    (mutex-unlock! m)


	    (debug:print 2 "Output from running " fullrunscript ", pid " (vector-ref exit-info 0) " in work area " 
			 work-area ":\n====\n exit code " (vector-ref exit-info 2) "\n" "====\n")


	    (if (not (vector-ref exit-info 1))
		(exit 4)))))))

;; set up the very basics needed for doing anything here.
(define (setup-for-run)
  ;; 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 

;; Copyright 2006-2013, 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.

;;======================================================================
;; launch a task - this runs on the originating host, tests themselves
;;
;;======================================================================

(use sqlite3 srfi-18)
(import (prefix sqlite3 sqlite3:))

(declare (unit lock-queue))
(declare (uses common))

;;======================================================================
;; attempt to prevent overlapping updates of rollup files by queueing
;; update requests in an sqlite db
;;======================================================================

(define (lock-queue:open-db fname)
  (let* ((actualfname (conc fname ".lockdb"))
	 (dbexists (file-exists? actualfname))
	 (db       (sqlite3:open-database actualfname))
	 (handler  (make-busy-timeout 3600)))
    (if dbexists
	db
	(begin
	  (sqlite3:execute 
	   db
	   "CREATE TABLE IF NOT EXISTS queue (
  	      id         INTEGER PRIMARY KEY,
              test_id    INTEGER,
              start_time INTEGER,
              state      TEXT,
              CONSTRAINT queue_constraint UNIQUE (test_id));")
	  (sqlite3:execute
	   db
	   "CREATE TABLE IF NOT EXISTS runlocks (
              id         INTEGER PRIMARY KEY,
              test_id    INTEGER,
              run_lock   TEXT,
              CONSTRAINT runlock_constraint UNIQUE (run_lock));")))
    (sqlite3:set-busy-handler! db handler)
    db))

(define (lock-queue:set-state db test-id newstate)
  (sqlite3:execute db "UPDATE queue SET state=? WHERE test_id=?;"
		   newstate
		   test-id))

(define (lock-queue:any-younger? db mystart test-id)
  (let ((res #f))
    (sqlite3:for-each-row
     (lambda (tid)
       ;; Actually this should not be needed as mystart cannot be simultaneously less than and test-id same as 
       (if (not (equal? tid test-id)) 
	   (set! res tid)))
     db
     "SELECT test_id FROM queue WHERE start_time > ?;" mystart)
    res))

(define (lock-queue:get-lock db test-id)
  (let ((res       #f)
	(lckqry    (sqlite3:prepare db "SELECT test_id,run_lock FROM runlocks WHERE run_lock='locked';"))
	(mklckqry  (sqlite3:prepare db "INSERT INTO runlocks (test_id,run_lock) VALUES (?,'locked');")))
    (let ((result 
	   (handle-exceptions
	    exn
	    #f
	    (sqlite3:with-transaction
	     db
	     (lambda ()
	       (sqlite3:for-each-row (lambda (tid lockstate)
				       (set! res (list tid lockstate)))
				     lckqry)
	       (if res
		   (if (equal? (car res) test-id)
		       #t ;; already have the lock
		       #f)
		   (begin
		     (sqlite3:execute mklckqry test-id)
		     ;; if no error handled then return #t for got the lock
		     #t)))))))
      (sqlite3:finalize! lckqry)
      (sqlite3:finalize! mklckqry)
      result)))

(define (lock-queue:release-lock fname test-id)
  (let ((db (lock-queue:open-db fname)))
    (sqlite3:execute db "DELETE FROM runlocks WHERE test_id=?;" test-id)
    (sqlite3:finalize! db)))

(define (lock-queue:steal-lock db test-id)
  (sqlite3:execute db "DELETE FROM runlocks WHERE run_lock='locked';")
  (lock-queue:get-lock db test-it))

;; returns #f if ok to skip the task
;; returns #t if ok to proceed with task
;; otherwise waits
;;
(define (lock-queue:wait-turn fname test-id)
  (let ((db      (lock-queue:open-db fname))
	(mystart (current-seconds)))
    (sqlite3:execute
     db
     "INSERT OR REPLACE INTO queue (test_id,start_time,state) VALUES (?,?,'waiting');"
     test-id mystart)
    (thread-sleep! 1) ;; give other tests a chance to register
    (let ((result 
	   (let loop ((younger-waiting (lock-queue:any-younger? db mystart test-id)))
	     (if younger-waiting
		 (begin
		   ;; no need for us to wait. mark in the lock queue db as skipping
		   (lock-queue:set-state db test-id "skipping")
		   #f) ;; let the calling process know that nothing needs to be done
		 (if (lock-queue:get-lock db test-id)
		     #t
		     (if (> (- (current-seconds) mystart) 36000) ;; waited too long, steal the lock
			 (lock-queue:steal-lock db test-id)
			 (begin
			   (thread-sleep! 1)
			   (loop (lock-queue:any-younger? db mystart test-id)))))))))
      (sqlite3:finalize! db)
      result)))
	  
