Megatest

(declare (uses db))
(declare (uses common))
(declare (uses items))
(declare (uses runconfig))
(declare (uses tests))
(declare (uses server))
(declare (uses mt))
(declare (uses archive))
;; (declare (uses filedb))

(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")
(include "run_records.scm")
(include "test_records.scm")

    (if (> (- currtime lasttime) waitval)
	(begin
	  (hash-table-set! *runs:denoise* key currtime)
	  #t)
	#f)))

(define (runs:can-run-more-tests run-id jobgroup max-concurrent-jobs)
  (thread-sleep! (cond
        	  ((> *runs:can-run-more-tests-count* 20) 2);; obviously haven't had any work to do for a while
        	  (else 0)))
  (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 (config-lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))
    (if (> (+ num-running num-running-in-jobgroup) 0)

    (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))

    (set-signal-handler! signal/int
			 (lambda (signum)
			   (signal-mask! signum)
			   (print "Received signal " signum ", cleaning up before exit. Please wait...")
			   (let ((tdbdat (tasks:open-db)))
			     (rmt:tasks-set-state-given-param-key task-key "killed"))
			   (print "Killed by signal " signum ". Exiting")
			   (exit)))

    ;; register this run in monitor.db
    (rmt:tasks-add "run-tests" user target runname test-patts task-key) ;; params)
    (rmt:tasks-set-state-given-param-key task-key "running")
    (runs:set-megatest-env-vars run-id inkeys: keys inrunname: runname) ;; these may be needed by the launching process
    (if (file-exists? runconfigf)
	(setup-env-defaults runconfigf run-id *already-seen-runconfig-info* keyvals target)
	(debug:print 0 "WARNING: You do not have a run config file: " runconfigf))

    ;; Now generate all the tests lists
    (set! all-tests-registry (tests:get-all))

    (if (not (null? required-tests))
	(debug:print-info 1 "Adding " required-tests " to the run queue"))
    ;; NOTE: these are all parent tests, items are not expanded yet.
    (debug:print-info 4 "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 (current-error-port))
					       (debug:print 0 "ERROR: failure in runs:run-tests-queue thread, error: " ((condition-property-accessor 'exn 'message) exn))
					       (if (> run-queue-retries 0)
						   (begin
						     (set! run-queue-retries (- run-queue-retries 1))
						     (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 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 ()				    
					    ;; (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 "error in calling find-and-mark-incomplete for run-id " run-id)
							       (rmt:find-and-mark-incomplete run-id #f)))) ;; ovr-deadtime)))
							run-ids)))
					  "runs: mark-incompletes")))
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th1)
	    (set! keep-going #f)
	    (thread-join! th2)
	    ;; if run-count > 0 call, set -preclean and -rerun STUCK/DEAD
	    (if (> run-count 0)
		(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"))
		  (runs:run-tests target runname test-patts user flags run-count: (- run-count 1)))))
	  (debug:print-info 0 "No tests to run")))
    (debug:print-info 4 "All done by here")
    (rmt:tasks-set-state-given-param-key task-key "done")
    ;; (sqlite3:finalize! tasks-db)
    ))


;; loop logic. These are used in runs:run-tests-queue to make it a bit more readable.
;;
;; If reg not full and have items in tal then loop with (car tal)(cdr tal) reg reruns

	     (newtal      (append tal (list hed)))
	     (regfull     (>= (length reg) reglen))
	     (num-running (rmt:get-count-tests-running-for-run-id run-id)))

	;; every couple minutes verify the server is there for this run
	(if (and (common:low-noise-print 60 "try start server"  run-id)
		 (tasks:need-server run-id))
	    (tasks:start-and-wait-for-server tdbdat run-id 10)) ;; NOTE: delay and wait is done under the hood
	
	(if (> num-running 0)
	  (set! last-time-some-running (current-seconds)))

      (if (> (current-seconds)(+ last-time-some-running 240))
	  (hash-table-set! *max-tries-hash* tfullname (+ (hash-table-ref/default *max-tries-hash* tfullname 0) 1)))
	;; (debug:print 0 "max-tries-hash: " (hash-table->alist *max-tries-hash*))

