Megatest

Check-in [8085b19fd6]
Login
Overview
Comment:Removed gratuitious call to get-count-tests-running-for-run-id and some indenting fixes.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | v1.65
Files: files | file ages | folders
SHA1: 8085b19fd663e362e7b3a7eda8ac8ff941df3828
User & Date: matt on 2020-10-25 22:54:48
Other Links: branch diff | manifest | tags
Context
2020-10-26
22:12
Added auto propagation of env. vars in steps check-in: 21d4a4c30b user: mrwellan tags: v1.65
2020-10-25
22:54
Removed gratuitious call to get-count-tests-running-for-run-id and some indenting fixes. check-in: 8085b19fd6 user: matt tags: v1.65
2020-10-24
23:23
Meld'd more of testdat check-in: 426aad0a60 user: matt tags: v1.65
Changes

Modified runs.scm from [d0c781d218] to [5009c715c7].

633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
    ;; run the run prehook if there are no tests yet run for this run:
    ;;
    (runs:run-pre-hook run-id)
    ;; mark all test launced flag as false in the meta table 
    (rmt:set-var (conc "lunch-complete-" run-id) "no")
    (debug:print-info 1 *default-log-port* "Setting end-of-run to no")
    (let* ((config-reruns      (let ((x (configf:lookup *configdat* "setup" "reruns")))
			       (if x (string->number x) #f)))
	  (config-rerun-cnt (if config-reruns
			config-reruns
			1)))
    (if (eq? config-rerun-cnt run-count)
      (rmt:set-var (conc "end-of-run-" run-id) "no")))

    (rmt:set-run-state-status run-id "new" "n/a")
    ;; now add non-directly referenced dependencies (i.e. waiton)
    ;;======================================================================
    ;; refactoring this block into tests:get-full-data
    ;;
    ;; What happended, this code is now duplicated in tests!?
    ;;







|
|
|
|
|
|
|







633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
    ;; run the run prehook if there are no tests yet run for this run:
    ;;
    (runs:run-pre-hook run-id)
    ;; mark all test launced flag as false in the meta table 
    (rmt:set-var (conc "lunch-complete-" run-id) "no")
    (debug:print-info 1 *default-log-port* "Setting end-of-run to no")
    (let* ((config-reruns      (let ((x (configf:lookup *configdat* "setup" "reruns")))
				 (if x (string->number x) #f)))
	   (config-rerun-cnt (if config-reruns
				 config-reruns
				 1)))
      (if (eq? config-rerun-cnt run-count)
	  (rmt:set-var (conc "end-of-run-" run-id) "no")))
    
    (rmt:set-run-state-status run-id "new" "n/a")
    ;; now add non-directly referenced dependencies (i.e. waiton)
    ;;======================================================================
    ;; refactoring this block into tests:get-full-data
    ;;
    ;; What happended, this code is now duplicated in tests!?
    ;;
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
	(debug:print-info 1 *default-log-port* "Adding \"" (string-intersperse required-tests " ") "\" to the run queue"))
    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print-info 4 *default-log-port* "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)
						  (print " message: " ((condition-property-accessor 'exn 'message) exn)))
					      (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 ()			 ;; BBQ: why are we visiting ALL runs here?	    
					    ;; (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 *default-log-port* "error in calling find-and-mark-incomplete for run-id " run-id ", exn=" exn)
								;; lets run this only if a run has been NOT seen for more than 900 seconds
								(if (> (- (current-seconds)(hash-table-ref/default *find-and-mark-incomplete-last-run* run-id 0)) 900)
								    (begin
								      (rmt:find-and-mark-incomplete run-id #f)
								      (hash-table-set! *find-and-mark-incomplete-last-run* run-id (current-seconds)))
								    )))) ;; ovr-deadtime))) ;; could be root of https://hsdes.intel.com/appstore/article/#/220546828/main -- Title: Megatest jobs show DEAD even though they are still running (1.64/27)
							run-ids)))
					  "runs: mark-incompletes")))
	    ;; (thread-start! th1)
	    (thread-start! th2)
	    ;; (thread-join! th1)
	    ;; just do the main stuff in the main thread
	    (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests
								    (any->number reglen) all-tests-registry)
	    (set! keep-going #f)
	    (thread-join! th2)
	    ;; if run-count > 0 call, set -preclean and -rerun STUCK/DEAD
	    (if (> run-count 0) ;; handle reruns
		(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"))
		  ;; recursive call to self
      (runs:run-tests target runname test-patts user flags run-count: (- run-count 1)))
                  (launch:end-of-run-check run-id)))
	  (debug:print-info 0 *default-log-port* "No tests to run")))
    (debug:print-info 4 *default-log-port* "All done by here")
    ;; TODO: try putting post hook call here
      
