Megatest

;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;  strftime('%m/%d/%Y %H:%M:%S','now','localtime')





(declare (unit runs))
(declare (uses db))
(declare (uses common))
(declare (uses debugprint))
(declare (uses commonmod))
(declare (uses items))
(declare (uses runconfig))
(declare (uses tests))
(declare (uses server))
(declare (uses mt))
(declare (uses archive))
(declare (uses mtargs))
(declare (uses rmtmod))
(declare (uses dbfile))
(declare (uses tcp-transportmod))

(use (prefix sqlite3 sqlite3:) srfi-1 posix regex regex-case srfi-69 (srfi 18) 
     posix-extras directory-utils pathname-expand typed-records format  sxml-serializer
     sxml-modifications matchable)



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

;; (include "debugger.scm")

(import commonmod
	debugprint
	rmtmod
	dbfile
	tcp-transportmod
	(prefix mtargs args:))

;; use this struct to facilitate refactoring
;;

(defstruct runs:dat
  reglen regfull
  runname max-concurrent-jobs run-id
  test-patts required-tests test-registry

  (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)
  



























;; 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

		(if (runs:lownoise "runners-softlock-wait" 360)
		    (debug:print-info 0 *default-log-port* "Other runners in flight, giving up some time..."))
		(thread-sleep! 2))
	      (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"))))

		  (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* "Call chain:")
		  (with-output-to-port *default-log-port*

		    (lambda ()
		      (print "*configdat* is >>"*configdat*"<<")
                      (pp *configdat*)
                      (pp call-chain)))
                  
		  (exit 1))))
          ;;(bb-check-path msg: "runs:set-megatest-env-vars block 1.5")
          (when (or (not *configdat*) (not (hash-table? *configdat*)))
              (debug:print 0 *default-log-port* "WARNING: *configdat* was inaccessible! This should never happen.  Brute force reread.")

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

(define *last-test-launch* 0)
(define *too-soon-delays* (make-hash-table))

;; to-soon delay, when matching event happened in less than dseconds delay wseconds
;;
(define (runs:too-soon-delay key dseconds wseconds)
  (let* ((last-time (hash-table-ref/default *too-soon-delays* key #f)))
    (if (and last-time
	     (< (- (current-seconds) last-time) dseconds))
	(begin
	  (if (runs:lownoise (conc "too-soon-delay"key) 60)
	      (debug:print-info 2 *default-log-port* "Polling throttle for "key))
	  (thread-sleep! wseconds)))
    (hash-table-set! *too-soon-delays* key (current-seconds))))

(define (runs:can-run-more-tests runsdat run-id jobgroup max-concurrent-jobs)

  ;; 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-for-run-id 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)))
    (if (not (eq? *last-num-running-tests* num-running))
	(begin
	  (debug:print 2 *default-log-port* "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)

;; 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)))



;;======================================================================
;; 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.
;;              FIXME: error out if /patt specified
;;            
;; run-count is passed from megatest.scm as configf:lookup *configdat* "setup" "reruns", or defaults to 1.
(define (runs:run-tests target runname test-patts user flags #!key (run-count 1)) ;; test-names
  (let* ((keys               (keys:config-get-fields *configdat*))
	 (keyvals            (keys:target->keyval keys target))
	 (run-id             (rmt:register-run keyvals runname "new" "n/a" user (args:get-arg "-contour")))  ;;  test-name)))
	 ;; (deferred          '()) ;; delay running these since they have a waiton clause
	 (runconfigf         (conc  *toppath* "/runconfigs.config"))
         (mtconfig           (conc  *toppath* "/megatest.config"))
         (readonly-mode      (not (file-write-access? mtconfig)))
	 (test-records       (make-hash-table))
	 ;; need to process runconfigs before generating these lists
	 (all-tests-registry #f)  ;; (tests:get-all)) ;; (tests:get-valid-tests (make-hash-table) test-search-path)) ;; all valid tests to check waiton names
	 (all-test-names     #f)  ;; (hash-table-keys all-tests-registry))
	 (test-names         #f)  ;; Generated by a call to (tests:filter-test-names all-test-names test-patts))
	 (required-tests     #f)  ;; Put fully qualified test/testpath names in this list to be done
         (waitors-upon       (make-hash-table)) ;; given a test, return list of tests waiting upon this test.
	 (task-key           (conc (hash-table->alist flags) " " (get-host-name) " " (current-process-id)))
	 ;; (tdbdat             (tasks:open-db))
	 (config-reruns      (let ((x (configf:lookup *configdat* "setup" "reruns")))
			       (if x (string->number x) #f)))
	 (allowed-tests      #f)
	 (runconf            #f))



    ;; check if readonly
    (when readonly-mode
      (debug:print-error 0 *default-log-port* "Megatest database is readonly.  Cannot proceed.")
      (exit 1))

    ;; per user request. If less than 100Meg space on dbdir partition, bail out with error
    ;; this will reduce issues in database corruption
    (common:check-db-dir-and-exit-if-insufficient)

    ;; override the number of reruns from the configs
    ;; this needs to be done at the place where is first runs:run-tests called 
    ;(if (and config-reruns
	  ;   (> run-count config-reruns))
 	  ;(set! run-count config-reruns))

