Megatest

(define (set-item-env-vars itemdat)
  (for-each (lambda (item)
	      (debug:print 2 "setenv " (car item) " " (cadr item))
	      (setenv (car item) (cadr item)))
	    itemdat))

(define *last-num-running-tests* 0)

;; Every time can-run-more-tests is called increment the delay
;; if the cou
(define *runs:can-run-more-tests-count* 0)
(define (runs:shrink-can-run-more-tests-count)
  (set! *runs:can-run-more-tests-count* 0)) ;; (/ *runs:can-run-more-tests-count* 2)))

(define (runs:can-run-more-tests test-record 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* ((tconfig                 (tests:testqueue-get-testconfig test-record))
	 (jobgroup                (config-lookup tconfig "requirements" "jobgroup"))




	 (num-running             (cdb:remote-run db:get-count-tests-running #f))
	 (num-running-in-jobgroup (cdb:remote-run db:get-count-tests-running-in-jobgroup #f jobgroup))




	 (job-group-limit         (config-lookup *configdat* "jobgroups" jobgroup)))
    (if (> (+ num-running num-running-in-jobgroup) 0)
	(set! *runs:can-run-more-tests-count* (+ *runs:can-run-more-tests-count* 1)))



    (if (not (eq? *last-num-running-tests* num-running))
	(begin
	  (debug:print 2 "max-concurrent-jobs: " max-concurrent-jobs ", num-running: " num-running)
	  (set! *last-num-running-tests* num-running)))
    (if (not (eq? 0 *globalexitstatus*))
	(list #f num-running num-running-in-jobgroup max-concurrent-jobs job-group-limit)
	(let ((can-not-run-more (cond

;;======================================================================
;; New methodology. These routines will replace the above in time. For
;; now the code is duplicated. This stuff is initially used in the monitor
;; based code.
;;======================================================================
































;; This is a duplicate of run-tests (which has been deprecated). Use this one instead of run tests.
;; keyvals.
;;
;;  test-names: Comma separated patterns same as test-patts but used in selection 
;;              of tests to run. The item portions are not respected.
;;              FIXME: error out if /patt specified
;;            
(define (runs:run-tests target runname test-names test-patts user flags)
  (common:clear-caches) ;; clear all caches
  (let* ((db          #f)
	 (keys        (cdb:remote-run db:get-keys #f))
	 (keyvallst   (keys:target->keyval keys target))
	 (run-id      (cdb:remote-run db:register-run #f keys keyvallst runname "new" "n/a" user))  ;;  test-name)))
	 (keyvals     (if run-id (cdb:remote-run db:get-key-vals #f run-id) #f))
	 (deferred    '()) ;; delay running these since they have a waiton clause
	 ;; keepgoing is the defacto modality now, will add hit-n-run a bit later
	 ;; (keepgoing   (hash-table-ref/default flags "-keepgoing" #f))
	 (runconfigf   (conc  *toppath* "/runconfigs.config"))
	 (required-tests '())
	 (test-records (make-hash-table)))

;; test-records is a hash table testname:item_path => vector < testname testconfig waitons priority items-info ... >
(define (runs:run-tests-queue run-id runname test-records keyvallst flags test-patts)
    ;; At this point the list of parent tests is expanded 
    ;; NB// Should expand items here and then insert into the run queue.
  (debug:print 5 "test-records: " test-records ", keyvallst: " keyvallst " flags: " (hash-table->alist flags))
  (let ((sorted-test-names (tests:sort-by-priority-and-waiton test-records))
	(test-registery    (make-hash-table))
	(registery-mutex   (make-mutex))
	(num-retries        0)
	(max-retries       (config-lookup *configdat* "setup" "maxretries"))
	(max-concurrent-jobs     (let ((mcj (config-lookup *configdat* "setup"     "max_concurrent_jobs")))
				    (if (and mcj (string->number mcj))
					(string->number mcj)
					1))))
    (set! max-retries (if (and max-retries (string->number max-retries))(string->number max-retries) 100))
    (if (not (null? sorted-test-names))
	(let loop ((hed         (car sorted-test-names))
		   (tal         (cdr sorted-test-names))
		   (reruns      '()))
	  (if (not (null? reruns))(debug:print-info 4 "reruns=" reruns))
	  ;; (print "Top of loop, hed=" hed ", tal=" tal " ,reruns=" reruns)

