Megatest

(declare (unit common))
(declare (uses commonmod))
(import commonmod)

(include "common_records.scm")


;; (require-library margs)
;; (include "margs.scm")

;; (define old-exit exit)
;; 
;; (define (exit . code)
;;   (if (null? code)
;;       (old-exit)
;;       (old-exit code)))

(define (stop-the-train)
  (thread-start! (make-thread (lambda ()
				(let loop ()
				  (if (and *toppath*
					   (file-exists? (conc *toppath*"/stop-the-train")))
				      (begin
					(debug:print 0 *default-log-port* "ERROR: found file "*toppath*"/stop-the-train, exiting immediately")

					(exit 1)))
				  (thread-sleep! 5)
				  (loop))))))

;; execute thunk, return value.  If exception thrown, trap exception, return #f, and emit nonfatal condition note to *default-log-port* .
;; arguments - thunk, message
(define (common:fail-safe thunk warning-message-on-exception)

                                 (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 30000))
				 (if sync-mode
				     (sqlite3:execute db (conc "PRAGMA synchronous = "sync-mode";")))
				 (if journal-mode
				     (sqlite3:execute db (conc "PRAGMA journal_mode = "journal-mode";")))
				 (if (and init-proc (not db-exists))
				     (init-proc db))
				 db)))

                            (begin
			      (if (file-exists? fname )
                                  (let ((db (sqlite3:open-database fname)))
				    ;; pragmas synchronous not needed because this db is used read-only
				    ;; (sqlite3:execute db (conc "PRAGMA synchronous = "mode";")
				    (sqlite3:set-busy-handler! db (sqlite3:make-busy-timeout 30000)) ;; read-only but still need timeout
				    db )
                                  (print "file doesn't exist: " fname))))
			(exn (io-error)
			     (dbfile:print-err exn "ERROR: i/o error with " fname ". Check permissions, disk space etc. and try again.")
			     (retry))
			(exn (corrupt)
			     (dbfile:print-err exn "ERROR: database " fname " is corrupt. Repair it to proceed.")
			     (retry))
			(exn (busy)

	  (thread-sleep! 0.25)
	  (if (file-exists? fname)
	      (handle-exceptions exn
                #f 
                (with-input-from-file fname
	  	  (lambda ()
		    (equal? key-string (read-line)))))


	      #f)


       )
    )
  )
)

(define (dbfile:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))

	      (list  "MT_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
	      (list  "MT_TESTSUITENAME" (common:get-testsuite-name))))
          ;;(bb-check-path msg: "launch:execute post block 3")




	  (if mt-bindir-path (setenv "PATH" (conc "\""(getenv "PATH")":"mt-bindir-path"\"")))
          ;;(bb-check-path msg: "launch:execute post block 4")
	  ;; (change-directory top-path)
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

	  
	  ;; environment overrides are done *before* the remaining critical envars.

		       runname
		       (common:file-exists? fulldir))
		  (let ((tmpfile  (conc fulldir "/.megatest.cfg." (current-seconds)))
			(targfile (conc fulldir "/.megatest.cfg-"  megatest-version "-" megatest-fossil-hash))
			(rconfig  (conc fulldir "/.runconfig." megatest-version "-" megatest-fossil-hash)))
		    (if (common:file-exists? rconfig) ;; only cache megatest.config AFTER runconfigs has been cached
			(begin
			  (debug:print-info 0 *default-log-port* "Caching megatest.config in " tmpfile)
                          (if (not (common:in-running-test?))
                              (configf:write-alist *configdat* tmpfile))
			  (system (conc "ln -sf " tmpfile " " targfile))))
		    )))
	    (debug:print-info 1 *default-log-port* "No linktree yet, no caching configs.")))))

(dbfile:db-init-proc db:initialize-main-db)

;; load the ~/.megatestrc file, put (use trace)(trace-call-sites #t)(trace function-you-want-to-trace) in this file
;;
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.megatestrc")))
  (if (common:file-exists? debugcontrolf)



      (load debugcontrolf)))




;; usage logging, careful with this, it is not designed to deal with all real world challenges!
;;
(if (and *usage-log-file*
         (file-write-access? *usage-log-file*))
    (with-output-to-file
        *usage-log-file*

(declare (unit portlogger))
(declare (uses db))

;; lsof -i

(define (portlogger:open-db fname)
  (let* ((avail    (tasks:wait-on-journal fname 5 remove: #t)) ;; wait up to about 10 seconds for the journal to go away

