Megatest

Check-in [ef4d658125]
Login
Overview
Comment:Improved lock out on running tests. Added weak work-around for issue with preclean
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | v1.60
Files: files | file ages | folders
SHA1: ef4d6581250ee9656b764ea2ac028384686bf516
User & Date: matt on 2014-10-13 01:09:51
Other Links: branch diff | manifest | tags
Context
2014-10-13
10:43
Wrapped db init with transaction. Dunno why I didn't think of that a long time ago. check-in: d494a8975c user: mrwellan tags: v1.60
01:09
Improved lock out on running tests. Added weak work-around for issue with preclean check-in: ef4d658125 user: matt tags: v1.60
2014-10-10
10:49
bumped version check-in: 69da7e37ab user: mrwellan tags: v1.60
Changes

Modified http-transport.scm from [6063d55ba0] to [6e6db43f0a].

413
414
415
416
417
418
419


420
421
422
423
424
425
426
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428







+
+







      (mutex-lock! *heartbeat-mutex*)
      (set! last-access *last-db-access*)
      (mutex-unlock! *heartbeat-mutex*)

      ;; (debug:print 11 "last-access=" last-access ", server-timeout=" server-timeout)
      ;;
      ;; no_traffic, no running tests, if server 0, no running servers
      ;;
      ;; (let ((wait-on-running (configf:lookup *configdat* "server" "wait-on-running"))) ;; wait on running tasks (if not true then exit on time out)
      ;;
      (if (and *server-run*
	       (or (> (+ last-access server-timeout)
		      (current-seconds))
		   (and (eq? run-id 0)
			(> (tasks:num-servers-non-zero-running tdb) 0))
		   (and (not (eq? run-id 0)) ;; only makes sense in non-zero run-id servers

Modified launch.scm from [8269db9adc] to [15a97345bd].

89
90
91
92
93
94
95
96
97
98
99
100
101
102
103








104
105
106
107
108
109
110
89
90
91
92
93
94
95








96
97
98
99
100
101
102
103
104
105
106
107
108
109
110







-
-
-
-
-
-
-
-
+
+
+
+
+
+
+
+







	                                  (if (and (file-exists? fulln)
                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
	       (rollup-status 0))
	  (change-directory top-path)




	  ;; ADD here - is test already RUNNING? If so ---- ABORT RUN ATTEMPT




	  ;; Do not run the test if it is REMOVING, RUNNING, KILLREQ or REMOTEHOSTSTART,
	  ;; Mark the test as REMOTEHOSTSTART *IMMEDIATELY*
	  ;;
	  (let ((test-info (rmt:get-testinfo-state-status run-id test-id)))
	    (if (not (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ")))
		(tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
		(debug:print 0 "ERROR: test state is " (db:test-get-state test-info) ", cannot proceed")))
	  
	  (debug:print 2 "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
	  (set! keys       (rmt:get-keys))
	  ;; (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals) ;; these may be needed by the launching process
	  ;; one of these is defunct/redundant ...
	  (if (not (launch:setup-for-run force: #t))
	      (begin
		(debug:print 0 "Failed to setup, exiting") 
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
171
172
173
174
175
176
177


178
179
180
181
182
183
184







-
-







	  (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:set-full-meta-info test-id run-id 0 work-area)
	  (tests:set-full-meta-info #f test-id run-id 0 work-area 10)

	  ;; (tests:test-set-status! test-id "REMOTEHOSTSTART" "n/a" (args:get-arg "-m") #f)
	  (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
	  (thread-sleep! 0.3) ;; NFS slowness has caused grief here

	  (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
522
523
524
525
526
527
528
529

530
531
532
533
534
535
536
520
521
522
523
524
525
526

527
528
529
530
531
532
533
534







-
+







;; 
;;  <linkdir> - <target> - <testname> [ - <itempath> ]
;; 
;; All log file links should be stored relative to the top of link path
;;  
;; <target> - <testname> [ - <itempath> ] 
;;
(define (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat)
(define (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat #!key (remtries 2))
  (let* ((item-path (item-list->path itemdat))
	 (runname  (db:get-value-by-header (db:get-rows run-info)
					   (db:get-header run-info)
					   "runname"))
	 ;; convert back to db: from rdb: - this is always run at server end
	 (target   (string-intersperse (map cadr keyvals) "/"))

544
545
546
547
548
549
550
551
552
553




554
555
556
557
558
559
560
542
543
544
545
546
547
548



549
550
551
552
553
554
555
556
557
558
559







-
-
-
+
+
+
+







	 (toptest-path (conc disk-path "/" testtop-base))
	 (test-path    (conc disk-path "/" test-base))

