Megatest

(defstruct runs:dat
  reglen regfull
  runname max-concurrent-jobs run-id
  test-patts required-tests test-registry
  registry-mutex flags keyvals run-info all-tests-registry
  can-run-more-tests
  ((can-run-more-tests-count 0) : fixnum)
  (last-fuel-check         0)  ;; time when we last checked fuel
  (beginning-of-time       (current-seconds))
  )

(defstruct runs:testdat
  hed tal reg reruns  test-record
  test-name item-path jobgroup
  waitons testmode  newtal itemmaps prereqs-not-met)
  
;; look in the $MT_RUN_AREA_HOME/.softlocks directory for key-host-pid.softlock files
;;  - remove any that are over 3600 seconds old
;;  - if there are any that are younger than 10 seconds
;;      * sleep 10 seconds
;;      * touch my key-host-pid.softlock file
;;      * return
;;  - if there are no files younger than 10 seconds
;;     * touch my key-host-pid.softlock file
;;     * return
;;  
(define (runs:wait-on-softlock rdat key)
  (if (not (and *toppath* (file-exists? *toppath*))) ;; don't seem to have toppath yet
      (debug:print-info 0 *default-log-port* "Can't create softlocks - don't see MTRAH yet.")
      (let* ((softlocks-dir (conc *toppath* "/.softlocks")))
	(if (not (file-exists? softlocks-dir))
	    (create-directory softlocks-dir #t))
	(let* ((my-lock-file   (conc softlocks-dir "/" key "-" (get-host-name) "-" (current-process-id) ".softlock"))
	       (lock-files     (filter (lambda (x)
					 (not (equal? x my-lock-file)))
				       (glob (conc softlocks-dir "/" key "*.softlock"))))
	       (fresh-locks    (any (lambda (x) ;; do we have any locks younger than 10 seconds
				      (let* ((mod-time (file-modification-time x))
					     (age      (- (current-seconds) mod-time)))
					(cond
					 ((> age 3600) ;; too old to keep, remove it
					  (delete-file* x) #f)
					 ((< age 10)  #t)
					 (else #f))))
				    lock-files)))
	  (if fresh-locks
	      (begin
		(if (runs:lownoise "runners-softlock-wait" 360)
		    (debug:print-info 0 *default-log-port* "Other runners in flight, giving up some time..."))
		(thread-sleep! 10))
	      (begin
		(if (runs:lownoise "runners-softlock-nowait" 360)
		    (debug:print-info 0 *default-log-port* "No runners in flight, updating softlock"))
		(let* ((ouf (open-output-file my-lock-file)))
		  (with-output-to-port ouf
		    (lambda ()(print (current-seconds))))
		  (close-output-port ouf))))
	  (runs:dat-last-fuel-check-set! rdat (current-seconds))))))
  
;; Fourth try, do accounting through time
;;
(define (runs:parallel-runners-mgmt rdat)
  (let ((time-to-check (configf:lookup-number *configdat* "runners" "time-to-check" default: 10)) ;; 28
	(time-to-wait  (configf:lookup-number *configdat* "runners" "time-to-wait" default: 30))
	(now-time      (current-seconds)))
    (if (> (- now-time (runs:dat-last-fuel-check rdat)) time-to-check) ;; time to check
	(runs:wait-on-softlock rdat "runners"))))

;; To test parallel-runners management start a repl:
;;  megatest -repl
;; then run:
;;  (runs:test-parallel-runners 60)
;;
(define (runs:test-parallel-runners duration #!optional (proc #f))
  (let* ((rdat   (make-runs:dat))
	 (rtime  0)
	 (startt (current-seconds))
	 (endt   (+ startt duration)))
    ((or proc runs:parallel-runners-mgmt) rdat)
    (let loop ()
      (let* ((wstart (current-seconds)))
	(if (< wstart endt)
	    (let* ((work-time (random 10)))
	      #;(debug:print-info 0 *default-log-port* "working for " work-time
				" seconds. Total work: " rtime ", elapsed time: " (- wstart startt))
	      (thread-sleep! work-time)
	      (set! rtime (+ rtime work-time))
	      ((or proc runs:parallel-runners-mgmt) rdat)
	      (loop)))))
    (let* ((done-time (current-seconds)))
      (debug:print-info 0 *default-log-port* "DONE: rtime=" rtime ", elapsed time=" (- done-time startt)
			", ratio=" (/ rtime (- done-time startt))))))

