Megatest

    (let fatal-loop ((count 0)) 
      (handle-exceptions
	  exn
	  (let ((call-chain (get-call-chain))
		(msg        ((condition-property-accessor 'exn 'message) exn)))
	    (if (< count 5)
		(begin ;; this call is colliding, do some crude stuff to fix it.
		  (debug:print 0 *default-log-port* "ERROR: *configdat* was inaccessible! This should never happen. Retry #" count)
		  (debug:print 0 *default-log-port* "ERROR: *configdat* was inaccessible! This should never happen. Retry #" count
			       ", exn=" exn)
		  (launch:setup force-reread: #t)
		  (fatal-loop (+ count 1))) 
		(begin
		  (debug:print 0 *default-log-port* "FATAL: *configdat* was inaccessible! This should never happen. Retried " count " times. Message: " msg)
		  (debug:print 0 *default-log-port* "FATAL: *configdat* was inaccessible! This should never happen. Retried " count
			       " times. Message: " msg)
		  (debug:print 0 *default-log-port* "Call chain:")
		  (with-output-to-port *default-log-port*

                    (lambda ()
                      (print "*configdat* is >>"*configdat*"<<")
                      (pp *configdat*)
                      (pp call-chain)))

	   (args:get-arg "-one-pass"))
      (exit 0))

  (thread-sleep! (cond ;; BB: check with Matt.  Should this sleep move
		       ;; 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)?
		  ;; seems like always)?
		  ((< (- (current-seconds)(runs:dat-beginning-of-time runsdat)) 30) ;; for the first 30 seconds do not throttle in any way
		   0)
        	  ((> (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)
;;		   )))

           (full-log-fname  (conc log-dir "/" log-file)))
      (if run-pre-hook
          (if (null? existing-tests)
              (let* ((use-log-dir (if (not (directory-exists? log-dir))
                                      (handle-exceptions
                                       exn
                                       (begin
                                         (debug:print 0 *default-log-port* "WARNING: Failed to create " log-dir)
                                         (debug:print 0 *default-log-port* "WARNING: Failed to create " log-dir ", exn=" exn)
                                         #f)
                                       (create-directory log-dir #t)
                                       #t)
                                      #t))
                     (start-time   (current-seconds))
                     (actual-logf  (if use-log-dir full-log-fname log-file)))
                (handle-exceptions
                 exn
                 (begin
                   (print-call-chain *default-log-port*)
                   (debug:print 0 *default-log-port* "Message: " ((condition-property-accessor 'exn 'message) exn))
                   (debug:print 0 *default-log-port* "Message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn)
                   (debug:print 0 *default-log-port* "ERROR: failed to run pre-hook " run-pre-hook ", check the log " log-file))
                 (debug:print-info 0 *default-log-port* "running run-pre-hook: \"" run-pre-hook "\", log is " actual-logf)
                 (system (conc run-pre-hook " >> " actual-logf " 2>&1"))
                 (debug:print-info 0 *default-log-port* "pre-hook \"" run-pre-hook "\" took " (- (current-seconds) start-time) " seconds to run.")))
              (debug:print 0 *default-log-port* "Skipping pre-hook call \"" run-pre-hook "\" as there are existing tests for this run.")))))
    
(define (runs:run-post-hook run-id)

      (if run-post-hook
          ;; (if (null? existing-tests)
          ;;    (debug:print 0 *default-log-port* "Skipping post-hook call \"" run-post-hook "\" as there are existing tests for this run.")))))
	  (let* ((use-log-dir (if (not (directory-exists? log-dir))
				  (handle-exceptions
				      exn
				      (begin
					(debug:print 0 *default-log-port* "WARNING: Failed to create " log-dir)
					(debug:print 0 *default-log-port* "WARNING: Failed to create " log-dir ", exn=" exn)
					#f)
				    (create-directory log-dir #t)
				    #t)
				  #t))
		 (start-time   (current-seconds))
		 (actual-logf  (if use-log-dir full-log-fname log-file)))
	    (handle-exceptions
		exn
		(begin
		  (print-call-chain *default-log-port*)
		  (debug:print 0 *default-log-port* "Message: " ((condition-property-accessor 'exn 'message) exn))
		  (debug:print 0 *default-log-port* "Message: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn)
		  (debug:print 0 *default-log-port* "ERROR: failed to run post-hook " run-post-hook ", check the log " log-file))
	      (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)