	    (if (member test-name waitons)
		(begin
		  (debug:print 0 "ERROR: test " test-name " has listed itself as a waiton, please correct this!")
		  (set! waiton (filter (lambda (x)(not (equal? x hed))) waitons))))

	    (cond ;; OUTER COND
	     ((not items) ;; when false the test is ok to be handed off to launch (but not before)
	      (let* ((run-limits-info         (runs:can-run-more-tests test-record max-concurrent-jobs)) ;; look at the test jobgroup and tot jobs running
		     (have-resources          (car run-limits-info))
		     (num-running             (list-ref run-limits-info 1))
		     (num-running-in-jobgroup (list-ref run-limits-info 2))
		     (max-concurrent-jobs     (list-ref run-limits-info 3))
		     (job-group-limit         (list-ref run-limits-info 4))
		     (prereqs-not-met         (db:get-prereqs-not-met run-id waitons item-path mode: testmode))
		     (fails                   (runs:calc-fails prereqs-not-met))
		     (non-completed           (runs:calc-not-completed prereqs-not-met)))
		(debug:print-info 8 "have-resources: " have-resources " prereqs-not-met: " 
			     (string-intersperse 
			      (map (lambda (t)
				     (if (vector? t)
					 (conc (db:test-get-state t) "/" (db:test-get-status t))

		 ((not (tests:match test-patts (tests:testqueue-get-testname test-record) item-path)) ;; This test/itempath is not to be run
		  ;; else the run is stuck, temporarily or permanently
		  ;; but should check if it is due to lack of resources vs. prerequisites
		  (debug:print-info 1 "Skipping " (tests:testqueue-get-testname test-record) " " item-path " as it doesn't match " test-patts)
		  ;; (thread-sleep! *global-delta*)
		  (if (not (null? tal))
		      (loop (car tal)(cdr tal) reruns)))
		 ;; Registery has been started for this test but has not yet completed
		 ;; this should be rare, the case where there are only a couple of tests and the db is slow
		 ;; delay a short while and continue
		 ;; ((eq? (hash-table-ref/default test-registery (runs:make-full-test-name test-name item-path) #f) 'start)
		 ;;  (thread-sleep! 0.01)
		 ;;  (loop (car newtal)(cdr newtal) reruns))
		 ;; count number of 'done, if more than 100 then skip on through.
		 (;; (and (< (length (filter (lambda (x)(eq? x 'done))(hash-table-values test-registery))) 100) ;; why get more than 200 ahead?
		  (not (hash-table-ref/default test-registery (runs:make-full-test-name test-name item-path) #f)) ;; ) ;; too many changes required. Implement later.