	 ;; ensure this exists first as links to subtests must be created there
	 (linktree  (let ((rd (config-lookup *configdat* "setup" "linktree")))
		      (if rd rd (conc *toppath* "/runs"))))

	 (lnkbase  (conc linktree "/" target "/" runname))
	 (lnkpath  (conc lnkbase "/" testname))
	 (lnkpathf (conc lnkpath (if not-iterated "" "/") item-path)))
	 (lnkbase   (conc linktree "/" target "/" runname))
	 (lnkpath   (conc lnkbase "/" testname))
	 (lnkpathf  (conc lnkpath (if not-iterated "" "/") item-path))
	 (lnktarget (conc lnkpath "/" item-path)))

    ;; Update the rundir path in the test record for all, rundir=physical, shortdir=logical
    ;;                                                 rundir   shortdir
    (rmt:general-call 'test-set-rundir-shortdir run-id lnkpathf test-path testname item-path)

    (debug:print 2 "INFO:\n       lnkbase=" lnkbase "\n       lnkpath=" lnkpath "\n  toptest-path=" toptest-path "\n     test-path=" test-path)
    (if (not (file-exists? linktree))
629
630
631
632
633
634
635
636

637
638
639
640
641
642
643
628
629
630
631
632
633
634

635
636
637
638
639
640
641
642







-
+







		(debug:print-info 2 "Creating " toptest-path " and link " lnkpath)
		(create-directory toptest-path #t)
		(hash-table-set! *toptest-paths* testname toptest-path)))))

    ;; The toptest path has been created, the link to the test in the linktree has
    ;; been created. Now, if this is an iterated test the real test dir must be created
    (if (not not-iterated) ;; this is an iterated test
	(let ((lnktarget (conc lnkpath "/" item-path)))
	(begin ;; (let ((lnktarget (conc lnkpath "/" item-path)))
	  (debug:print 2 "Setting up sub test run area")
	  (debug:print 2 " - creating run area in " test-path)
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print 0 "ERROR:  Failed to create directory " test-path ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit 1))
651
652
653
654
655
656
657



658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673





674

675
676
677
678
679
680
681
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680

681
682
683
684
685
686
687
688







+
+
+
















+
+
+
+
+
-
+







	   exn
	   (begin
	     (debug:print 0 "ERROR:  Failed to re-create link " linktarget ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit))
	   (if (symbolic-link? lnktarget)     (delete-file lnktarget))
	   (if (not (file-exists? lnktarget)) (create-symbolic-link test-path lnktarget)))))

    (if (not (directory? test-path))
	(create-directory test-path #t)) ;; this is a hack, I don't know why out of the blue this path does not exist sometimes

    (if (directory? test-path)
	(begin
	  (let* ((ovrcmd (let ((cmd (config-lookup *configdat* "setup" "testcopycmd")))
			   (if cmd
			       ;; substitute the TEST_SRC_PATH and TEST_TARG_PATH
			       (string-substitute "TEST_TARG_PATH" test-path
						  (string-substitute "TEST_SRC_PATH" test-src-path cmd #t) #t)
			       #f)))
		 (cmd    (if ovrcmd 
			     ovrcmd
			     (conc "rsync -av" (if (debug:debug-mode 1) "" "q") " " test-src-path "/ " test-path "/"
				   " >> " test-path "/mt_launch.log 2>> " test-path "/mt_launch.log")))
		 (status (system cmd)))
	    (if (not (eq? status 0))
		(debug:print 2 "ERROR: problem with running \"" cmd "\"")))
	  (list lnkpathf lnkpath ))
	(if (> remtries 0)
	    (begin
	      (debug:print 0 "ERROR: Failed to create work area at " test-path " with link at " lnktarget ", remaining attempts " remtries)
	      ;; 
	      (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat remtries: (- remtries 1)))
	(list #f #f))))
	    (list #f #f)))))

;; 1. look though disks list for disk with most space
;; 2. create run dir on disk, path name is meaningful
;; 3. create link from run dir to megatest runs area 
;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
739
740
741
742
743
744
745





746
747
748
749
750
751
752
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764







+
+
+
+
+







    (if launcher (set! launcher (string-split launcher)))
    ;; set up the run work area for this test
    (if (and (args:get-arg "-preclean") ;; user has requested to preclean for this run
	     (not (member (db:test-get-rundir testinfo)(list "n/a" "/tmp/badname")))) ;; n/a is a placeholder and thus not a read dir
	(begin
	  (debug:print-info 0 "attempting to preclean directory " (db:test-get-rundir testinfo) " for test " test-name "/" item-path)
	  (runs:remove-test-directory #f testinfo #t))) ;; remove data only, do not perturb the record