            
;; (use trace)
;; (trace lock-queue:get-lock lock-queue:release-lock lock-queue:wait-turn lock-queue:any-younger? lock-queue:set-state)

	      (let ((logfname (args:get-arg "-setlog")))
		(cdb:test-set-log! *runremote* test-id logfname)))
	  (if (args:get-arg "-set-toplog")
	      ;; DO NOT run remote
	      (tests:test-set-toplog! db run-id test-name (args:get-arg "-set-toplog")))
	  (if (args:get-arg "-summarize-items")
	      ;; DO NOT run remote
	      (tests:summarize-items db run-id test-id test-name #t)) ;; do force here
	  (if (args:get-arg "-runstep")
	      (if (null? remargs)
		  (begin
		    (debug:print 0 "ERROR: nothing specified to run!")
		    (if db (sqlite3:finalize! db))
		    (exit 6))
		  (let* ((stepname   (args:get-arg "-runstep"))

(define (mt:get-prereqs-not-met run-id waitons ref-item-path #!key (mode 'normal))
  (db:get-prereqs-not-met run-id waitons ref-item-path mode: mode))

(define (mt:get-run-stats)
  (cdb:remote-run db:get-run-stats #f))

;;======================================================================
;;  S T A T E   A N D   S T A T U S   F O R   T E S T S 
;;======================================================================

(define (mt:roll-up-pass-fail-counts run-id test-name item-path status)
  (if (and (not (equal? item-path ""))
	   (member status '("PASS" "WARN" "FAIL" "WAIVED" "RUNNING" "CHECK" "SKIP")))
      (begin
	(cdb:update-pass-fail-counts *runremote* run-id test-name)
	(if (equal? status "RUNNING")
	    (cdb:top-test-set-running *runremote* run-id test-name)
	    (cdb:top-test-set-per-pf-counts *runremote* run-id test-name))
	#f)
      #f))

;; speed up for common cases with a little logic
(define (mt:test-set-state-status-by-id test-id newstate newstatus newcomment)
  (cond
   ((and newstate newstatus newcomment)
    (cdb:client-call *runremote* 'state-status-msg #t *default-numtries* newstate newstatus newcomment test-id))
   ((and newstate newstatus)
    (cdb:client-call serverdat 'state-status #t *default-numtries* newstate newstatus test-id))
   (else
    (if newstate   (cdb:client-call *runremote* 'set-test-state #t *default-numtries* newstate test-id))
    (if newstatus  (cdb:client-call *runremote* 'set-test-status #t *default-numtries* newstatus test-id))
    (if newcomment (cdb:client-call *runremote* 'set-test-comment #t *default-numtries* newcomment test-id))))
   (db:process-triggers test-id newstate newstatus))

           (runs:update-test_meta test-name test-conf)))
    
    ;; itemdat => ((ripeness "overripe") (temperature "cool") (season "summer"))
    (let* ((new-test-path (string-intersperse (cons test-path (map cadr itemdat)) "/"))
	   (test-id       (cdb:remote-run db:get-test-id #f  run-id test-name item-path))
	   (testdat       (cdb:get-test-info-by-id *runremote* test-id)))
      (if (not testdat)
	  (let loop ()
	    ;; ensure that the path exists before registering the test
	    ;; NOPE: Cannot! Don't know yet which disk area will be assigned....
	    ;; (system (conc "mkdir -p " new-test-path))
	    ;;
	    ;; (open-run-close tests:register-test db run-id test-name item-path)
	    ;;
	    ;; NB// for the above line. I want the test to be registered long before this routine gets called!
	    ;;
	    (set! test-id (cdb:remote-run db:get-test-id #f run-id test-name item-path))
	    (if (not test-id)
		(begin
		  (debug:print 2 "WARN: Test not pre-created? test-name=" test-name ", item-path=" item-path ", run-id=" run-id)
		  (cdb:tests-register-test *runremote* run-id test-name item-path)
		  (set! test-id (cdb:remote-run db:get-test-id #f run-id test-name item-path))))
	    (debug:print-info 4 "test-id=" test-id ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
	    (set! testdat (cdb:get-test-info-by-id *runremote* test-id))
	    (if (not testdat)
		(begin
		  (debug:print-info 0 "WARNING: server is overloaded, trying again in one second")
		  (thread-sleep! 1)
		  (loop)))))
      (if (not testdat) ;; should NOT happen
	  (debug:print 0 "ERROR: failed to get test record for test-id " test-id))
      (set! test-id (db:test-get-id testdat))
      (if (file-exists? test-path)
	  (change-directory test-path)
	  (begin
	    (debug:print "ERROR: test run path not created before attempting to run the test. Perhaps you are running -remove-runs at the same time?")

