Megatest

Check-in [81f57a1e46]
Login
Overview
Comment:merged v1.6569-gt-crash-fix and fixed launch.scm to match
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | v1.6569-refactor-server-key-chk
Files: files | file ages | folders
SHA1: 81f57a1e465b27eddfc92b1bac36327bcb94d930
User & Date: mmgraham on 2021-01-26 18:34:36
Other Links: branch diff | manifest | tags
Context
2021-01-27
15:30
fix for crash with error (match) no matching pattern check-in: 4c4d63c3e0 user: pjhatwal tags: v1.6569-refactor-server-key-chk
2021-01-26
18:34
merged v1.6569-gt-crash-fix and fixed launch.scm to match check-in: 81f57a1e46 user: mmgraham tags: v1.6569-refactor-server-key-chk
2021-01-23
16:08
Updated megatest version to 1.6581 check-in: 36196086c3 user: mmgraham tags: v1.6569-refactor-server-key-chk
Changes

Modified db.scm from [5fbac67d93] to [f2d817bbad].

3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
       (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 ;; (sdb:qry 'getid newcomment)
				       test-id))))))
  (mt:process-triggers dbstruct run-id test-id newstate newstatus))

;; NEW BEHAVIOR: Count tests running in all runs!
;;
(define (db:get-count-tests-running dbstruct run-id fastmode)
  (let* ((qry (if fastmode
		  "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND NOT (uname = 'n/a' AND item_path = '') LIMIT 1;"
		  "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND NOT (uname = 'n/a' AND item_path = '');")))
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (db)
     (let* ((stmth (db:get-cache-stmth dbstruct  db qry)))
       (sqlite3:first-result stmth))))))







|
|
|
|







3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
       (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 ;; (sdb:qry 'getid newcomment)
				       test-id))))))
  (mt:process-triggers dbstruct run-id test-id newstate newstatus))

;; NEW BEHAVIOR: Count tests running in all runs!
;;
(define (db:get-count-tests-running dbstruct run-id) ;; fastmode)
  (let* ((qry ;; (if fastmode
		;;   "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND NOT (uname = 'n/a' AND item_path = '') LIMIT 1;"
		  "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND NOT (uname = 'n/a' AND item_path = '');")) ;; )
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (db)
     (let* ((stmth (db:get-cache-stmth dbstruct  db qry)))
       (sqlite3:first-result stmth))))))
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
      ;; "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id NOT IN (SELECT id FROM runs WHERE state='deleted') AND NOT (uname = 'n/a' AND item_path = '');")
      "SELECT count(id) FROM tests WHERE state in ('RUNNING','REMOTEHOSTSTART','LAUNCHED') AND run_id=?;" 
      run-id)))) ;; NOT IN (SELECT id FROM runs WHERE state='deleted');")

;; NEW BEHAVIOR: Look only at single run with run-id
;; 
;; (define (db:get-running-stats dbstruct run-id)
(define (db:get-count-tests-running-for-run-id dbstruct run-id fastmode)
  (let* ((qry (if fastmode
		  "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id=? LIMIT 1;"
		  "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id=?;")))
    (db:with-db
     dbstruct
     run-id
     #f
     (lambda (db)
       (let* ((stmth (db:get-cache-stmth dbstruct db qry)))
	 (sqlite3:first-result stmth run-id))))))







|
|
|
|







3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
      ;; "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id NOT IN (SELECT id FROM runs WHERE state='deleted') AND NOT (uname = 'n/a' AND item_path = '');")
      "SELECT count(id) FROM tests WHERE state in ('RUNNING','REMOTEHOSTSTART','LAUNCHED') AND run_id=?;" 
      run-id)))) ;; NOT IN (SELECT id FROM runs WHERE state='deleted');")

;; NEW BEHAVIOR: Look only at single run with run-id
;; 
;; (define (db:get-running-stats dbstruct run-id)
(define (db:get-count-tests-running-for-run-id dbstruct run-id) ;; fastmode)
  (let* ((qry ;; (if fastmode
		 ;;  "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id=? LIMIT 1;"
		  "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id=?;")) ;; )
    (db:with-db
     dbstruct
     run-id
     #f
     (lambda (db)
       (let* ((stmth (db:get-cache-stmth dbstruct db qry)))
	 (sqlite3:first-result stmth run-id))))))

Modified launch.scm from [d0067277fa] to [e8093b3e63].