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

    (let ((sighand (lambda (signum)
		     ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
		     (set! *time-to-exit* #t)
		     (debug:print 0 *default-log-port* "Received signal " signum ", cleaning up before exit. Please wait...")
		     (let ((th1 (make-thread (lambda ()
					       ;; (let ((tdbdat (tasks:open-db)))
						 (rmt:tasks-set-state-given-param-key task-key "killed") ;; )
					       (debug:print 0 *default-log-port* "Killed by signal " signum ". Exiting")
					       (thread-sleep! 3)
					       (exit))))
			   (th2 (make-thread (lambda ()
					       (thread-sleep! 5)
					       (debug:print 0 *default-log-port* "Done")
					       (exit 4)))))
		       (thread-start! th2)

    ;; (set! required-tests     (lset-intersection equal? test-names all-test-names))
    
    ;; look up all tests matching the comma separated list of globs in
    ;; test-patts (using % as wildcard)

    ;; (set! test-names (delete-duplicates (tests:get-valid-tests *toppath* test-patts)))
    (debug:print-info 0 *default-log-port* "tests search path: " (string-intersperse (tests:get-tests-search-path *configdat*) " "))
    (debug:print-info 2 *default-log-port* "all tests:         " (string-intersperse (sort all-test-names string<) " "))
    (debug:print-info 0 *default-log-port* "test names:        " (string-intersperse (sort test-names string<) " "))
    (debug:print-info 0 *default-log-port* "required tests:    " (string-intersperse (sort required-tests string<) " "))

    ;; on the first pass or call to run-tests set FAILS to NOT_STARTED if
    ;; -keepgoing is specified
    (if (eq? *passnum* 0)
	(begin

    ;; run the run prehook if there are no tests yet run for this run:
    ;;
    (runs:run-pre-hook run-id)
    ;; mark all test launched 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!?
    ;;
    ;;======================================================================
    
    (if (not (null? test-names)) ;; BEGIN test-names loop
	(let loop ((hed (car test-names))   ;; NOTE: This is the main loop that iterates over the test-names
		   (tal (cdr test-names)))         ;; 'return-procs tells the config reader to prep running system but return a proc
          (debug:print-info 4 *default-log-port* "\n\ntestpatt elaboration loop => hed="hed " tal="tal" test-patts="test-patts" test-names="test-names)
	  (change-directory *toppath*) ;; PLEASE OPTIMIZE ME!!! I think this should be a no-op but there are several places where change-directories could be happening.
	  (setenv "MT_TEST_NAME" hed) ;; 
	  (let*-values (((waitons waitors config) (tests:get-waitons hed all-tests-registry (tests:get-global-waitons *runconfigdat*)))

			;; NOTE: Have the config - can extract [waitons] section
			
                        ((hed-mode)
                         (let ((m (configf:lookup config "requirements" "mode")))
                           (if m (map string->symbol (string-split m)) '(normal))))
                        ((hed-itemized-waiton) ;; are items in hed waiting on items of waiton?

                          (waiton-record   (hash-table-ref/default test-records waiton #f))
			  (waiton-tconfig  (if waiton-record (vector-ref waiton-record 1) #f))
			  (waiton-itemized (and waiton-tconfig
						(or (hash-table-ref/default waiton-tconfig "items" #f)
						    (hash-table-ref/default waiton-tconfig "itemstable" #f))))
			  (itemmaps        (tests:get-itemmaps config))  ;; (configf:lookup config "requirements" "itemmap"))
			  (new-test-patts  (tests:extend-test-patts test-patts hed waiton itemmaps hed-itemized-waiton))) 
		     (debug:print-info 2 *default-log-port* "Test " waiton " has " (if waiton-record "a" "no") " waiton-record and" (if waiton-itemized " " " no ") "items")
		     ;; need to account for test-patt here, if I am test "a", selected with a test-patt of "hed/b%"
		     ;; and we are waiting on "waiton" we need to add "waiton/,waiton/b%" to test-patt
		     ;; is this satisfied by merely appending "/" to the waiton name added to the list?
		     ;;
		     ;; This approach causes all of the items in an upstream test to be run 
		     ;; if we have this waiton already processed once we can analzye it for extending
		     ;; tests to be run, since we can't properly process waitons unless they have been

                                   (set! test-names (cons waiton test-names)) ;; need to process this one, only add once the waiton tconfig read
                                   (set! required-tests (cons (conc waiton "/") required-tests))
                                   (set! test-patts new-test-patts))
                                 (begin
                                   (debug:print-info 0 *default-log-port* "Waitor(s) not yet on testpatt for " waiton ", setting up to re-process it")
                                   (set! tal (append (cons waiton tal)(list hed)))))
                             (begin
                               (debug:print-info 2 *default-log-port* "Adding non-itemized test " waiton " to required-tests")
                               (set! required-tests (cons waiton required-tests))
                               (set! test-patts new-test-patts)))
			 (begin
			   (debug:print-info 0 *default-log-port* "No testconfig info yet for " waiton ", setting up to re-process it")
			   (set! tal (append (cons waiton tal)(list hed))))) ;; (cons (conc waiton "/") required-tests))
		     ;; NOPE: didn't work. required needs to be plain test names. Try tacking on to test-patts
		     ;;  - doesn't work