    ;  (debug:print-info 2 *default-log-port* " run-count " run-count)
    ;  (runs:run-post-hook run-id))
    ;  (debug:print-info 2 *default-log-port* "Not calling post hook runcount = " run-count ))   







<
<
<
<
<
<
<
<
<
















<

<
<












|
|







757
758
759
760
761
762
763









764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779

780


781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
	(debug:print-info 1 *default-log-port* "Adding \"" (string-intersperse required-tests " ") "\" to the run queue"))
    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print-info 4 *default-log-port* "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)









		 (th2        (make-thread (lambda ()			 ;; BBQ: why are we visiting ALL runs here?	    
					    ;; (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 *default-log-port* "error in calling find-and-mark-incomplete for run-id " run-id ", exn=" exn)
								;; lets run this only if a run has been NOT seen for more than 900 seconds
								(if (> (- (current-seconds)(hash-table-ref/default *find-and-mark-incomplete-last-run* run-id 0)) 900)
								    (begin
								      (rmt:find-and-mark-incomplete run-id #f)
								      (hash-table-set! *find-and-mark-incomplete-last-run* run-id (current-seconds)))
								    )))) ;; ovr-deadtime))) ;; could be root of https://hsdes.intel.com/appstore/article/#/220546828/main -- Title: Megatest jobs show DEAD even though they are still running (1.64/27)
							run-ids)))
					  "runs: mark-incompletes")))

	    (thread-start! th2)


	    (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests
								    (any->number reglen) all-tests-registry)
	    (set! keep-going #f)
	    (thread-join! th2)
	    ;; if run-count > 0 call, set -preclean and -rerun STUCK/DEAD
	    (if (> run-count 0) ;; handle reruns
		(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"))
		  ;; recursive call to self
		  (runs:run-tests target runname test-patts user flags run-count: (- run-count 1)))
		(launch:end-of-run-check run-id)))
	  (debug:print-info 0 *default-log-port* "No tests to run")))
    (debug:print-info 4 *default-log-port* "All done by here")
    ;; TODO: try putting post hook call here
      
    ;  (debug:print-info 2 *default-log-port* " run-count " run-count)
    ;  (runs:run-post-hook run-id))
    ;  (debug:print-info 2 *default-log-port* "Not calling post hook runcount = " run-count ))   
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
		   (equal? (configf:lookup *configdat* "setup" "run-wait") "yes"))
	       (> num-running 0))
	  (begin
	    ;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
	    ;; (debug:print 0 *default-log-port* "Got here eh! num-running=" num-running " (> num-running 0) " (> num-running 0))
	    (if (> (- (current-seconds)(hash-table-ref/default *find-and-mark-incomplete-last-run* run-id 0)) 900)
		;; (begin(if (> (current-seconds)(+ last-time-incomplete 900))
		(let ((actual-num-running (rmt:get-count-tests-running-for-run-id run-id)))
		  (debug:print-info 0 *default-log-port* "Marking stuck tests as INCOMPLETE while waiting for run " run-id
				    ". Running as pid " (current-process-id) " on " (get-host-name))
		  ;; (set! last-time-incomplete (current-seconds)) ;; FIXME, this might be causing slow down - use of set!
		  (rmt:find-and-mark-incomplete run-id #f)
		  (hash-table-set! *find-and-mark-incomplete-last-run* run-id (current-seconds))
		  (debug:print-info 0 *default-log-port* "run-wait specified, waiting on " actual-num-running
				    " tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at "







|







1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
		   (equal? (configf:lookup *configdat* "setup" "run-wait") "yes"))
	       (> num-running 0))
	  (begin
	    ;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
	    ;; (debug:print 0 *default-log-port* "Got here eh! num-running=" num-running " (> num-running 0) " (> num-running 0))
	    (if (> (- (current-seconds)(hash-table-ref/default *find-and-mark-incomplete-last-run* run-id 0)) 900)
		;; (begin(if (> (current-seconds)(+ last-time-incomplete 900))
		(let ((actual-num-running num-running)) ;; (rmt:get-count-tests-running-for-run-id run-id))) ;; why call it again?
		  (debug:print-info 0 *default-log-port* "Marking stuck tests as INCOMPLETE while waiting for run " run-id
				    ". Running as pid " (current-process-id) " on " (get-host-name))
		  ;; (set! last-time-incomplete (current-seconds)) ;; FIXME, this might be causing slow down - use of set!
		  (rmt:find-and-mark-incomplete run-id #f)
		  (hash-table-set! *find-and-mark-incomplete-last-run* run-id (current-seconds))
		  (debug:print-info 0 *default-log-port* "run-wait specified, waiting on " actual-num-running
				    " tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at "