(define (runs:get-mt-env-alist run-id runname target testname itempath)
  ;;(bb-check-path msg: "runs:set-megatest-env-vars entry")
  `(("MT_TEST_NAME"     . ,testname)
    
    ("MT_ITEMPATH"      . ,itempath)

         (cons  (car key) (cadr key)))
       (keys:target->keyval (rmt:get-keys) target))
    
    ,@(map (lambda (var)
             (let ((val (configf:lookup *configdat* "env-override" var)))
               (cons var val)))
           (configf:section-vars *configdat* "env-override"))))





    
;; set up needed environment variables given a run-id and optionally a target, itempath etc.
;;
(define (runs:set-megatest-env-vars run-id #!key (inkeys #f)(inrunname #f)(inkeyvals #f)(intarget #f)(testname #f)(itempath #f))
  ;;(bb-check-path msg: "runs:set-megatest-env-vars entry")
  (let* ((target    (or intarget 
			(common:args-get-target)

		       ;; to cond clauses below where we determine we
		       ;; have too many jobs running rather than each
		       ;; time the and condition above is true (which
		       ;; seems like always)?
        	  ((> (runs:dat-can-run-more-tests-count runsdat) 20) ;; original intent was - save cycles, wait a long time
		   (if (runs:lownoise "waiting on tasks" 60)(debug:print-info 2 *default-log-port* "waiting for tasks to complete, sleeping briefly ..."))
		   10)  ;; obviously haven't had any work to do for a while
		  (else 0)))
;;		   ;; if have a number for inter-test-delay, use it, else don't delay much, maybe even zero?
;;		   (configf:lookup-number *configdat* "setup" "inter-test-delay" default: 0.01)
;;		   )))
  
  (let* ((num-running             (rmt:get-count-tests-running run-id #f)) ;; fastmode=no
	 (num-running-in-jobgroup (rmt:get-count-tests-running-in-jobgroup run-id jobgroup))
	 (job-group-limit         (let ((jobg-count (configf:lookup *configdat* "jobgroups" jobgroup)))
				    (if (string? jobg-count)
					(string->number jobg-count)
					jobg-count))))
    (if (> (+ num-running num-running-in-jobgroup) 0)
	(runs:inc-can-run-more-tests-count runsdat)) ;; (set! *runs:can-run-more-tests-count* (+ *runs:can-run-more-tests-count* 1)))

   run-info: #f
   runname: #f
   target: #f
   )
  )

(define (runs:incremental-print-results run-id)
  (let ((curr-sec    (current-seconds))
	(last-update (runs:gendat-inc-results-last-update *runs:general-data*)))
    (if (> (- curr-sec last-update) 5) ;; at least five seconds since last update
	(let* ((run-dat  (or (runs:gendat-run-info *runs:general-data*)(rmt:get-run-info run-id)))
	       (runname  (or (runs:gendat-runname *runs:general-data*)
			     (db:get-value-by-header (db:get-rows run-dat)
						     (db:get-header run-dat) "runname")))
	       (target   (or (runs:gendat-target *runs:general-data*)(rmt:get-target run-id)))
	       (testsdat (let ((res (rmt:get-tests-for-run
				     run-id "%" '() '() ;; run-id testpatt states statuses
				     #f #f ;; offset limit
				     #f ;; not-in
				     #f ;; sort-by
				     #f ;; sort-order
				     #f ;; get full data (not 'shortlist)
				     last-update
				     'dashboard)))
			   (if (list? res)
			       res
			       (begin
				 (debug:print-error
				  0 *default-log-port*
				  "FAILED TO GET DATA using rmt:get-tests-for-run. Notify developers if you see this. result: " res)
				 '())))))
	  (runs:gendat-inc-results-last-update-set! *runs:general-data* (- curr-sec 1))
	  (if (not (runs:gendat-run-info *runs:general-data*))
	      (runs:gendat-run-info-set! *runs:general-data* run-dat))
	  (if (not (runs:gendat-runname  *runs:general-data*))
	      (runs:gendat-runname-set! *runs:general-data* runname))
	  (if (not (runs:gendat-target *runs:general-data*))
	      (runs:gendat-target-set! *runs:general-data* target))
	  (for-each

			     state
			     status
			     dtime
			     (seconds->hr-min-sec duration)
			     (conc "lt/" target "/" runname "/" test-name (if (string-null? item-path) "" (conc "/" item-path))))
		     (hash-table-set! (runs:gendat-inc-results *runs:general-data*) (conc run-id "," test-id) testdat)))))
	   testsdat)))