		 (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)
							       (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

     ;; If no resources are available just kill time and loop again
     ;;
     ((not have-resources) ;; simply try again after waiting a second
      (if (runs:lownoise "no resources" 60)
	  (debug:print-info 1 *default-log-port* "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)
      (thread-sleep! 0.25)
      ;; could have done hed tal here but doing car/cdr of newtal to rotate tests
      (list (car newtal)(cdr newtal) reg reruns))
     
     ;; This is the final stage, everything is in place so launch the test
     ;;
     ((and have-resources
	   (or (null? prereqs-not-met)

      ;; jobtools maxload is useful for where the full Megatest run is done on one machine
      (if maxload ;; only gate if maxload is specified, NOTE: maxload is normalized, i.e. load=1 means all cpus fully utilized
          (common:wait-for-normalized-load maxload "Waiting for load to drop before starting more tests" #f))

      ;; jobtools maxhomehostload is intended to prevent overloading on the homehost which can cause database corruption issues
      (if maxhomehostload
          (common:wait-for-homehost-load maxhomehostload
          (common:wait-for-homehost-load maxhomehostload (conc "Waiting for homehost load to drop below normalized value of " maxhomehostload)))
					 (conc "Waiting for homehost load to drop below normalized value of " maxhomehostload)))
      
      (run:test run-id run-info keyvals runname test-record flags #f test-registry all-tests-registry)
      (runs:incremental-print-results run-id)
      (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'running)
      (runs:shrink-can-run-more-tests-count runsdat)  ;; DELAY TWEAKER (still needed?)
      ;; (thread-sleep! *global-delta*)
      (if (or (not (null? tal))(not (null? reg)))

    ))

;; 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))

(define (runs:pretty-long-list lst)
   (if (> (length lst) 8)(append (take lst 3)(list "...")) lst))

;;======================================================================
;; runs:run-tests-queue is called by runs:run-tests
;;======================================================================
;;
;; 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 keyvals flags test-patts required-tests reglen-in all-tests-registry)
  ;; At this point the list of parent tests is expanded 

			reruns))))
        ;; (loop (car tal)(cdr tal) reg reruns))))

	(runs:incremental-print-results run-id)
	(debug:print 4 *default-log-port* "TOP OF LOOP => "
		     "test-name: " test-name
		     "\n  hed:         " hed
		     "\n  tal:         " tal
		     "\n  tal:         " (runs:pretty-long-list tal)
		     "\n  reg:         " reg
                     "\n  test-record  " test-record
                     "\n  itemdat:     " itemdat
		     "\n  items:       " items
		     "\n  item-path:   " item-path
		     "\n  waitons:     " waitons
		     "\n  num-retries: " num-retries

	 ((not items)
          (debug:print-info 4 *default-log-port* "cond branch - "  "rtq-2")
	  (debug:print-info 4 *default-log-port* "OUTER COND: (not items)")
	  (if (and (not (tests:match test-patts (tests:testqueue-get-testname test-record) item-path required: required-tests))
		   (not (null? tal)))
	      (loop (car tal)(cdr tal) reg reruns))
	  (runs:testdat-prereqs-not-met-set! testdat (rmt:get-prereqs-not-met run-id waitons hed item-path mode: testmode itemmaps: itemmaps))

	  ;; This would be a good place to block on homehost load



	  
	  (runs:dat-can-run-more-tests-set! runsdat (runs:can-run-more-tests runsdat run-id jobgroup max-concurrent-jobs)) ;; look at the test jobgroup and tot jobs running
	  (let ((loop-list (runs:process-expanded-tests runsdat testdat)))
            (if loop-list (apply loop loop-list))))

	 ;; items processed into a list but not came in as a list been processed
	 ;;
	 ((and (list? items)     ;; thus we know our items are already calculated

		     (begin
		       (mt:test-set-state-status-by-id run-id test-id "COMPLETED" "SKIP" skip-test)
		       (debug:print-info 1 *default-log-port* "SKIPPING Test " full-test-name " due to " skip-test))
		     ;;
		     ;; Here the test is handed off to launch.scm for launch-test to complete the launch process
		     ;;
		     (if (not (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat flags))
	 		 (begin
			 (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 *default-log-port* "NOTE: " full-test-name " is already running or was explictly killed, use -force to launch it.")
	 (hash-table-set! test-registry (db:test-make-full-name test-name test-path) 'DONOTRUN)) ;; KILLED))
	((LAUNCHED REMOTEHOSTSTART RUNNING)  