;; fields are passing in through 
;; action:
;;    'remove-runs
;;    'set-state-status
;;
;; NB// should pass in keys?
;;
(define (runs:operate-on action target runnamepatt testpatt #!key (state #f)(status #f)(new-state-status #f)(mode 'remove-all)(options '()))
  (common:clear-caches) ;; clear all caches
  (let* ((db           #f)
	 (tdbdat       (tasks:open-db))
	 (keys         (rmt:get-keys))
	 (rundat       (mt:get-runs-by-patt keys runnamepatt target))
	 (header       (vector-ref rundat 0))
	 (runs         (vector-ref rundat 1))

							       (else          'event_time))))))
	 (let* ((run-id    (db:get-value-by-header run header "id"))
		(run-state (db:get-value-by-header run header "state"))
		(run-name  (db:get-value-by-header run header "runname"))
		(tests     (if (not (equal? run-state "locked"))
			       (proc-get-tests run-id)
			       '()))
		(lasttpath "/does/not/exist/I/hope")
		(worker-thread #f))
	   (debug:print-info 4 "runs:operate-on run=" run ", header=" header)
	   (if (not (null? tests))
	       (begin
		 (case action
		   ((remove-runs)
		    (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    ;; seek and kill in flight -runtests with % as testpatt here
		    (if (equal? testpatt "%")
			(tasks:kill-runner target run-name)
			(debug:print 0 "not attempting to kill any run launcher processes as testpatt is " testpatt))
		    (debug:print 1 "Removing tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((set-state-status)
		    (if (tasks:need-server run-id)(tasks:start-and-wait-for-server tdbdat run-id 10))
		    (debug:print 1 "Modifying state and staus for tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((print-run)
		    (debug:print 1 "Printing info for run " runkey ", run=" run ", tests=" tests ", header=" header)
		    action)
		   ((run-wait)
		    (debug:print 1 "Waiting for run " runkey ", run=" runnamepatt " to complete"))
		   ((archive)
		    (debug:print 1 "Archiving data for run: " runkey " " (db:get-value-by-header run header "runname"))
		    (set! worker-thread (make-thread (lambda ()
						       (case (string->symbol (args:get-arg "-archive"))
							 ((save save-remove keep-html)(archive:run-bup (args:get-arg "-archive") run-id run-name tests))
							 ((restore)(archive:bup-restore (args:get-arg "-archive") run-id run-name tests))
							 (else (debug:print 0 "ERROR: unrecognised sub command to -archive. Run \"megatest\" to see help"))))
						     "archive-bup-thread"))
		    (thread-start! worker-thread))
		   (else
		    (debug:print-info 0 "action not recognised " action)))
		 
		 ;; actions that operate on one test at a time can be handled below
		 ;;
		 (let ((sorted-tests     (sort tests (lambda (a b)(let ((dira ;; (rmt:sdb-qry 'getstr 
									 (db:test-get-rundir a)) ;; )  ;; (filedb:get-path *fdb* (db:test-get-rundir a)))
									(dirb ;; (rmt:sdb-qry 'getstr 
									 (db:test-get-rundir b))) ;; ) ;; ((filedb:get-path *fdb* (db:test-get-rundir b))))
								    (if (and (string? dira)(string? dirb))
									(> (string-length dira)(string-length dirb))
									#f)))))