			 (set! skip-test "Skipping due to previous tests running"))))
		  ((and skip-check
			(configf:lookup test-conf "skip" "fileexists"))
		   (if (file-exists? (configf:lookup test-conf "skip" "fileexists"))
		       (set! skip-test (conc "Skipping due to existance of file " (configf:lookup test-conf "skip" "fileexists"))))))
		 (if skip-test
		     (begin
		       (mt:test-set-state-status-by-id test-id "COMPLETED" "SKIP" skip-test)
		       (debug:print-info 1 "SKIPPING Test " full-test-name " due to " skip-test))
		     (if (not (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat flags))
			 (begin
			   (print "ERROR: Failed to launch the test. Exiting as soon as possible")
			   (set! *globalexitstatus* 1) ;; 
			   (process-signal (current-process-id) signal/kill))))))))
	((KILLED) 

					  (hash-table-set! test-retry-time test-fulln (current-seconds))))
				    (if (> (- (current-seconds)(hash-table-ref test-retry-time test-fulln)) allow-run-time)
				      ;; This test is not in a correct state for cleaning up. Let's try some graceful shutdown steps first
				      ;; Set the test to "KILLREQ" and wait five seconds then try again. Repeat up to five times then give
				      ;; up and blow it away.
				      (begin
					(debug:print 0 "WARNING: could not gracefully remove test " test-fulln ", tried to kill it to no avail. Forcing state to FAILEDKILL and continuing")
					(mt:test-set-state-status-by-id (db:test-get-id test) "FAILEDKILL" "n/a" #f)
					(thread-sleep! 1))
				      (begin
					(mt:test-set-state-status-by-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
				    (mt:test-set-state-status-by-id (db:test-get-id test) "REMOVING" "LOCKED" #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)

					    ))
				    ;; Only delete the records *after* removing the directory. If things fail we have a record 
				    (cdb:remote-run db:delete-test-records db #f (db:test-get-id test))
				    (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 (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)

;; Tests
;;======================================================================

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

(declare (unit tests))
(declare (uses lock-queue))
(declare (uses db))
(declare (uses common))
(declare (uses items))
(declare (uses runconfig))

(include "common_records.scm")
(include "key_records.scm")

			   type     )))
	    ;; This was run remote, don't think that makes sense.
	    (db:csv->test-data #f test-id
				dat))))
      
    ;; need to update the top test record if PASS or FAIL and this is a subtest
    (if (not (equal? item-path ""))
	(mt:roll-up-pass-fail-counts run-id test-name item-path status))

    (if (or (and (string? comment)
		 (string-match (regexp "\\S+") comment))
	    waived)
	(let ((cmt  (if waived waived comment)))
	  (cdb:remote-run db:test-set-comment #f test-id cmt)))
    ))


(define (tests:test-set-toplog! db run-id test-name logf) 
  (cdb:client-call *runremote* 'tests:test-set-toplog #t 2 logf run-id test-name))

(define (tests:summarize-items db run-id test-id test-name force)
  ;; if not force then only update the record if one of these is true:
  ;;   1. logf is "log/final.log
  ;;   2. logf is same as outputfilename
  (let* ((outputfilename (conc "megatest-rollup-" test-name ".html"))
	 (orig-dir       (current-directory))
	 (logf-info      (cdb:remote-run db:test-get-logfile-info #f run-id test-name))
	 (logf           (if logf-info (cadr logf-info) #f))
	 (path           (if logf-info (car  logf-info) #f)))
    ;; This query finds the path and changes the directory to it for the test
    (if (directory? path)
	(begin
	  (debug:print 4 "Found path: " path)
	  (change-directory path))
	;; (set! outputfilename (conc path "/" outputfilename)))
	(print "No such path: " path))
    (debug:print 4 "summarize-items with logf " logf ", outputfilename " outputfilename " and force " force)
    (if (or (equal? logf "logs/final.log")
	    (equal? logf outputfilename)
	    force)
	(begin
	  (if ;; (not (obtain-dot-lock outputfilename 1 5 7)) ;; retry every second for 20 seconds, call it dead after 30 seconds and steal the lock
              (not (lock-queue:wait-turn outputfilename test-id))
	      (print "Not updating " outputfilename " as another test item has signed up for the job")
	      (begin
		(print "Obtained lock for " outputfilename)
		(let ((oup    (open-output-file outputfilename))
		      (counts (make-hash-table))
		      (statecounts (make-hash-table))
		      (outtxt "")
		      (tot    0)

