Megatest

Check-in [d125f9632a]
Login
Overview
Comment:Corrected exit delays, added attempted repair on discovering a corrupted db
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | v1.60
Files: files | file ages | folders
SHA1: d125f9632a36ba7019e7d90fa22863df8219b8b4
User & Date: matt on 2015-05-28 00:18:04
Other Links: branch diff | manifest | tags
Context
2015-05-28
22:55
Added basis for test of legacy vs. current db check-in: 0457c113cf user: matt tags: v1.60
06:57
Added basic template for getting changed runs. Fixed exit handling, exit promptly after cleanup check-in: 33295b94b3 user: matt tags: v1.60
00:18
Corrected exit delays, added attempted repair on discovering a corrupted db check-in: d125f9632a user: matt tags: v1.60
2015-05-27
11:40
Removed exit from on-exit call. check-in: b4a4de8e9e user: mrwellan tags: v1.60, v1.6014a
Changes

Modified common.scm from [9be5018f3c] to [005a48a275].

279
280
281
282
283
284
285
286
287
288



289
290
291
292
293
294
295
279
280
281
282
283
284
285



286
287
288
289
290
291
292
293
294
295







-
-
-
+
+
+







					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  (vector-set! *task-db* 0 #f)))))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
				  (thread-sleep! 5) ;; give the clean up few seconds to do it's stuff
				  (thread-sleep! 1))
			      (debug:print 0 "       Done.")
				  (thread-sleep! 0.1) ;; give the clean up few seconds to do it's stuff
				  (thread-sleep! 4))
			      (debug:print 4 " ... done")
			      )
			    "clean exit")))
      (thread-start! th2)
      (thread-start! th1)
      (thread-join! th2))))

(define (std-signal-handler signum)

Modified db.scm from [5f7db1b5aa] to [9ce696701e].

160
161
162
163
164
165
166
167
168
169
170
171
172
173







174
175
176
177
178
179
180
181
182
183
184
185















186
187
188
189
190
191
192
160
161
162
163
164
165
166







167
168
169
170
171
172
173












174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195







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







    (sqlite3:execute db (conc "PRAGMA synchronous = " (or syncprag 1) ";"))))

;; open an sql database inside a file lock
;;
;; returns: db existed-prior-to-opening
;;
(define (db:lock-create-open fname initproc)
  (if (file-exists? fname)
      (let ((db (sqlite3:open-database fname)))
	(sqlite3:set-busy-handler! db (make-busy-timeout 136000))
	(db:set-sync db) ;; (sqlite3:execute db "PRAGMA synchronous = 0;")
	db)
      (let* ((parent-dir   (pathname-directory fname))
	     (dir-writable (file-write-access? parent-dir)))
  ;; (if (file-exists? fname)
  ;;     (let ((db (sqlite3:open-database fname)))
  ;;       (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
  ;;       (db:set-sync db) ;; (sqlite3:execute db "PRAGMA synchronous = 0;")
  ;;       db)
  (let* ((parent-dir   (pathname-directory fname))
	 (dir-writable (file-write-access? parent-dir))
	(if dir-writable
	    (let ((exists  (file-exists? fname))
		  (lock    (obtain-dot-lock fname 1 5 10))
		  (db      (sqlite3:open-database fname)))
	      (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
	      (db:set-sync db) ;; (sqlite3:execute db "PRAGMA synchronous = 0;")
	      (if (not exists)(initproc db))
	      (release-dot-lock fname)
	      db)
	    (begin
	      (debug:print 0 "ERROR: no such db in non-writable dir " fname)
	      (sqlite3:open-database fname))))))
	 (file-exists  (file-exists? fname))
	 (file-write   (if file-exists
			   (file-write-access? fname)
			   dir-writable )))
    (if file-write ;; dir-writable
	(let ((lock    (obtain-dot-lock fname 1 5 10))
	      (db      (sqlite3:open-database fname)))
	  (sqlite3:set-busy-handler! db (make-busy-timeout 136000))
	  (db:set-sync db) ;; (sqlite3:execute db "PRAGMA synchronous = 0;")
	  (if (not file-exists)(initproc db))
	  (release-dot-lock fname)
	  db)
	(begin
	  (debug:print 0 "ERROR: no such db in non-writable dir " fname)
	  (sqlite3:open-database fname))))) ;; )

;; This routine creates the db. It is only called if the db is not already opened
;; 
(define (db:open-rundb dbstruct run-id #!key (attemptnum 0)(do-not-open #f)) ;;  (conc *toppath* "/megatest.db") (car *configinfo*)))
  (let* ((local  (dbr:dbstruct-get-local dbstruct))
	 (rdb    (if local
		     (dbr:dbstruct-get-localdb dbstruct run-id)
484
485
486
487
488
489
490
491











































































492
493
494
495
496
497
498
499

500
501
502
503
504
505
506

507





508


509
510
511
512




513
514
515
516
517
518
519
487
488
489
490
491
492
493

494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585

586
587
588
589
590
591
592
593




594
595
596
597
598
599
600
601
602
603
604







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








+







+
-
+
+
+
+
+

+
+
-
-
-
-
+
+
+
+







	   '("description"    #f)
	   '("reviewed"       #f)
	   '("iterated"       #f)
	   '("avg_runtime"    #f)
	   '("avg_disk"       #f)
	   '("tags"           #f)
	   '("jobgroup"       #f)))))
    

