Megatest

  (last-jobs-check-time    0)
  )

(defstruct runs:testdat
  hed tal reg reruns  test-record
  test-name item-path jobgroup
  waitons testmode  newtal itemmaps prereqs-not-met)

(module runsmod
    (
     runs:wait-if-seen-recently
     )
  
(import scheme chicken data-structures extras files)
(import posix typed-records srfi-18 srfi-69
	  md5 message-digest
	  regex srfi-1)

(define *last-seen-ht* (make-hash-table))

(define (runs:wait-if-seen-recently wait-until . keys)
  (let* ((full-key   (string-intersperse keys "-"))
	 (last-seen  (hash-table-ref/default *last-seen-ht* full-key 0))
	 (now        (current-seconds))
	 (delta      (- now last-seen))
	 (needed     (if (< delta wait-until)
			 0
			 (- wait-until delta))))
    (if (> needed 0)(thread-sleep! needed))
    (hash-table-set! *last-seen-ht* full-key (current-seconds))
    needed))
)

(import runsmod)
    
;; 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

  ;; Take advantage of a good place to exit if running the one-pass methodology
  (if (and (> (runs:dat-can-run-more-tests-count runsdat) 20)
	   (args:get-arg "-one-pass"))
      (exit 0))

  (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))

  (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 (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)))

	      (debug:print-info 0 *default-log-port* "running run-post-hook: \"" run-post-hook "\", log is " actual-logf)
	      (system (conc run-post-hook " >> " actual-logf " 2>&1"))
	      (debug:print-info 0 *default-log-port* "post-hook \"" run-post-hook "\" took " (- (current-seconds) start-time) " seconds to run."))))))

;; return #t when all items in waitors-upon list are represented in test-patt, #f otherwise.
(define (runs:testpatts-mention-waitors-upon? test-patt waitors-upon)
  (null? (tests:filter-test-names-not-matched waitors-upon test-patt)))

(define *find-and-mark-incomplete-last-run* (make-hash-table))

;;======================================================================
;; runs:run-tests is called from megatest.scm and itself
;;======================================================================
;;
;;  test-names: Comma separated patterns same as test-patts but used in selection 
;;              of tests to run. The item portions are not respected.

    ;; run the run prehook if there are no tests yet run for this run:
    ;;
    (runs:run-pre-hook run-id)
    ;; mark all test launced flag as false in the meta table 
    (rmt:set-var (conc "lunch-complete-" run-id) "no")
    (debug:print-info 1 *default-log-port* "Setting end-of-run to no")
    (let* ((config-reruns      (let ((x (configf:lookup *configdat* "setup" "reruns")))
				 (if x (string->number x) #f)))
	   (config-rerun-cnt (if config-reruns
				 config-reruns
				 1)))
      (if (eq? config-rerun-cnt run-count)
	  (rmt:set-var (conc "end-of-run-" run-id) "no")))
    
    (rmt:set-run-state-status run-id "new" "n/a")
    ;; now add non-directly referenced dependencies (i.e. waiton)
    ;;======================================================================
    ;; refactoring this block into tests:get-full-data
    ;;
    ;; What happended, this code is now duplicated in 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)









		 (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)
								;; lets run this only if a run has been NOT seen for more than 900 seconds
								(if (> (- (current-seconds)(hash-table-ref/default *find-and-mark-incomplete-last-run* run-id 0)) 900)
								    (begin
								      (rmt:find-and-mark-incomplete run-id #f)
								      (hash-table-set! *find-and-mark-incomplete-last-run* run-id (current-seconds)))
								    )))) ;; 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! th2)


	    (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" "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 ))   

         (num-retries           0)
         (max-retries           (configf:lookup *configdat* "setup" "maxretries"))
         (max-concurrent-jobs   (configf:lookup-number *configdat* "setup" "max_concurrent_jobs" default: 50))
         (reglen                (if (number? reglen-in) reglen-in 1))
         (last-time-incomplete  (- (current-seconds) 900)) ;; force at least one clean up cycle
         (last-time-some-running (current-seconds))
         ;; (tdbdat                (tasks:open-db))
	 (misc-data             (make-hash-table)) ;; use as needed
         (runsdat (make-runs:dat
                   ;; hed: hed
                   ;; tal: tal
                   ;; reg: reg
                   ;; reruns: reruns
                   reglen: reglen
                   regfull: #f ;; regfull

      ;;
      (if (> (current-seconds)(+ last-time-incomplete 900))
          (begin
            (set! last-time-incomplete (current-seconds))
            ;; (rmt:find-and-mark-incomplete-all-runs)
	    ))

      ;; WAIT FOR TIME ON TIGHT LOOP
      (if (< (- (current-milliseconds)(hash-table-ref/default misc-data "tight-loop-last-time" 0))
	     100) ;; less than 1/100 second since came through the loop
	  (thread-sleep! 0.1)) ;; wait a 1/100 seconds
      (hash-table-set! misc-data "tight-loop-last-time" (current-milliseconds))
      