	 (exists   (common:file-exists? fname))
	 (db       (if avail 
		       (sqlite3:open-database fname)
		       (begin
			 (system (conc "rm -f " fname))
			 (sqlite3:open-database fname))))
	 (handler  (sqlite3:make-busy-timeout 136000))

            port INTEGER PRIMARY KEY,
            state TEXT DEFAULT 'not-used',
            fail_count INTEGER DEFAULT 0,
            update_time TIMESTAMP DEFAULT (strftime('%s','now')) );")
    db))

(define (portlogger:open-run-close proc . params)
  (let* ((fname  (conc "/tmp/." (current-user-name) "-portlogger.db"))
	 (avail  (tasks:wait-on-journal fname 10))) ;; wait up to about 10 seconds for the journal to go away
    (handle-exceptions
     exn
     (begin
       ;; (release-dot-lock fname)
       (debug:print-error 0 *default-log-port* "portlogger:open-run-close failed. " proc " " params)
       (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
       (debug:print 5 *default-log-port* "exn=" (condition->list exn))

    (debug:print-info 13 *default-log-port* "rmt:send-receive, case  9. conninfo=" conninfo " dat=" dat " runremote = " runremote)
    (mutex-unlock! *rmt-mutex*)
    (if success ;; success only tells us that the transport was
	;; successful, have to examine the data to see if
	;; there was a detected issue at the other end
	(extras-transport-succeded *default-log-port* *rmt-mutex* attemptnum runremote res params rid cmd)
	(begin
           (debug:print-error 0 *default-log-port* " dat=" dat) 
           (extras-transport-failed *default-log-port* *rmt-mutex* attemptnum runremote cmd rid params))
	)))

(define (rmt:print-db-stats)
  (let ((fmtstr "~40a~7-d~9-d~20,2-f")) ;; "~20,2-f"
    (debug:print 18 *default-log-port* "DB Stats\n========")
    (debug:print 18 *default-log-port* (format #f "~40a~8a~10a~10a" "Cmd" "Count" "TotTime" "Avg"))

	 (all-dat    (cdr run-dat))
	 (tests-data (alist-ref "data" all-dat equal?))
	 (run-meta   (alist-ref "meta" all-dat equal?))
         (run-id     (string->number (alist-ref "id"   run-meta equal?))))
    (rmt:insert-run run-id target runname run-meta)
    (if (list? tests-data)
      (begin

        (for-each
          (lambda (test-dat)
            (let* ((test-id  (car test-dat))
	      (test-rec (cdr test-dat)))
	      (rmt:insert-test run-id test-rec)))
         tests-data)
      )

	 (item-path (alist-ref "item_path" test-rec equal?))
         (test-id (rmt:get-test-id run-id testname item-path))
         )

    (if test-id
       (debug:print 0 *default-log-port* "test "testname"/"item-path " already exists in run-id " run-id)
       (begin
         (debug:print 0 *default-log-port* " Insert test in run "run-id": "testname"/"item-path)
         (rmt:send-receive 'insert-test run-id test-rec)
       )
    )
  )
)

    (if (not (null? required-tests))
	(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)
							       (rmt:find-and-mark-incomplete run-id #f)))) ;; 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" "ABORT,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 ))   

     (lambda (key)
       (let* ((idx (cadr key))
	      (fld (car  key))
	      (val (configf:lookup test-conf "test_meta" fld)))
	 ;; (debug:print 5 *default-log-port* "idx: " idx " fld: " fld " val: " val)
	 (if (and val (not (equal? (vector-ref currrecord idx) val)))
	     (begin
	       (debug:print 0 *default-log-port* "Updating " test-name " " fld " to " val)
	       (rmt:testmeta-update-field test-name fld val)))))
     '(("author" 2)("owner" 3)("description" 4)("reviewed" 5)("tags" 9)("jobgroup" 10)))))

;; find tests with matching tags, tagpatt is a string "tagpatt1,tagpatt2%, ..."
;;
(define (runs:get-tests-matching-tags tagpatt)
  (let* ((tagdata (rmt:get-tests-tags))

		(if (and tcfg cache-file) (hash-table-set! tcfg "have fulldata" #t)) ;; mark this as fully read data
		(if tcfg (hash-table-set! *testconfigs* test-full-name tcfg))
		(if (and testexists
			 cache-file
			 (file-write-access? cache-path)
			 allow-write-cache)
		    (let ((tpath (conc cache-path "/.testconfig")))
		      (debug:print-info 1 *default-log-port* "Caching testconfig for " test-name " in " tpath)
                      (if (and tcfg (not (common:in-running-test?)))
                          (configf:write-alist tcfg tpath))))
		tcfg))))))
  
;; sort tests by priority and waiton
;; Move test specific stuff to a test unit FIXME one of these days
(define (tests:sort-by-priority-and-waiton test-records)