729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
;; 100% COMPLETED/ (PASS,FAIL,ABORT etc.) ==> COMPLETED / X where X is same as itemized rollup
;; > 3 RUNNING with not test_dead do nothing (run should already be RUNNING/ na
;; > 0 RUNNING and test_dead then send KILLREQ ==> COMPLETED
;; 0 RUNNING ==> this is actually the first condition, should not get here

(define (launch:end-of-run-check run-id )
    (let* ((not-completed-cnt (rmt:get-not-completed-cnt run-id))  
           (running-cnt (rmt:get-count-tests-running-for-run-id run-id #f)) ;; fastmode=no
           (all-test-launched (rmt:get-var (conc "lunch-complete-" run-id)))
           (current-state (rmt:get-run-state run-id))
           (current-status (rmt:get-run-status run-id)))
     ;;get-vars run-id to query metadata table to check if all completed. if all-test-launched = yes then only not-completed-cnt = 0 means everyting is completed if no entry found in the table do nothing 
     (debug:print 0 *default-log-port* "Running test cnt :" running-cnt)                      
     (rmt:set-state-status-and-roll-up-run  run-id current-state current-status)
     (runs:update-junit-test-reporter-xml run-id) 







|







729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
;; 100% COMPLETED/ (PASS,FAIL,ABORT etc.) ==> COMPLETED / X where X is same as itemized rollup
;; > 3 RUNNING with not test_dead do nothing (run should already be RUNNING/ na
;; > 0 RUNNING and test_dead then send KILLREQ ==> COMPLETED
;; 0 RUNNING ==> this is actually the first condition, should not get here

(define (launch:end-of-run-check run-id )
    (let* ((not-completed-cnt (rmt:get-not-completed-cnt run-id))  
           (running-cnt (rmt:get-count-tests-running-for-run-id run-id))
           (all-test-launched (rmt:get-var (conc "lunch-complete-" run-id)))
           (current-state (rmt:get-run-state run-id))
           (current-status (rmt:get-run-status run-id)))
     ;;get-vars run-id to query metadata table to check if all completed. if all-test-launched = yes then only not-completed-cnt = 0 means everyting is completed if no entry found in the table do nothing 
     (debug:print 0 *default-log-port* "Running test cnt :" running-cnt)                      
     (rmt:set-state-status-and-roll-up-run  run-id current-state current-status)
     (runs:update-junit-test-reporter-xml run-id) 

Modified rmt.scm from [88741f6ca9] to [2391f353fe].

671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
	   (map (lambda (run-id)
		  (rmt:send-receive 'test-get-paths-matching-keynames-target-new run-id (list run-id keynames target res testpatt statepatt statuspatt runname)))
	   run-ids))))

(define (rmt:get-prereqs-not-met run-id waitons ref-test-name ref-item-path #!key (mode '(normal))(itemmaps #f))
  (rmt:send-receive 'get-prereqs-not-met run-id (list run-id waitons ref-test-name ref-item-path mode itemmaps)))

(define (rmt:get-count-tests-running-for-run-id run-id fastmode)
  (rmt:send-receive 'get-count-tests-running-for-run-id run-id (list run-id fastmode)))

(define (rmt:get-not-completed-cnt run-id)
  (rmt:send-receive 'get-not-completed-cnt run-id (list run-id)))


;; Statistical queries

(define (rmt:get-count-tests-running run-id fastmode)
  (rmt:send-receive 'get-count-tests-running run-id (list run-id fastmode)))

(define (rmt:get-count-tests-running-for-testname run-id testname)
  (rmt:send-receive 'get-count-tests-running-for-testname run-id (list run-id testname)))

(define (rmt:get-count-tests-running-in-jobgroup run-id jobgroup)
  (rmt:send-receive 'get-count-tests-running-in-jobgroup run-id (list run-id jobgroup)))








|
|







|
|







671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
	   (map (lambda (run-id)
		  (rmt:send-receive 'test-get-paths-matching-keynames-target-new run-id (list run-id keynames target res testpatt statepatt statuspatt runname)))
	   run-ids))))

(define (rmt:get-prereqs-not-met run-id waitons ref-test-name ref-item-path #!key (mode '(normal))(itemmaps #f))
  (rmt:send-receive 'get-prereqs-not-met run-id (list run-id waitons ref-test-name ref-item-path mode itemmaps)))

(define (rmt:get-count-tests-running-for-run-id run-id)
  (rmt:send-receive 'get-count-tests-running-for-run-id run-id (list run-id)))

(define (rmt:get-not-completed-cnt run-id)
  (rmt:send-receive 'get-not-completed-cnt run-id (list run-id)))


;; Statistical queries

(define (rmt:get-count-tests-running run-id)
  (rmt:send-receive 'get-count-tests-running run-id (list run-id)))

(define (rmt:get-count-tests-running-for-testname run-id testname)
  (rmt:send-receive 'get-count-tests-running-for-testname run-id (list run-id testname)))

(define (rmt:get-count-tests-running-in-jobgroup run-id jobgroup)
  (rmt:send-receive 'get-count-tests-running-in-jobgroup run-id (list run-id jobgroup)))

Modified runs.scm from [030b929939] to [a0489a126c].