      ;; (print "Top of loop, hed=" hed ", tal=" tal " ,reruns=" reruns)
      (let* ((test-record (hash-table-ref test-records hed))
	     (test-name   (tests:testqueue-get-testname test-record))
	     (tconfig     (tests:testqueue-get-testconfig test-record))
	     (jobgroup    (configf:lookup tconfig "test_meta" "jobgroup"))
	     (testmode    (let ((m (configf:lookup tconfig "requirements" "mode")))
			    (if m (map string->symbol (string-split m)) '(normal))))

				       ))
				   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))
	     (testdat     (make-runs:testdat
			   hed: hed
			   tal: tal
			   reg: reg
			   reruns: reruns
			   test-record: test-record
			   test-name:   test-name

				  ;; wait for load here
				  (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
				  (loop-can-run-more (runs:can-run-more-tests runsdat run-id jobgroup max-concurrent-jobs)
						     (- remtries 1)))))))
		       )))))

	  ;; I'm not clear on why prereqs are gathered here TODO: verfiy this is needed
	  (let ((waited (runs:wait-if-seen-recently 5 "prereqs-not-met" hed item-path))) ;; if we've been down this path in the past 5 seconds - wait out the difference
	    (if (> waited 0)(debug:print 0 *default-log-port* "Waited for prereqs-not-met-"hed"-"item-path" for " waited "seconds.")))
	  
	  (runs:testdat-prereqs-not-met-set! testdat (rmt:get-prereqs-not-met run-id waitons hed item-path mode: testmode itemmaps: itemmaps))

	  ;; I'm not clear on why we'd capture running job counts here TODO: verify this is needed
	  (runs:dat-can-run-more-tests-set! runsdat (runs:can-run-more-tests runsdat run-id jobgroup max-concurrent-jobs))

	  (let ((loop-list (runs:process-expanded-tests runsdat testdat))) ;; in process-expanded-tests ultimately run:test -> launch-test -> test actually running
            (if loop-list (apply loop loop-list))))

    (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)
    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
    (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))
		    (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)(hash-table-ref/default *find-and-mark-incomplete-last-run* run-id 0)) 900)
		;; (begin(if (> (current-seconds)(+ last-time-incomplete 900))
		(let ((actual-num-running num-running)) ;; (rmt:get-count-tests-running-for-run-id run-id))) ;; why call it again?
		  (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)
		  (hash-table-set! *find-and-mark-incomplete-last-run* run-id (current-seconds))
		  (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))))))
	    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
	    (thread-sleep! 5) ;; (if (>= num-running max-concurrent-jobs) 5 1))
	    (wait-loop (rmt:get-count-tests-running-for-run-id run-id)
		       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")))

    (path-out (conc "/" (string-intersperse (take dparts (- (length dparts) 1)) "/")))
    )
    path-out
  )
)





















(define (runs:operate-on action target runnamepatt testpatt #!key (state #f)(status #f)(new-state-status #f)(mode #f)(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))

                                            (loop new-test-dat tal)
                                            (loop (car tal)(append tal (list new-test-dat)))))
                                      (begin
                                       (let ((rundir (db:test-get-rundir new-test-dat)))
                                        (if (and (not (string=  rundir "/tmp/badname")) 
                                             (file-exists? rundir)
                                             (substring-index run-name rundir)
                                             (tests:glob-like-match (conc "%/" target "/%") rundir)
                                             )
                                          (begin
                                            (set! lasttpath (db:test-get-rundir new-test-dat)) ;; remember this path for run removal
                                            (set! lastrealpath (remove-last-path-directory (resolve-pathname lasttpath)))
                                            (hash-table-set! run-paths-hash lastrealpath 1)
                                            (runs:remove-test-directory new-test-dat mode) ;; 'remove-all)
                                          )
                                          (begin
                                            (debug:print 2 *default-log-port* "Not removing directory " rundir " because either it doesn't exist or has a bad name")
                                            (debug:print 2 *default-log-port* "Is /tmp/badname: " (string=  rundir "/tmp/badname"))
                                            (debug:print 2 *default-log-port* "Exists: " (file-exists? rundir))
                                            (debug:print 2 *default-log-port* "Has run-name: " (substring-index run-name rundir))
                                            (debug:print 2 *default-log-port* "Has target: " (tests:glob-like-match (conc "%/" target "/%") rundir))
                                            (debug:print 2 *default-log-port* "Target: " target)

                                            ;;PJH remove record from db no need to cleanup directory
                                            (case mode
                                               ((remove-data-only)(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) (db:test-get-state test)(db:test-get-status test) #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))))

                                          )