;; use bunch of Unix commands to try to break the lock and recreate the db
;;
(define (db:move-and-recreate-db dbdat)
  (let* ((dbpath   (db:dbdat-get-path        dbdat))
	 (dbdir    (pathname-directory       dbpath))
	 (fname    (pathname-strip-directory dbpath))
	 (fnamejnl (conc fname "-journal"))
	 (tmpname  (conc fname "." (current-process-id)))
	 (tmpjnl   (conc fnamejnl "." (current-process-id))))
    (debug:print 0 "ERROR: " fname " appears corrupted. Making backup \"old/" fname "\"")
    (system (conc "cd " dbdir ";mkdir -p old;cat " fname " > old/" tmpname))
    (system (conc "rm -f " dbpath))
    (if (file-exists? fnamejnl)
	(begin
	  (debug:print 0 "ERROR: " fnamejnl " found, moving it to old dir as " tmpjnl)
	  (system (conc "cd " dbdir ";mkdir -p old;cat " fnamejnl " > old/" tmpjnl))
	  (system (conc "rm -f " dbdir "/" fnamejnl))))
    ;; attempt to recreate database
    (system (conc "cd " dbdir ";sqlite3 old/" tmpname " .dump | sqlite3 " fname))))
    
;; return #f to indicate the dbdat should be closed/reopened
;; else return dbdat
;;
(define (db:repair-db dbdat #!key (numtries 1))
  (let* ((dbpath   (db:dbdat-get-path        dbdat))
	 (dbdir    (pathname-directory       dbpath))
	 (fname    (pathname-strip-directory dbpath)))
    (debug:print-info 0 "Checking db " dbpath " for errors.")
    (cond
     ((not (file-write-access? dbdir))
      (debug:print 0 "WARNING: can't write to " dbdir ", can't fix " fname)
      #f)

     ;; handle special cases, megatest.db and monitor.db
     ;; 
     ;;  NOPE: apply this same approach to all db files
     ;;
     (else ;; ((equal? fname "megatest.db") ;; this file can be regenerated if needed
      (handle-exceptions
       exn
       (begin
	 (db:move-and-recreate-db dbdat)
	 (if (> numtries 0)
	     (db:repair-db dbdat numtries: (- numtries 1))
	     #f)
	 (debug:print 0 "FATAL: file " dbpath " was found corrupted, an attempt to fix has been made but you must start over.")
	 (debug:print 0
		      "   check the following:\n"
		      "      1. full directories, look in ~/ /tmp and " dbdir "\n"
		      "      2. write access to " dbdir "\n\n"
		      "   if the automatic recovery failed you may be able to recover data by doing \"" 
		      (if (member fname '("megatest.db" "monitor.db"))
			  "megatest -cleanup-db"
			  "megatest -import-megatest.db;megatest -cleanup-db")
		      "\"\n")
	 (exit) ;; we can not safely continue when a db was corrupted - even if fixed.
	 )
       ;; test read/write access to the database
       (let ((db (sqlite3:open-database dbpath)))
	 (cond
	  ((equal? fname "megatest.db")
	   (sqlite3:execute db "DELETE FROM tests WHERE state='DELETED';"))
	  ((equal? fname "main.db")
	   (sqlite3:execute db "DELETE FROM runs WHERE state='deleted';"))
	  ((string-match "\\d.db" fname)
	   (sqlite3:execute db "UPDATE tests SET state='DELETED' WHERE state='DELETED';"))
	  ((equal? fname "monitor.db")
	   (sqlite3:execute "DELETE FROM servers WHERE state LIKE 'defunct%';"))
	  (else
	   (sqlite3:execute db "vacuum;")))
	 
	 (finalize! db)
	 #t))))))
    
;; tbls is ( ("tablename" ( "field1" [#f|proc1] ) ( "field2" [#f|proc2] ) .... ) )
;; db's are dbdat's
;;
(define (db:sync-tables tbls fromdb todb . slave-dbs)
  (mutex-lock! *db-sync-mutex*)
  (handle-exceptions
   exn
   (begin
     (mutex-unlock! *db-sync-mutex*)
     (debug:print 0 "EXCEPTION: database probably overloaded or unreadable in db:sync-tables.")
     (print-call-chain (current-error-port))
     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
     (print "exn=" (condition->list exn))
     (debug:print 0 " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
     (debug:print 0 " src db:  " (db:dbdat-get-path fromdb))
     (for-each (lambda (dbdat)
		 (let ((dbpath (db:dbdat-get-path dbdat)))
		 (debug:print 0 " dbpath:  " (db:dbdat-get-path dbdat)))
		   (debug:print 0 " dbpath:  " dbpath)
		   (if (not (db:repair-db dbdat))
		       (begin
			 (debug:print 0 "ERROR: Failed to rebuild " dbpath ", exiting now.")
			 (exit)))))
	       (cons todb slave-dbs))
     
     0)
     (if *server-run* ;; we are inside a server, throw a sync-failed error
	 (signal (make-composite-condition
		 (make-property-condition 'sync-failed 'message "db:sync-tables failed in a server context.")))
	 0)) ;; return zero for num synced
;;      (if *server-run* ;; we are inside a server, throw a sync-failed error
;; 	 (signal (make-composite-condition
;; 		 (make-property-condition 'sync-failed 'message "db:sync-tables failed in a server context.")))
;; 	 0)) ;; return zero for num synced

	 ;; (set! *time-to-exit* #t) ;; let watch dog know that it is time to die.
	 ;; (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "shutting-down")
	 ;; (portlogger:open-run-close portlogger:set-port port "released")
	 ;; (exit 1)))
   (cond
    ((not fromdb) (debug:print 3 "WARNING: db:sync-tables called with fromdb missing") -1)