319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
  ;; Take advantage of a good place to exit if running the one-pass methodology
  (if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
	   (args:get-arg "-one-pass"))
      (exit 0))

  (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))

  (let* ((num-running             (rmt:get-count-tests-running run-id #f)) ;; fastmode=no
	 (num-running-in-jobgroup (rmt:get-count-tests-running-in-jobgroup run-id jobgroup))
	 (job-group-limit         (let ((jobg-count (configf:lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))
    (if (> (+ num-running num-running-in-jobgroup) 0)
	(runs:inc-can-run-more-tests-count runsdat)) ;; (set! *runs:can-run-more-tests-count* (+ *runs:can-run-more-tests-count* 1)))







|







319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
  ;; Take advantage of a good place to exit if running the one-pass methodology
  (if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
	   (args:get-arg "-one-pass"))
      (exit 0))

  (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))

  (let* ((num-running             (rmt:get-count-tests-running run-id))
	 (num-running-in-jobgroup (rmt:get-count-tests-running-in-jobgroup run-id jobgroup))
	 (job-group-limit         (let ((jobg-count (configf:lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))
    (if (> (+ num-running num-running-in-jobgroup) 0)
	(runs:inc-can-run-more-tests-count runsdat)) ;; (set! *runs:can-run-more-tests-count* (+ *runs:can-run-more-tests-count* 1)))
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
				       ))
				   extras)
				  extras)
				'())))
	     (waitons     (delete-duplicates (append (tests:testqueue-get-waitons test-record) extra-waits) equal?))
	     (newtal      (append tal (list hed)))
	     (regfull     (>= (length reg) reglen))
	     (num-running (rmt:get-count-tests-running-for-run-id run-id #t)) ;; fastmode=yes
	     (testdat     (make-runs:testdat
			   hed: hed
			   tal: tal
			   reg: reg
			   reruns: reruns
			   test-record: test-record
			   test-name:   test-name







|







1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
				       ))
				   extras)
				  extras)
				'())))
	     (waitons     (delete-duplicates (append (tests:testqueue-get-waitons test-record) extra-waits) equal?))
	     (newtal      (append tal (list hed)))
	     (regfull     (>= (length reg) reglen))
	     (num-running (rmt:get-count-tests-running-for-run-id run-id))
	     (testdat     (make-runs:testdat
			   hed: hed
			   tal: tal
			   reg: reg
			   reruns: reruns
			   test-record: test-record
			   test-name:   test-name
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
    (rmt:set-var (conc "lunch-complete-" run-id) "yes")  
        
    ;; now *if* -run-wait we wait for all tests to be done
    ;; Now wait for any RUNNING tests to complete (if in run-wait mode)
    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
    (thread-sleep! 10) ;; I think there is a race condition here. Let states/statuses settle
    
    (let wait-loop ((num-running      (rmt:get-count-tests-running-for-run-id run-id #t)) ;; fastmode=yes
		    (prev-num-running 0))
      ;; (debug:print-info 13 *default-log-port* "num-running=" num-running ", prev-num-running=" prev-num-running)
      (if (and (or (args:get-arg "-run-wait")
		   (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)(+ last-time-incomplete 900))
		(let ((actual-num-running (rmt:get-count-tests-running-for-run-id run-id #f))) ;; fastmode=no
		  (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)
		  (debug:print-info 0 *default-log-port* "run-wait specified, waiting on " actual-num-running
				    " tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at "
				    (time->string (seconds->local-time (current-seconds))))))
	    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
	    (thread-sleep! 5) ;; (if (>= num-running max-concurrent-jobs) 5 1))
	    (wait-loop (rmt:get-count-tests-running-for-run-id run-id #t) ;; fastmode=yes
		       num-running))))
    ;; LET* ((test-record
    ;; we get here on "drop through". All done!
    ;; this is moved to runs:run-testes since this function is getting called twice to ensure everthing is completed. 
    ;; (debug:print-info 0 *default-log-port* "Calling Post Hook")    
    ;; (runs:run-post-hook run-id)
    (debug:print-info 1 *default-log-port* "All tests launched")))







|









|









|







1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
    (rmt:set-var (conc "lunch-complete-" run-id) "yes")  
        
    ;; now *if* -run-wait we wait for all tests to be done
    ;; Now wait for any RUNNING tests to complete (if in run-wait mode)
    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
    (thread-sleep! 10) ;; I think there is a race condition here. Let states/statuses settle
    
    (let wait-loop ((num-running      (rmt:get-count-tests-running-for-run-id run-id))
		    (prev-num-running 0))
      ;; (debug:print-info 13 *default-log-port* "num-running=" num-running ", prev-num-running=" prev-num-running)
      (if (and (or (args:get-arg "-run-wait")
		   (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)(+ 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)
		  (debug:print-info 0 *default-log-port* "run-wait specified, waiting on " actual-num-running
				    " tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at "
				    (time->string (seconds->local-time (current-seconds))))))
	    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
	    (thread-sleep! 5) ;; (if (>= num-running max-concurrent-jobs) 5 1))
	    (wait-loop (rmt:get-count-tests-running-for-run-id run-id)
		       num-running))))
    ;; LET* ((test-record
    ;; we get here on "drop through". All done!
    ;; this is moved to runs:run-testes since this function is getting called twice to ensure everthing is completed. 
    ;; (debug:print-info 0 *default-log-port* "Calling Post Hook")    
    ;; (runs:run-post-hook run-id)
    (debug:print-info 1 *default-log-port* "All tests launched")))