    (if (not (null? required-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* ()


















	    (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests
				  (any->number reglen) all-tests-registry)








	    (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" "ABORT,STUCK/DEAD,n/a,ZERO_ITEMS"))

                  (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 ))   

;; return value of runs:expand-items is passed back to runs-tests-queue and is fed to named loop with this signature:
;;    (let loop ((hed         (car sorted-test-names))
;;	         (tal         (cdr sorted-test-names))
;;	         (reg         '()) ;; registered, put these at the head of tal 
;;	         (reruns      '()))
(define (runs:expand-items hed tal reg reruns regfull newtal jobgroup max-concurrent-jobs run-id waitons item-path testmode test-record can-run-more items runname tconfig reglen test-registry test-records itemmaps)
  (let* ((loop-list       (list hed tal reg reruns))
         (junk (debug:print-info 4 *default-log-port* "expand-items calling rmt:get-prereqs-not-met"))
	 (prereqs-not-met (let ((res (rmt:get-prereqs-not-met run-id waitons hed item-path mode: testmode itemmaps: itemmaps)))
			    (if (list? res)
				res
				(begin
				  (debug:print 0 *default-log-port*
					       "ERROR: rmt:get-prereqs-not-met returned non-list!\n"
					       "  res=" res " run-id=" run-id " waitons=" waitons " hed=" hed " item-path=" item-path " testmode=" testmode " itemmaps=" itemmaps)

	  (if (list? items-list)
	      (begin
		(if (null? items-list)
		    (let ((test-id   (rmt:get-test-id run-id test-name ""))
			  (num-items (rmt:test-toplevel-num-items run-id test-name)))
		      (if (and test-id
			       (not (> num-items 0)))
			  (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "ZERO_ITEMS" (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails))))))
		(tests:testqueue-set-items! test-record items-list)
		(list hed tal reg reruns))
	      (begin
		(debug:print-error 0 *default-log-port* "The proc from reading the items table did not yield a list - please report this")
		(exit 1))))))

    ((and (null? fails)

			    (set! give-up #t)))
		      prereqstrs))

	(if (and give-up
		 (not (and (null? tal)(null? reg))))
	    (let ((trimmed-tal (mt:discard-blocked-tests run-id hed tal test-records))
		  (trimmed-reg (mt:discard-blocked-tests run-id hed reg test-records)))
	      (debug:print 1 *default-log-port* "WARNING: test " hed " has discarded prerequisites: " prereqstrs ", removing it from the queue")

	      (let ((test-id (rmt:get-test-id run-id hed "")))
		(if test-id (mt:test-set-state-status-by-id run-id test-id "COMPLETED" "PREQ_DISCARDED" (conc "Failed to run due to discarded prerequisites: " prereqstrs))))
	      
	      (if (and (null? trimmed-tal)
		       (null? trimmed-reg))
		  #f
                  (runs:loop-values trimmed-tal trimmed-reg reglen regfull reruns)
                  ))
	      (list (car newtal)(append (cdr newtal) reg) '() reruns))))

    ((and (null? fails) ;; have not-started tests, but unable to run them.  everything looks completed with no prospect of unsticking something that is stuck.  we should mark hed as moribund and exit or continue if there are more tests to consider
	   (null? prereq-fails)
	   (null? non-completed))
     (debug:print-info 4 *default-log-port* "cond branch - "  "ei-4")
      (if  (runs:can-keep-running? hed 20)
	  (begin
	    (runs:inc-cant-run-tests hed)
	    (debug:print-info 0 *default-log-port* "no fails in prerequisites (" (runs:pretty-string prereqs-not-met) ") for " hed " but also none running, keeping " hed " for now. Try count: " (hash-table-ref/default *seen-cant-run-tests* hed 0) ", going to wait 60 sec.") ;; 
	    ;; getting here likely means the system is way overloaded, kill a full minute before continuing
	    ;; (thread-sleep! 60) ;; TODO: gate by normalized server load > 1.0 (maxload config thing) CHECKTHIS!!!
	    ;; No runsdat, can't do this yet
	    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
	    ;;
	    (thread-sleep! 5) ;; TODO: gate by normalized server load > 1.0 (maxload config thing)
	    ;; num-retries code was here
	    ;; we use this opportunity to move contents of reg to tal
	    (list (car newtal)(append (cdr newtal) reg) '() reruns)) ;; an issue with prereqs not yet met?
	  (begin
	    (debug:print-info 1 *default-log-port* "no fails in prerequisites (" (runs:pretty-string prereqs-not-met) ") for " hed " but nothing seen running in a while, dropping test " hed " from the run queue")
	    (let ((test-id (rmt:get-test-id run-id hed "")))
	      (if test-id (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "TIMED_OUT" (conc "Prerequisites (" (runs:pretty-string prereqs-not-met) ") not seen running in a while."))))
            (runs:loop-values tal reg reglen regfull reruns)
            )))