		  (debug:print-info 4 "Pre-registering test " test-name "/" item-path " to create placeholder" )
		  ;; NEED TO THREADIFY THIS
		  (let ((th (make-thread (lambda ()
		        		   (mutex-lock! registery-mutex)
		        		   (hash-table-set! test-registery (runs:make-full-test-name test-name item-path) 'start)
		        		   (mutex-unlock! registery-mutex)
					   ;; If haven't done it before register a top level test if this is an itemized test
					   (if (not (eq? (hash-table-ref/default test-registery (runs:make-full-test-name test-name "") #f) 'done))
					       (cdb:tests-register-test *runremote* run-id test-name ""))
					   (cdb:tests-register-test *runremote* run-id test-name item-path)
		        		   (mutex-lock! registery-mutex)
					   (hash-table-set! test-registery (runs:make-full-test-name test-name item-path) 'done)
		        		   (mutex-unlock! registery-mutex))
		        		 (conc test-name "/" item-path))))
		    (thread-start! th))
		  ;; TRY (thread-sleep! *global-delta*)
		  (runs:shrink-can-run-more-tests-count)   ;; DELAY TWEAKER (still needed?)
		  (loop (car newtal)(cdr newtal) reruns))
		 ;; At this point *all* test registrations must be completed.
		 ((not (null? (filter (lambda (x)(eq? 'start x))(hash-table-values test-registery))))
		  (debug:print-info 0 "Waiting on test registrations: " (string-intersperse 
									 (filter (lambda (x)
										   (eq? (hash-table-ref/default test-registery x #f) 'start))
										 (hash-table-keys test-registery))
									 ", "))
		  (thread-sleep! 0.1)
		  (loop hed tal reruns))
		 ((not have-resources) ;; simply try again after waiting a second
		  (debug:print-info 1 "no resources to run new tests, waiting ...")
		  ;; Have gone back and forth on this but db starvation is an issue.
		  ;; wait one second before looking again to run jobs.
		  (thread-sleep! 1) ;; (+ 2 *global-delta*))
		  ;; could have done hed tal here but doing car/cdr of newtal to rotate tests
		  (loop (car newtal)(cdr newtal) reruns))
		 ((and have-resources
		       (or (null? prereqs-not-met)
			   (and (eq? testmode 'toplevel)
				(null? non-completed))))
		  (run:test run-id runname keyvallst test-record flags #f)
		  (hash-table-set! test-registery (runs:make-full-test-name test-name item-path) 'running)
		  (runs:shrink-can-run-more-tests-count)  ;; DELAY TWEAKER (still needed?)
		  ;; (thread-sleep! *global-delta*)
		  (if (not (null? tal))
		      (loop (car tal)(cdr tal) reruns)))
		 (else ;; must be we have unmet prerequisites
		    (debug:print 4 "FAILS: " fails)
		    ;; If one or more of the prereqs-not-met are FAIL then we can issue
		    ;; a message and drop hed from the items to be processed.
		    (if (null? fails)
			(begin
			  ;; couldn't run, take a breather
			  (debug:print-info 4 "Shouldn't really get here, race condition? Unable to launch more tests at this moment, killing time ...")
			  ;; (thread-sleep! (+ 0.01 *global-delta*)) ;; long sleep here - no resources, may as well be patient
			  ;; we made new tal by sticking hed at the back of the list
			  (loop (car newtal)(cdr newtal) reruns))
			;; the waiton is FAIL so no point in trying to run hed ever again
			(if (not (null? tal))
			    (if (vector? hed)
				(begin 
				  (debug:print 1 "WARN: Dropping test " (db:test-get-testname hed) "/" (db:test-get-item-path hed)
					       " from the launch list as it has prerequistes that are FAIL")
				  (runs:shrink-can-run-more-tests-count) ;; DELAY TWEAKER (still needed?)
				  ;; (thread-sleep! *global-delta*)
				  (hash-table-set! test-registery (runs:make-full-test-name test-name item-path) 'removed)
				  (loop (car tal)(cdr tal) (cons hed reruns)))
				(begin
				  (debug:print 1 "WARN: Test not processed correctly. Could be a race condition in your test implementation? " hed) ;;  " as it has prerequistes that are FAIL. (NOTE: hed is not a vector)")
				  (runs:shrink-can-run-more-tests-count) ;; DELAY TWEAKER (still needed?)
				  ;; (thread-sleep! (+ 0.01 *global-delta*))
				  (loop hed tal reruns))))))))) ;; END OF INNER COND
	     
	     ;; case where an items came in as a list been processed
	     ((and (list? items)     ;; thus we know our items are already calculated
		   (not   itemdat)) ;; and not yet expanded into the list of things to be done
	      (if (and (debug:debug-mode 1) ;; (>= *verbosity* 1)