		      (print "</td></td></tr></table>")
		      
		      (print "<table cellspacing=\"0\" border=\"1\">" 
			     "<tr><td>Item</td><td>State</td><td>Status</td><td>Comment</td>"
			     outtxt "</table></body></html>")
		      (release-dot-lock outputfilename)))
		  (close-output-port oup)
		  (lock-queue:release-lock outputfilename test-id)
		  (change-directory orig-dir)
		  ;; NB// tests:test-set-toplog! is remote internal...
		  (tests:test-set-toplog! db run-id test-name outputfilename)
		  )))))))

;;======================================================================
;; Gather data from test/task specifications

	 (lambda (count)
	   (set! res count))
	 tdb
	 "SELECT count(id) FROM test_rundat;")
	res))
  0)

(define (tests:update-central-meta-info test-id cpuload diskfree minutes uname hostname)
  ;; This is a good candidate for threading the requests to enable
  ;; transactionized write at the server
  (cdb:tests-update-cpuload-diskfree *runremote* test-id cpuload diskfree)





    (if minutes 
	(cdb:tests-update-run-duration *runremote* test-id minutes))

  (if (and uname hostname)
      (cdb:tests-update-uname-host *runremote* test-id uname hostname)))


  

(define (tests:set-full-meta-info db test-id run-id minutes work-area)
  ;; DOES cdb:remote-run under the hood!

  (let* ((num-records 0) ;; (test:tdb-get-rundat-count tdb))
	 (cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))

	 (uname    (get-uname "-srvpio"))
	      (hostname (get-host-name)))
    (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
    (tests:update-central-meta-info test-id cpuload diskfree minutes uname hostname)))
	  
(define (tests:set-partial-meta-info db test-id run-id minutes work-area)
  ;; DOES cdb:remote-run under the hood!
  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory))))
    (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
    ;; Update central with uname and hostname = #f
    (tests:update-central-meta-info test-id cpuload diskfree minutes #f #f)))
	 
(define (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
  (let ((tdb         (db:open-test-db-by-test-id db test-id work-area: work-area)))
    (sqlite3:execute tdb "INSERT INTO test_rundat (update_time,cpuload,diskfree,run_duration) VALUES (strftime('%s','now'),?,?,?);"
		     cpuload diskfree minutes)
    (sqlite3:finalize! tdb)))
	  
;;======================================================================
;; A R C H I V I N G
;;======================================================================

	   (cd fullrun;$(MEGATEST) -runtests priority_10_waiton_1 -target $$targ :runname $(RUNNAME) &); done
	for sys in ubuntu suse redhat debian;do \
	  for fs in afs nfs zfs; do \
	     for dpath in none tmp; do \
	        (cd fullrun;$(MEGATEST) -runtests priority_10_waiton_1 -target $$sys/$$fs/$$dpath :runname $(RUNNAME) &);\
	     done;done;done

test11 :
	 cd fullrun;time (for a in 1 2 3 4 5 6 7 8 9 10  1 2 3 4 5 6 7 8 9 10  1 2 3 4 5 6 7 8 9 10  1 2 3 4 5 6 7 8 9 10 ;do   (megatest -test-paths -target %/%/% > /dev/null ) & done; wait; )

minsetup : 
	cd ..;make && make install
	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 20 &

[fields]
SYSTEM TEXT
RELEASE TEXT

[setup]
# Adjust max_concurrent_jobs to limit how much you load your machines
max_concurrent_jobs 150

# This is your link path, you can move it but it is generally better to keep it stable
linktree #{shell readlink -f #{getenv PWD}/../simplelinks}

[include testqa/configs/megatest.abc.config]

	   ($(MEGATEST) -runtests % -target a/b :runname $tn & ) ; \
	done

bigrun :
	$(MEGATEST) -runtests bigrun -target a/bigrun :runname a

bigrun2 :
	$(MEGATEST) -runtests bigrun2 -target a/bigrun2 :runname a -transport http

dashboard : 
	$(DASHBOARD) -rows 20 &

compile :
	(cd ../../..;make && make install)

[setup]
testcopycmd cp --remove-destination -rlv TEST_SRC_PATH/. TEST_TARG_PATH/. >> TEST_TARG_PATH/mt_launch.log 2>> TEST_TARG_PATH/mt_launch.log
runqueue 20
transport http
launchwait no

[include ../fdk.config]

[server]
port 9080

home #{shell readlink -f $MT_RUN_AREA_HOME}
parent #{shell readlink -f $MT_RUN_AREA_HOME/..}

[tests-paths]
1 #{get misc parent}/simplerun/tests

[setup]
# Set launchwait to no to use the more agressive code that does not wait for the launch to complete before proceeding

launchwait no

# Use http instead of direct filesystem access
transport http


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