    ((and 
       (or (not (null? fails))
	   (not (null? prereq-fails)))
       (member 'normal testmode))
      (debug:print-info 4 *default-log-port* "cond branch - "  "ei-5")
      (debug:print-info 1 *default-log-port* "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")
      (let ((test-id (rmt:get-test-id run-id hed "")))
	(if test-id
	    (if (not (null? prereq-fails))
		(mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_DISCARDED" (conc "Failed to run due to prior failed prerequisites: "(runs:pretty-string prereq-fails)))
		(mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL"      (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails ))))))
      ;; (debug:print 4 *default-log-port*"BB> set PREQ_FAIL on "hed)
      ;; (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_FAIL"      "Failed to run due to failed prerequisites")))) ;; BB: this works, btu equivalent for itemwait mode does not work.
      (if (or (not (null? reg))(not (null? tal)))
	  (begin
	    (hash-table-set! test-registry hed 'CANNOTRUN)
            (runs:loop-values tal reg reglen regfull (cons hed reruns))
            )
	  #f)) ;; #f flags do not loop
      
     ((and (not (null? fails))(member 'toplevel testmode))
      (debug:print-info 4 *default-log-port* "cond branch - "  "ei-6")
      (if (or (not (null? reg))(not (null? tal)))
	   (list (car newtal)(append (cdr newtal) reg) '() reruns)
	  #f)) 
     ((null? runnables)
      (debug:print-info 4 *default-log-port* "cond branch - "  "ei-7")
      #f) ;; if we get here and non-completed is null then it is all over.
     (else
      (debug:print-info 4 *default-log-port* "cond branch - "  "ei-8")
      (debug:print 2 *default-log-port* "WARNING: FAILS or incomplete tests maybe preventing completion of this run. Watch for issues with test " hed ", continuing for now")
      (list (car newtal)(cdr newtal) reg reruns)))))

(define (runs:mixed-list-testname-and-testrec->list-of-strings inlst)
  (if (null? inlst)
      '()
      (map (lambda (t)
	     (cond

	 (registry-mutex         (runs:dat-registry-mutex runsdat))
	 (flags                  (runs:dat-flags runsdat))
	 (keyvals                (runs:dat-keyvals runsdat))
	 (run-info               (runs:dat-run-info runsdat))
	 (all-tests-registry     (runs:dat-all-tests-registry runsdat))
	 (run-limits-info        (runs:dat-can-run-more-tests runsdat))
	 ;; (runs:can-run-more-tests run-id jobgroup max-concurrent-jobs)) ;; look at the test jobgroup and tot jobs running
	 (have-resources         (and #; (if *journal-stats*
					  (let* ((dbfname (conc
							   (dbfile:run-id->dbnum run-id)
							   ".db"))
						 (load (tt:get-journal-stats dbfname)))
					    (if (> load 0.1) ;; dbs too busy to start more tests
						(begin
						  (debug:print-info 0 *default-log-port* "Gating launch due to db load "load" based on journal file observations for "dbfname)
						 #f)
						#t))
					  (begin
					    (debug:print-info 0 *default-log-port* "Journal gating not started for "run-id)
					    #t)) ;; if journal monitoring not started do not gate
				      (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        (rmt:get-prereqs-not-met run-id waitons hed item-path mode: testmode itemmaps: itemmaps))
	 ;; (prereqs-not-met         (mt:lazy-get-prereqs-not-met run-id waitons item-path mode: testmode itemmap: itemmap))
         (prereqs-running        (runs:calc-prereqs-running prereqs-not-met))
	 (fails                  (if (list? prereqs-not-met) ;; TODO: rename fails to failed-prereqs
				      (runs:calc-fails prereqs-not-met)
				      (begin
					(debug:print-error 0 *default-log-port* "prereqs-not-met is not a list! " prereqs-not-met)
					'())))
	 (non-completed           (filter (lambda (x)             ;; remove hed from not completed list, duh, of course it is not completed!
					    (not (equal? x hed)))

	   (if (and (not (common:on-homehost?))
		    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
					      (conc "Waiting for homehost load to drop below normalized value of " maxhomehostload)))

	   ;; lastly lets check the servers are not overloaded by looking at threads
	   (tt:wait-on-server-load run-id *ttdat*)
	   
	   )))
    
    
    (if (and (not (null? prereqs-not-met))
	     (runs:lownoise (conc "waiting on tests " prereqs-not-met hed) 60))
	(debug:print-info 2 *default-log-port* "waiting on tests; " (string-intersperse (runs:mixed-list-testname-and-testrec->list-of-strings prereqs-not-met) ", ")))

    ;; Don't know at this time if the test have been launched at some time in the past
    ;; i.e. is this a re-launch?

			 ", "))
      (thread-sleep! 0.051)
      (list hed tal reg reruns))
     
     ;; 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" 600)
	  (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! 0.25)
      
      ;; new logic.
      ;; If it has been more than 10 seconds since we were last here don't wait at all
      ;; otherwise sleep 2 seconds to give db a rest and let dashboard read data
      (if (runs:lownoise "frequent-no-resources" 10)
	  (thread-sleep! 0.25) ;; no significant delay
	  (thread-sleep! 2))
      ;; 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)
	       (and (member 'toplevel testmode) ;;  'toplevel)
		    (null? non-completed)
		    (not (member 'exclusive testmode)))))
      ;; (hash-table-delete! *max-tries-hash* (db:test-make-full-name test-name item-path))
      ;; we are going to reset all the counters for test retries by setting a new hash table
      ;; this means they will increment only when nothing can be run
      (set! *max-tries-hash* (make-hash-table))
      