		    (debug:print-info 4 "End of items list, looping with next after short delay")
                    ;; (thread-sleep! (+ 0.01 *global-delta*))
		    (loop (car tal)(cdr tal) reruns))))

	     ;; if items is a proc then need to run items:get-items-from-config, get the list and loop 
	     ;;    - but only do that if resources exist to kick off the job
	     ((or (procedure? items)(eq? items 'have-procedure))
	      (let ((can-run-more    (runs:can-run-more-tests test-record max-concurrent-jobs)))
		(if (and (list? can-run-more)
			 (car can-run-more))
		    (let* ((prereqs-not-met (db:get-prereqs-not-met run-id waitons item-path mode: testmode))
			   (fails           (runs:calc-fails prereqs-not-met))
			   (non-completed   (runs:calc-not-completed prereqs-not-met)))
		      (debug:print-info 8 "can-run-more: " can-run-more
				   "\n testname:        " hed
				   "\n prereqs-not-met: " (runs:pretty-string prereqs-not-met)
				   "\n non-completed:   " (runs:pretty-string non-completed) 
				   "\n fails:           " (runs:pretty-string fails)

				  (debug:print 0 "ERROR: The proc from reading the setup did not yield a list - please report this")
				  (exit 1))))))
		       ((null? fails)
			(debug:print-info 4 "fails is null, moving on in the queue but keeping " hed " for now")
			;; only increment num-retries when there are no tests runing
			(if (eq? 0 (list-ref can-run-more 1))
			    (begin
			      ;; TRY (if (> num-retries 100) ;; first 100 retries are low time cost
			      ;; TRY     (thread-sleep! (+ 2 *global-delta*))
			      ;; TRY     (thread-sleep! (+ 0.01 *global-delta*)))
			      (set! num-retries (+ num-retries 1))))
			(if (> num-retries  max-retries)
			    (if (not (null? tal))
				(loop (car tal)(cdr tal) reruns))
			    (loop (car newtal)(cdr newtal) reruns))) ;; an issue with prereqs not yet met?
		       ((and (not (null? fails))(eq? testmode 'normal))
			(debug:print-info 1 "test "  hed " (mode=" testmode ") has failed prerequisite(s); "
				     (string-intersperse (map (lambda (t)(conc (db:test-get-testname t) ":" (db:test-get-state t)"/"(db:test-get-status t))) fails) ", ")
				     ", removing it from to-do list")
			(if (not (null? tal))
			    (begin
                              ;; (thread-sleep! *global-delta*)
			      (loop (car tal)(cdr tal)(cons hed reruns)))))
		       (else
			(debug:print 8 "ERROR: No handler for this condition.")
			;; TRY (thread-sleep! (+ 1 *global-delta*))
			(loop (car newtal)(cdr newtal) reruns)))) ;; END OF IF CAN RUN MORE

		    ;; if can't run more just loop with next possible test
		    (begin
		      (debug:print-info 4 "processing the case with a lambda for items or 'have-procedure. Moving through the queue without dropping " hed)
		      ;; (thread-sleep! (+ 2 *global-delta*))
		      (loop (car newtal)(cdr newtal) reruns))))) ;; END OF (or (procedure? items)(eq? items 'have-procedure))

	    ;;
	    ;; NB// for the above line. I want the test to be registered long before this routine gets called!
	    ;;
	    (set! test-id (open-run-close db:get-test-id db run-id test-name item-path))
	    (if (not test-id)
		(begin
		  (debug:print 2 "WARN: Test not pre-created? test-name=" test-name ", item-path=" item-path ", run-id=" run-id)
		  (cdb:tests-register-test *runremote* run-id test-name item-path)
		  (set! test-id (open-run-close db:get-test-id db run-id test-name item-path))))
	    (debug:print-info 4 "test-id=" test-id ", run-id=" run-id ", test-name=" test-name ", item-path=\"" item-path "\"")
	    (set! testdat (cdb:get-test-info-by-id *runremote* test-id))))
      (set! test-id (db:test-get-id testdat))
      (change-directory test-path)
      (case (if force ;; (args:get-arg "-force")
		'NOT_STARTED