					    (mt:test-set-state-status-by-id run-id (db:test-get-id test) "KILLREQ" "n/a" #f)
						  (thread-sleep! 1)))
					    ;; NOTE: This is suboptimal as the testdata will be used later and the state/status may have changed ...
					    (if (null? tal)
						(loop new-test-dat tal)
						(loop (car tal)(append tal (list new-test-dat)))))
					  (begin
					    (runs:remove-test-directory new-test-dat mode) ;; 'remove-all)
					    (if (not (null? tal))
						(loop (car tal)(cdr tal))))))))
			       ((set-state-status)
				(debug:print-info 2 "new state " (car state-status) ", new status " (cadr state-status))
				(mt:test-set-state-status-by-id run-id (db:test-get-id test) (car state-status)(cadr state-status) #f)
				(if (not (null? tal))
				    (loop (car tal)(cdr tal))))
			       ((run-wait)
				(debug:print-info 2 "still waiting, " (length tests) " tests still running")
				(thread-sleep! 10)
				(let ((new-tests (proc-get-tests run-id)))
				  (if (null? new-tests)
				      (debug:print-info 1 "Run completed according to zero tests matching provided criteria.")
				      (loop (car new-tests)(cdr new-tests)))))
			       ((archive)
				(if (not toplevel-with-children)
				    (begin
				      (debug:print-info 0 "Estimating disk space usage for " test-fulln)
				      (debug:print-info 0 "   " (common:get-disk-space-used (conc run-dir "/")))))
				(if (not (null? tal))
				    (loop (car tal)(cdr tal))))
			       )))
		       )
		     (if worker-thread (thread-join! worker-thread))))))
	   ;; remove the run if zero tests remain
	   (if (eq? action 'remove-runs)
	       (let ((remtests (mt:get-tests-for-run (db:get-value-by-header run header "id") #f '("DELETED") '("n/a") not-in: #t)))
		 (if (null? remtests) ;; no more tests remaining
		     (let* ((dparts  (string-split lasttpath "/"))
			    (runpath (conc "/" (string-intersperse 
						(take dparts (- (length dparts) 1))

		       )))))
	 ))
     runs)
    ;; (sqlite3:finalize! (db:delay-if-busy tdbdat))
    )
  #t)

(define (runs:remove-test-directory test mode) ;; 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)))
    (case mode
      ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "CLEANING" "LOCKED" #f))
      ((remove-all)      (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "REMOVING" "LOCKED" #f))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVE_REMOVING" #f #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)
	  (if (file-exists? real-dir)

		 (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 
    (case mode
      ((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))
      ((archive-remove)  (mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "ARCHIVED" #f #f))
      (else (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 
;; this wrapper is used to reduce the replication of code
(define (general-run-call switchname action-desc proc)
  (let ((runname (or (args:get-arg "-runname")(args:get-arg ":runname")))
	(target  (common:args-get-target)))
    (cond
     ((not target)
      (debug:print 0 "ERROR: Missing required parameter for " switchname ", you must specify the target with -target")
      (exit 3))
     ((not runname)
      (debug:print 0 "ERROR: Missing required parameter for " switchname ", you must specify the run name with -runname runname")
      (exit 3))
     (else
      (let (;; (db   #f)
	    (keys #f))
	(if (launch:setup-for-run)
	    (launch:cache-config)
	    (begin 
	      (debug:print 0 "Failed to setup, exiting")
	      (exit 1)))
	;; (if (args:get-arg "-server")
	;;     (cdb:remote-run server:start db (args:get-arg "-server")))
	(set! keys (keys:config-get-fields *configdat*))
	;; have enough to process -target or -reqtarg here
	(if (args:get-arg "-reqtarg")
	    (let* ((runconfigf (conc  *toppath* "/runconfigs.config")) ;; DO NOT EVALUATE ALL 
		   (runconfig  (read-config runconfigf #f #t environ-patt: #f)))
	      (if (hash-table-ref/default runconfig (args:get-arg "-reqtarg") #f)
		  (keys:target-set-args keys (args:get-arg "-reqtarg") args:arg-hash)
		    
		  (begin
		    (debug:print 0 "ERROR: [" (args:get-arg "-reqtarg") "] not found in " runconfigf)
		    ;; (if db (sqlite3:finalize! db))
		    (exit 1)
		    )))
	    (if (args:get-arg "-target")
		(keys:target-set-args keys (args:get-arg "-target" args:arg-hash) args:arg-hash)))
	(if (not (car *configinfo*))
	    (begin
	      (debug:print 0 "ERROR: Attempted to " action-desc " but run area config file not found")
	      (exit 1))
	    ;; Extract out stuff needed in most or many calls
	    ;; here then call proc
	    (let* ((keyvals    (keys:target->keyval keys target)))
	      (proc target runname keys keyvals)))
	;; (if db (sqlite3:finalize! db))
	(set! *didsomething* #t))))))

;;======================================================================
;; Lock/unlock runs
;;======================================================================

(define (runs:handle-locking target keys runname lock unlock user)