    ;; I don't think this should be here? -- Matt
    #;(runs:gendat-inc-results-last-update-set! *runs:general-data* (- curr-sec 10))

    ))

;; every time though the loop increment the test/itempatt val.
;; when the min is > max-allowed and none running then force exit
;;
(define *max-tries-hash* (make-hash-table))

;;======================================================================

				       ))
				   extras)
				  extras)
				'())))
	     (waitons     (delete-duplicates (append (tests:testqueue-get-waitons test-record) extra-waits) equal?))
	     (newtal      (append tal (list hed)))
	     (regfull     (>= (length reg) reglen))
	     (num-running (rmt:get-count-tests-running-for-run-id run-id #t)) ;; fastmode=yes
	     (testdat     (make-runs:testdat
			   hed: hed
			   tal: tal
			   reg: reg
			   reruns: reruns
			   test-record: test-record
			   test-name:   test-name

		     "\n  waitons:     " waitons
		     "\n  num-retries: " num-retries
                     "\n  reruns:      " reruns
		     "\n  regfull:     " regfull
		     "\n  reglen:      " reglen
		     "\n  length reg:  " (length reg)
                     )

	(runs:parallel-runners-mgmt runsdat)

	;; check for hed in waitons => this would be circular, remove it and issue an
	;; error
	(if (member test-name waitons)
	    (begin
	      (debug:print-error 0 *default-log-port* "test " test-name " has listed itself as a waiton, please correct this!")
	      (set! waiton (filter (lambda (x)(not (equal? x hed))) waitons))))

	 ))) ;; end loop on sorted test names
    ;; this is the point where everything is launched and now you can mark the run in metadata table as all launched 
    (rmt:set-var (conc "lunch-complete-" run-id) "yes")  
        
    ;; now *if* -run-wait we wait for all tests to be done
    ;; Now wait for any RUNNING tests to complete (if in run-wait mode)
    (thread-sleep! 10) ;; I think there is a race condition here. Let states/statuses settle
    (let wait-loop ((num-running      (rmt:get-count-tests-running-for-run-id run-id #t)) ;; fastmode=yes
		    (prev-num-running 0))
      ;; (debug:print-info 13 *default-log-port* "num-running=" num-running ", prev-num-running=" prev-num-running)
      (if (and (or (args:get-arg "-run-wait")
		   (equal? (configf:lookup *configdat* "setup" "run-wait") "yes"))
	       (> num-running 0))
	  (begin
	    ;; Here we mark any old defunct tests as incomplete. Do this every fifteen minutes
	    ;; (debug:print 0 *default-log-port* "Got here eh! num-running=" num-running " (> num-running 0) " (> num-running 0))
	    (if (> (current-seconds)(+ last-time-incomplete 900))
		(let ((actual-num-running (rmt:get-count-tests-running-for-run-id run-id #f))) ;; fastmode=no
		  (debug:print-info 0 *default-log-port* "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)) ;; FIXME, this might be causing slow down - use of set!
		  (rmt:find-and-mark-incomplete run-id #f)

		  (debug:print-info 0 *default-log-port* "run-wait specified, waiting on " actual-num-running
				    " tests in RUNNING, REMOTEHOSTSTART or LAUNCHED state at "
				    (time->string (seconds->local-time (current-seconds))))))
	    (thread-sleep! 5)
	    (wait-loop (rmt:get-count-tests-running-for-run-id run-id #t) ;; fastmode=yes
		       num-running))))
    ;; LET* ((test-record
    ;; we get here on "drop through". All done!
    ;; this is moved to runs:run-testes since this function is getting called twice to ensure everthing is completed. 
    ;; (debug:print-info 0 *default-log-port* "Calling Post Hook")    
    ;; (runs:run-post-hook run-id)
    (debug:print-info 1 *default-log-port* "All tests launched")))

                   )
		       )))))
	 ))
     runs)
    ;; special case - archive get
    (if (equal? (args:get-arg "-archive") "get")
	(archive:bup-get-data "get" #f #f test-records rp-mutex bup-mutex))
    (if (or (equal? (args:get-arg "-archive") "save") (equal? (args:get-arg "-archive") "save-remove"))
        (begin 
             (print "db archive started")  
             (archive:megatest-db target runnamepatt)
             (print "db archived")))
    )
  #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 (common:file-exists? run-dir)