	   (if (not runflag)
	       (if (not parent-test)
		   (debug:print 1 "NOTE: Not starting test " new-test-name " as it is state \"" (test:get-state testdat) 
				"\" and status \"" (test:get-status testdat) "\", use -rerun \"" (test:get-status testdat)
                                "\" or -force to override"))
	       ;; NOTE: No longer be checking prerequisites here! Will never get here unless prereqs are
	       ;;       already met.
	       ;; This would be a great place to do the process-fork
	       (if (not (launch-test #f run-id runname test-conf keyvallst test-name test-path itemdat flags))
		   (begin
		     (print "ERROR: Failed to launch the test. Exiting as soon as possible")
		     (set! *globalexitstatus* 1) ;; 
		     (process-signal (current-process-id) signal/kill))))))
	((KILLED) 
	 (debug:print 1 "NOTE: " new-test-name " is already running or was explictly killed, use -force to launch it."))

	      runs)))
;;======================================================================
;; Rollup runs
;;======================================================================

;; Update the test_meta table for this test
(define (runs:update-test_meta db test-name test-conf)
  (let ((currrecord (cdb:remote-run db:testmeta-get-record db test-name)))
    (if (not currrecord)
	(begin
	  (set! currrecord (make-vector 10 #f))
	  (cdb:remote-run db:testmeta-add-record db test-name)))
    (for-each 
     (lambda (key)
       (let* ((idx (cadr key))
	      (fld (car  key))
	      (val (config-lookup test-conf "test_meta" fld)))
	 ;; (debug:print 5 "idx: " idx " fld: " fld " val: " val)
	 (if (and val (not (equal? (vector-ref currrecord idx) val)))
	     (begin
	       (print "Updating " test-name " " fld " to " val)
	       (cdb:remote-run db:testmeta-update-field db test-name fld val)))))
     '(("author" 2)("owner" 3)("description" 4)("reviewed" 5)("tags" 9)))))

;; Update test_meta for all tests
(define (runs:update-all-test_meta db)
  (let ((test-names (get-all-legal-tests)))
    (for-each 
     (lambda (test-name)
       (let* ((test-path    (conc *toppath* "/tests/" test-name))
	      (test-configf (conc test-path "/testconfig"))
	      (testexists   (and (file-exists? test-configf)(file-read-access? test-configf)))
	      ;; read configs with tricks turned off (i.e. no system)
	      (test-conf    (if testexists (read-config test-configf #f #f)(make-hash-table))))
	 ;; use the open-run-close instead of passing in db
	 (runs:update-test_meta #f test-name test-conf)))
     test-names)))

;; This could probably be refactored into one complex query ...
(define (runs:rollup-run keys keyvallst runname user) ;; was target, now keyvallst
  (debug:print 4 "runs:rollup-run, keys: " keys " keyvallst: " keyvallst " :runname " runname " user: " user)
  (let* ((db              #f) ;; (keyvalllst      (keys:target->keyval keys target))
	 (new-run-id      (cdb:remote-run db:register-run #f keys keyvallst runname "new" "n/a" user))
	 (prev-tests      (open-run-close test:get-matching-previous-test-run-records db new-run-id "%" "%"))
	 (curr-tests      (open-run-close db:get-tests-for-run db new-run-id "%/%" '() '()))
	 (curr-tests-hash (make-hash-table)))
    (open-run-close db:update-run-event_time db new-run-id)
    ;; index the already saved tests by testname and itemdat in curr-tests-hash
    (for-each
     (lambda (testdat)