      (run:test run-id run-info keyvals runname test-record flags #f test-registry all-tests-registry runsdat testdat)
      (set! *last-test-launch* (current-seconds))
      (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)))
	  (runs:loop-values tal reg reglen regfull reruns) ;; hed should be dropped at this time
	  #f))

	    
	    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
	    (thread-sleep! 5)
	    (list (car newtal)(cdr newtal) reg reruns))
	  ;; the waiton is FAIL so no point in trying to run hed ever again
	  (begin
            (let ((my-test-id (rmt:get-test-id run-id test-name item-path)))
              (mt:test-set-state-status-by-id-unless-completed run-id my-test-id "COMPLETED" "PREQ_FAIL" (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails))))            
            (if (or (not (null? reg))(not (null? tal)))
                (if (vector? hed)
		  (begin
		    (debug:print 1 *default-log-port* "WARNING: Dropping test " test-name "/" item-path
				 " from the launch list as it has prerequistes that are FAIL: " (runs:pretty-string fails))
		    (let ((test-id (rmt:get-test-id run-id hed "")))
		      (if test-id (mt:test-set-state-status-by-id-unless-completed run-id test-id "COMPLETED" "PREQ_FAIL" (conc "Failed to run due to failed prerequisites: " (runs:pretty-string fails)) )))
		    (runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
		    ;; (thread-sleep! *global-delta*)
		    ;; This next is for the items

                    (if (not (null? fails))

                        (rmt:set-state-status-and-roll-up-items run-id test-name item-path "NOT_STARTED" "PREQ_FAIL" (runs:pretty-string fails)) 

                        (rmt:set-state-status-and-roll-up-items run-id test-name item-path "NOT_STARTED" "BLOCKED" #f) )
		    (hash-table-set! test-registry (db:test-make-full-name test-name item-path) 'removed)
		    (runs:loop-values tal reg reglen regfull reruns))
		  (let ((nth-try (hash-table-ref/default test-registry hed 0))) ;; hed not a vector...
                    (debug:print 2 *default-log-port* "nth-try("hed")="nth-try)
		    (cond
		     ;;((member "RUNNING" (map db:test-get-state prereqs-not-met))
                      ((> 0 (length prereqs-running))
		      (if (runs:lownoise (conc "possible RUNNING prerequisites " hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has possible RUNNING prerequisites: " prereqs-running ", don't give up on it yet."))
		      (thread-sleep! 0.1)
		      (runs:loop-values tal reg reglen regfull reruns))
		     ((or (not nth-try) ;; BB: condition on subsequent tries, condition below fires on first try 
			  (and (number? nth-try)
			       (< nth-try 2)))
		      (hash-table-set! test-registry hed (if (number? nth-try)
							     (+ nth-try 1)
							     0))
		      (if (runs:lownoise (conc "not removing test " hed) 60)
			  (debug:print 1 *default-log-port* "WARNING: not removing test " hed " from queue although it may not be runnable due to FAILED prerequisites: "(runs:pretty-string fails)))
		      ;; may not have processed correctly. Could be a race condition in your test implementation? Dropping test " hed) ;;  " as it has prerequistes that are FAIL. (NOTE: hed is not a vector)")
		      (runs:shrink-can-run-more-tests-count runsdat) ;; DELAY TWEAKER (still needed?)
		      (runs:loop-values newtal reg reglen regfull reruns))
		     ((symbol? nth-try) ;; BB: 'done matches here in one case where prereq itemwait failed.  This is first "try"
		      (if (eq? nth-try 'removed) ;; removed is removed - drop it NOW
			  (if (null? tal)
			      #f ;; yes, really
			      (list (car tal)(cdr tal) reg reruns))
			  (begin
			    (if (runs:lownoise (conc "FAILED prerequisites or other issue" hed) 60)
				(debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequisites (" (runs:pretty-string fails)") or other issue. Internal state >" nth-try "< will be overridden and we'll retry."))
			    (let* ((test-id      (rmt:get-test-id run-id hed item-path))
				   (test-info    (rmt:get-testinfo-state-status run-id test-id)) ;; we need *current* info
				   (status       (db:test-get-status test-info)))
			      (if (equal? status "KEEP_TRYING")
				  (mt:test-set-state-status-by-testname-unless-completed run-id test-name item-path "COMPLETED" "PREQ_FAIL" #f)
				  (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "KEEP_TRYING" #f)))
			    (hash-table-set! test-registry hed 'removed) ;; was 0
                            (if (not (and (null? reg) (null? tal)))
                                (runs:loop-values tal reg reglen regfull reruns)
                                #f))))
		     (else
		      (if (runs:lownoise (conc "FAILED prerequitests and we tried" hed) 60)
			  (debug:print 0 *default-log-port* "WARNING: test " hed " has FAILED prerequitests: " (runs:pretty-string fails) "  and we've tried at least 10 times to run it. Giving up now."))
		      ;; (debug:print 0 *default-log-port* "         prereqs: " prereqs-not-met)
		      (hash-table-set! test-registry hed 'removed)
		      (mt:test-set-state-status-by-testname run-id test-name item-path "NOT_STARTED" "TEN_STRIKES" #f)
		      ;; I'm unclear on if this roll up is needed - it may be the root cause of the "all set to FAIL" bug.
		      (rmt:set-state-status-and-roll-up-items run-id test-name item-path #f "FAIL" #f) ;; treat as FAIL
		      (list (if (null? tal)(car newtal)(car tal))
			    tal