;;
(define (runs:get-hash-by-target target-patts runpatt)
  (let* ((targets (string-split target-patts ","))
	 (keys    (rmt:get-keys))
	 (res-ht  (make-hash-table))) ;; target -> ( runrecord1 runrecord2 ... )
    (for-each
     (lambda (target-patt)
       (let ((runs     (rmt:simple-get-runs runpatt #f #f target-patt)))
       (let ((runs     (rmt:simple-get-runs runpatt #f #f target-patt #f)))
	 (for-each
	  (lambda (run)
	    (let ((target (simple-run-target run)))
	      (hash-table-set! res-ht target (cons run (hash-table-ref/default res-ht target '())))))
	  runs)))
     targets)
    res-ht))

	    (debug:print 0 *default-log-port* "WARNING: directory " real-dir " does not exist")
	    (debug:print 0 *default-log-port* "WARNING: no real directory corrosponding to link " run-dir ", nothing done")))
    (if (symbolic-link? run-dir)
	(begin
	  (debug:print-info 1 *default-log-port* "Removing symlink " run-dir)
	  (handle-exceptions
	   exn
	   (debug:print-error 0 *default-log-port* " Failed to remove symlink " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue")
	   (debug:print-error 0 *default-log-port* " Failed to remove symlink " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue, exn=" exn)
	   (delete-file run-dir)))
	(if (directory? run-dir)
	    (if (> (directory-fold (lambda (f x)(+ 1 x)) 0 run-dir) 0)
		(debug:print 0 *default-log-port* "WARNING: refusing to remove " run-dir " as it is not empty")
		(handle-exceptions
		 exn
		 (debug:print-error 0 *default-log-port* " Failed to remove directory " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue")
		 (debug:print-error 0 *default-log-port* " Failed to remove directory " run-dir ((condition-property-accessor 'exn 'message) exn) ", attempting to continue, exn=" exn)
		 (delete-directory run-dir)))
	    (if (and run-dir
		     (not (member run-dir (list "n/a" "/tmp/badname"))))
		(debug:print 0 *default-log-port* "WARNING: not removing " run-dir " as it either doesn't exist or is not a symlink")
		(debug:print 0 *default-log-port* "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 

						  fail-cnt)))
 	      (if (null? tail)
		    (let* ((final-doc ((sxml-modify `("testsuite" insert-into (@ (name ,xml-ts-name) (tests ,tests-count) (errors ,error-cnt) (failures ,fail-cnt)))) new-doc)))
		    (debug:print 0 *default-log-port* "modify attrib error=" error-cnt " fail= " fail-cnt)
		    (handle-exceptions
		     exn
		     (let*	((msg	((condition-property-accessor 'exn 'message) exn)))
		       (debug:print 0 *default-log-port* (conc "WARNING: Failed to update file" xml-path". Message:" msg)))
		       (debug:print 0 *default-log-port* (conc "WARNING: Failed to update file" xml-path". Message:" msg ", exn=" exn)))
		     		   
		     (if (not (file-exists? xml-dir)) 
			 (create-directory xml-dir #t))
                     (if (not (rmt:no-sync-get/default keyname #f)) 
                       (begin
			 (rmt:no-sync-set  keyname "on")
			 (debug:print 0 *default-log-port* "creating xml at " xml-path)

		(debug:print-info 0 *default-log-port* "No cached megatest or runconfigs files found. None removed.")
		(begin
		  (debug:print-info 0 *default-log-port* "Removing cached files:\n    " (string-intersperse files "\n    "))
		  (for-each 
		   (lambda (f)
		     (handle-exceptions
			 exn
			 (debug:print 0 *default-log-port* "WARNING: Failed to remove file " f)
			 (debug:print 0 *default-log-port* "WARNING: Failed to remove file " f ", exn=" exn)
		       (delete-file f)))
		   files))))
	  (debug:print-error 0 *default-log-port* "-clean-cache requires -runname."))
      (debug:print-error 0 *default-log-port* "-clean-cache requires -target or -reqtarg")))