Megatest

;; 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