;; 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))

(define *last-loop-time-ms* 0)

;;======================================================================
;; 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)

         (registry-mutex        (make-mutex))
         (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))
	 (incoming-tests        '()) ;; queue up incoming tests here to tack on to tal when it gets low
         ;; (tdbdat                (tasks:open-db))

         (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)
	    ))







      ;; (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))))

			   test-name:   test-name
			   item-path:   item-path
			   jobgroup:    jobgroup
			   waitons:     waitons
			   testmode:    testmode
			   newtal:      newtal
			   itemmaps:    itemmaps
			   prereqs-not-met: '()
			   )))

	;; too-tight loop detection and delay, this might hide issues
	;; that occur in long run times. Consider commenting when debugging
	;; 
	(if (and (>= num-running max-concurrent-jobs)
		 (< (- (current-milliseconds) *last-loop-time-ms*) 500))
	    (begin
	      (if (runs:lownoise "too-tight-loop" 5)
		  (debug:print-info 2 *default-log-port* "Excessively fast loop, delaying 1/2 second"))
	      (thread-sleep! 0.5)))
	(set! *last-loop-time-ms* (current-milliseconds))
     
	(runs:dat-regfull-set! runsdat regfull)


	(if (> (- (current-seconds) *last-test-launch*) 5)        ;; be pretty aggressive for five seconds after
	    (runs:too-soon-delay (conc "loop delay " hed) 1 0.6)    ;; starting a test then apply more delay
	    (runs:too-soon-delay (conc "loop delay " hed) 1 0.1)) 
	
	(if (> num-running 0)
            (set! last-time-some-running (current-seconds)))

        (if (> (current-seconds)(+ last-time-some-running (or (configf:lookup *configdat* "setup" "give-up-waiting") 36000)))
            (hash-table-set! *max-tries-hash* tfullname (+ (hash-table-ref/default *max-tries-hash* tfullname 0) 1)))
	;; (debug:print 0 *default-log-port* "max-tries-hash: " (hash-table->alist *max-tries-hash*))

		  (debug:print-info 0 *default-log-port* "Skipping test " tfullname " as it has been marked do not run due to being completed or not runnable"))
	      (if (or (not (null? tal))(not (null? reg)))
		  (loop (runs:queue-next-hed tal reg reglen regfull)
			(runs:queue-next-tal tal reg reglen regfull)
			(runs:queue-next-reg tal reg reglen regfull)
			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:         " (runs:pretty-long-list tal)
		     "\n  reg:         " reg
                     "\n  test-record  " test-record

		 (let* ((jobgroup              (runs:testdat-jobgroup testdat-in))
			(can-run-more-tests    (runs:dat-can-run-more-tests runsdat))
			(last-jobs-check-time  (runs:dat-last-jobs-check-time runsdat))
			(should-check-jobs     (match can-run-more-tests
						 ((can-run-more-flag num-running nr-in-jobgroup max-concurrent-jobs . params)
						  (if (< (- max-concurrent-jobs num-running) 25)
						      (begin
							(debug:print-info 2 *default-log-port*
									  "less than 20 jobs headroom, ("max-concurrent-jobs
									  "-"num-running")>20. Forcing prelaunch check.")
							#t)
						      #f))
						 (else #f)))) ;; no record yet
		   (if should-check-jobs
		       (let loop-can-run-more

				  ;; 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



	  (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))))

                                   (my-item-path (item-list->path my-itemdat))

                                   (newtestname (db:test-make-full-name hed my-item-path)))    ;; test names are unique on testname/item-path
                              (tests:testqueue-set-items!     new-test-record #f)
                              (tests:testqueue-set-itemdat!   new-test-record my-itemdat)
                              (tests:testqueue-set-item_path! new-test-record my-item-path)
                              (hash-table-set! test-records newtestname new-test-record)
			      ;; BUG: This next line sucks up a lot of horsepower
			      ;; (set! tal (append tal (list newtestname)))
			      ;; (set! tal (cons newtestname tal)) ;; 4/6/2023 - try using cons, does it matter if the test gets added at the beginning?
			      (set! incoming-tests (cons newtestname incoming-tests))
			      ))  ;; since these are itemized create new test names testname/itempath
                          items-in-testpatt)))
          
	  (if (and (< (length tal) 20)
		   (not (null? incoming-tests)))
	      (begin
		(set! tal (append tal (reverse incoming-tests)))
		(set! incoming-tests '())))

	  ;; At this point we have possibly added items to tal but all must be handed off to 
	  ;; INNER COND logic. I think loop without rotating the queue 
	  ;; (loop hed tal reg reruns))
	  ;; (let ((newtal (append tal (list hed))))  ;; We should discard hed as it has been expanded into it's items? Yes, but only if this *is* an itemized test
	  ;; (loop (car newtal)(cdr newtal) reg reruns)
	  (if (null? tal)