    ;; prevent overlapping actions - set to LAUNCHED as early as possible
    ;;
    (tests:test-set-status! run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
    (rmt:roll-up-pass-fail-counts run-id test-name item-path "LAUNCHED")
    (set! diskpath (get-best-disk *configdat*))
    (if diskpath
	(let ((dat  (create-work-area run-id run-info keyvals test-id test-path diskpath test-name itemdat)))
	  (set! work-area (car dat))
	  (set! toptest-work-area (cadr dat))
	  (debug:print-info 2 "Using work area " work-area))
	(begin
772
773
774
775
776
777
778

779
780
781
782
783
784
785
786
787
788
789
790
784
785
786
787
788
789
790
791
792
793
794


795
796
797
798
799
800
801







+



-
-







				      (list 'ezsteps   ezsteps) 
				      (list 'target    mt_target)
				      (list 'runtlim   (if run-time-limit (common:hms-string->seconds run-time-limit) #f))
				      (list 'env-ovrd  (hash-table-ref/default *configdat* "env-override" '())) 
				      (list 'set-vars  (if params (hash-table-ref/default params "-setvars" #f)))
				      (list 'runname   runname)
				      (list 'mt-bindir-path mt-bindir-path))))))))

    ;; clean out step records from previous run if they exist
    ;; (rmt:delete-test-step-records run-id test-id)
    (change-directory work-area) ;; so that log files from the launch process don't clutter the test dir
    (tests:test-set-status! run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
    (rmt:roll-up-pass-fail-counts run-id test-name item-path "LAUNCHED")
    (cond
     ((and launcher hosts) ;; must be using ssh hostname
      (set! fullcmd (append launcher (car hosts)(list remote-megatest test-sig "-execute" cmdparms) debug-param)))
     ;; (set! fullcmd (append launcher (car hosts)(list remote-megatest test-sig "-execute" cmdparms))))
     (launcher
      (set! fullcmd (append launcher (list remote-megatest test-sig "-execute" cmdparms) debug-param)))
     ;; (set! fullcmd (append launcher (list remote-megatest test-sig "-execute" cmdparms))))

Modified runs.scm from [8478824409] to [93a7d492d9].

1068
1069
1070
1071
1072
1073
1074
1075

1076
1077
1078
1079
1080
1081
1082
1068
1069
1070
1071
1072
1073
1074

1075
1076
1077
1078
1079
1080
1081
1082







-
+







	  (begin
	    ;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
	    ;; (debug:print 0 "Got here eh! num-running=" num-running " (> num-running 0) " (> num-running 0))
	    (if (> (current-seconds)(+ last-time-incomplete 900))
		(begin
		  (debug:print-info 0 "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))
		  (rmt:find-and-mark-incomplete run-id)))
		  (rmt:find-and-mark-incomplete run-id #f)))
	    (if (not (eq? num-running prev-num-running))
		(debug:print-info 0 "run-wait specified, waiting on " num-running " tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at " (time->string (seconds->local-time (current-seconds)))))
	    (thread-sleep! 15)
	    ;; (wait-loop (rmt:get-count-tests-running-for-run-id run-id) num-running))))
	    (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!
1503
1504
1505
1506
1507
1508
1509
1510


1511
1512
1513
1514
1515
1516
1517
1503
1504
1505
1506
1507
1508
1509

1510
1511
1512
1513
1514
1515
1516
1517
1518







-
+
+







  #t)

(define (runs:remove-test-directory db test remove-data-only)
  (let* ((run-dir       (db:test-get-rundir test))    ;; run dir is from the link tree
	 (real-dir      (if (file-exists? run-dir)
			    (resolve-pathname run-dir)
			    #f)))
    (if (not remove-data-only)
    (if remove-data-only
	(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "CLEANING" "LOCKED" #f)
	(mt:test-set-state-status-by-id (db:test-get-run_id test)(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)
1537
1538
1539
1540
1541
1542
1543
1544


1545
1546
1547
1548
1549
1550
1551
1538
1539
1540
1541
1542
1543
1544

1545
1546
1547
1548
1549
1550
1551
1552
1553







-
+
+







		 (delete-directory run-dir)))
	    (if (and run-dir
		     (not (member run-dir (list "n/a" "/tmp/badname"))))
		(debug:print 0 "WARNING: not removing " run-dir " as it either doesn't exist or is not a symlink")
		(debug:print 0 "NOTE: the run dir for this test is undefined. Test may have already been deleted."))
	    ))
    ;; Only delete the records *after* removing the directory. If things fail we have a record 
    (if (not remove-data-only)
    (if remove-data-only
	(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "NOT_STARTED" "n/a" #f)
	(rmt:delete-test-records (db:test-get-run_id test) (db:test-get-id test)))))

;;======================================================================
;; Routines for manipulating runs
;;======================================================================

;; Since many calls to a run require pretty much the same setup