		  (if loop-list
		      (apply loop loop-list)
                      (debug:print-info 4 *default-log-port* " -- Can't expand hed="hed)
                      )
                  )
		;; if can't run more just loop with next possible test
		(loop (car newtal)(cdr newtal) reg reruns))))
        

	 ;; this case should not happen, added to help catch any bugs
	 ((and (list? items) itemdat)
          (debug:print-info 4 *default-log-port* "cond branch - "  "rtq-5")
	  (debug:print-error 0 *default-log-port* "Should not have a list of items in a test and the itemspath set - please report this")
	  (exit 1))

	 ((not (null? reruns))
          (debug:print-info 4 *default-log-port* "cond branch - "  "rtq-6")
	  (let* ((newlst (tests:filter-non-runnable run-id tal test-records)) ;; i.e. not FAIL, WAIVED, INCOMPLETE, PASS, KILLED,
		 (junked (lset-difference equal? tal newlst)))
	    (debug:print-info 4 *default-log-port* "full drop through, if reruns is less than 100 we will force retry them, reruns=" reruns ", tal=" tal)
	    (if (< num-retries max-retries)
		(set! newlst (append reruns newlst)))
	    (set! num-retries (+ num-retries 1))
	    ;; (thread-sleep! (+ 1 *global-delta*))
	    (if (not (null? newlst))
		;; since reruns have been tacked on to newlst create new reruns from junked
		(loop (car newlst)(cdr newlst) reg (delete-duplicates junked)))))

	 ((not (null? tal))
          (debug:print-info 4 *default-log-port* "cond branch - "  "rtq-7")
	  (debug:print-info 4 *default-log-port* "I'm pretty sure I shouldn't get here."))
	 ((not (null? reg)) ;; could we get here with leftovers?
          (debug:print-info 4 *default-log-port* "cond branch - "  "rtq-8")
	  (debug:print-info 0 *default-log-port* "Have leftovers!")
	  (loop (car reg)(cdr reg) '() reruns))
	 (else
          (debug:print-info 4 *default-log-port* "cond branch - "  "rtq-9")
	  (debug:print-info 4 *default-log-port* "Exiting loop with...\n  hed=" hed "\n  tal=" tal "\n  reruns=" reruns))
	 ))) ;; 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)
    ;; (if (runs:dat-load-mgmt-function runsdat)((runs:dat-load-mgmt-function runsdat)))
    (thread-sleep! 0.1) ;; 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)(+ last-time-incomplete 900))
		(let ((actual-num-running (rmt:get-count-tests-running-for-run-id run-id)))
		  (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))))))
	    ;; (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")))

(define (runs:calc-fails prereqs-not-met)
  (filter (lambda (test)
	    (and (vector? test) ;; not (string? test))
		 (member (db:test-get-state test) '("INCOMPLETE" "COMPLETED")) ;; TODO: pull from *common:stuff...*
		 (not (member (db:test-get-status test)
			      '("PASS" "WARN" "WAIVED" "SKIP")))))
	  prereqs-not-met))

(define (runs:calc-prereq-fail prereqs-not-met) ;; REMOVEME since NOT_STARTED/PREQ_FAIL is now COMPLETED/PREQ_FAIL
  (filter (lambda (test)
	    (and (vector? test) ;; not (string? test))
		 (equal? (db:test-get-state test) "NOT_STARTED")
		 (not (member (db:test-get-status test)

   (lambda (t)
     (or (not (vector? t))
	 (and (equal? "NOT_STARTED" (db:test-get-state t))
	      (member (db:test-get-status t)
		      '("n/a" "KEEP_TRYING")))
	 (and (equal? "RUNNING" (db:test-get-state t))))) ;; account for a test that is running
   prereqs-not-met))

(define (runs:calc-prereqs-running prereqs-not-met)
  (if (list? prereqs-not-met)
    (filter 
     (lambda (t)
       (or (not (vector? t))
         (member (db:test-get-state t) '("RUNNING" "LAUNCHED" "REMOTE_HOST_START"))
       ))
     prereqs-not-met)
    '()
  )
)

(define (runs:pretty-string lst)
  (map (lambda (t)
	 (if (not (vector? t))
	     (conc t)
	     (conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
       lst))

;; parent-test is there as a placeholder for when parent-tests can be run as a setup step
;;
(define (run:test run-id run-info keyvals runname test-record flags parent-test test-registry all-tests-registry runsdat testdat-rec)
  ;; All these vars might be referenced by the testconfig file reader
  ;; 
  ;; NEED to reprocess testconfig here, ensuring that item variables are available.
  ;; This is for Tal's issue with item-specific env vars not being set for use in skip.
  ;; HSD https://hsdes.intel.com/appstore/icf/index.html#/article?articleId=1408763273
  ;; Also later HSD https://hsdes.intel.com/appstore/article/#/14012138487
  ;;
  (let* ((test-name    (tests:testqueue-get-testname   test-record))
	 (test-waitons (tests:testqueue-get-waitons    test-record))
	 (itemdat      (tests:testqueue-get-itemdat    test-record))
	 (item-path     "")
	 (db           #f)
	 (full-test-name #f))

    ;; setting itemdat to a list if it is #f
    (if (not itemdat)(set! itemdat '()))
    (set! item-path (item-list->path itemdat))
    (set-item-env-vars itemdat)
    (set! full-test-name (db:test-make-full-name test-name item-path))
    (runs:set-megatest-env-vars run-id inrunname: runname testname: test-name itempath: item-path) ;; these may be needed by the launching process

   (let* ((test-conf ;; re-instated the tests:get-testconfig to fix HSD https://hsdes.intel.com/appstore/article/#/14012138487, need to be able to skip using [items], [itemstable] variables.
	   ;; (tests:testqueue-get-testconfig test-record )) ;; vector-ref test-record 3
           (tests:get-testconfig test-name item-path all-tests-registry #t force-create: #t))

	   (test-path    (hash-table-ref all-tests-registry test-name)) ;; (conc *toppath* "/tests/" test-name)) ;; could use tests:get-testconfig here ...

	   (force        (hash-table-ref/default flags "-force" #f))
	   (rerun        (hash-table-ref/default flags "-rerun" #f))
	   (keepgoing    (hash-table-ref/default flags "-keepgoing" #f))
	   (incomplete-timeout (string->number (or (configf:lookup *configdat* "setup" "incomplete-timeout") "x")))
	   )

    ;; Here is where the test_meta table is best updated
    ;; Yes, another use of a global for caching. Need a better way?
    ;;
    ;; There is now a single call to runs:update-all-test_meta and this 
    ;; per-test call is not needed. Given the delicacy of the move to 
    ;; v1.55 this code is being left in place for the time being.
    ;;
    (if (not (hash-table-exists? *test-meta-updated* test-name))
        (begin
          (hash-table-set! *test-meta-updated* test-name #t)
          (runs:update-test_meta test-name test-conf)))
    
    ;; itemdat => ((ripeness "overripe") (temperature "cool") (season "summer"))
    (let* ((new-test-path (string-intersperse (cons test-path (map cadr itemdat)) "/"))
	   (test-id       (rmt:get-test-id run-id test-name item-path))

	     (set! runflag #t))

	    (else (set! runflag #f)))
	   (debug:print 4 *default-log-port* "RUNNING => runflag: " runflag " STATE: " (test:get-state testdat) " STATUS: " (test:get-status testdat))
	   (if (not runflag)
	       (if (not parent-test)
		   (if (runs:lownoise (conc "not starting test" full-test-name) 60)
		       (debug:print 3 *default-log-port* "NOTE: Not starting test " full-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
	       ;; 
	       (let ((skip-test   #f)

		     (begin
		       ;; wait for less than max jobs here
		       (if (runs:dat-wait-for-jobs-function runsdat)
			   ((runs:dat-wait-for-jobs-function runsdat) testdat-rec))
		       
		       (if (not (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat flags))
			   (begin
			     (debug:print 0 *default-log-port* "ERROR: Failed to launch the test. Exiting as soon as possible")
			     (set! *globalexitstatus* 1) ;; 
			     (process-signal (current-process-id) signal/kill))
			   )
		       ;; wait again here?
		       ))))))
	((KILLED) 
	 (debug:print 1 *default-log-port* "NOTE: " full-test-name " is already running or was explictly killed, use -force to launch it.")

         (action-chk (member (string->symbol "remove-runs") actions)))
     ;; check the sequence of actions  archive must comme before remove-runs
     (if  (and action-chk (member (string->symbol "archive") action-chk))
          (begin
          (debug:print-error 0 *default-log-port* "action remove-runs must come after archive")
          (exit 1))) 
    (print "Actions: " actions " age: " age)












    (for-each
     (lambda (action)
        (for-each
         (lambda (target)
            (let* ((runs      (hash-table-ref runs-ht target))
	           (sorted    (sort runs (lambda (a b)(< (simple-run-event_time a)(simple-run-event_time b)))))
	           (to-remove (let* ((len      (length sorted))
                                     (trim-amt (- len num-to-keep)))
                                 (if (> trim-amt 0)
                                    (take sorted trim-amt)
                                    '()))))
	   (hash-table-set! runs-ht target to-remove)
           (print target ":")
           (for-each
            (lambda (run)
              (let ((remove (member run to-remove (lambda (a b)
                                                    (eq? (simple-run-id a)
                                                         (simple-run-id b))))))
  	        (if (and age (> (simple-run-event_time run) age-mark))
		     (print "Skipping handling of " target "/" (simple-run-runname run) " as it is younger than " (args:get-arg "-age"))
		     (case action
		       ((print)
			(print " " (simple-run-runname run)
			       " " (time->string (seconds->local-time (simple-run-event_time run)) "WW%V.%u %H:%M:%S")
			       " " (if remove "REMOVE" "")))
		       ((remove-runs)


			(if remove (system (conc precmd " megatest -remove-runs -target " target " -runname " (simple-run-runname run) " -testpatt %"
						 (if (member 'kill-runs actions) ;; if kill-runs is specified then set -kill-wait to 0
						     " -kill-wait 0"
						     "")))))


		       ((archive)

			(if remove (system (conc precmd " megatest -archive save-remove -target " target " -runname " (simple-run-runname run) " -testpatt %"))))


		       ((kill-runs)
			(if remove (system (conc precmd " megatest -kill-runs -target " target " -runname " (simple-run-runname run) " -testpatt %"))))))))


            sorted)))
         (hash-table-keys runs-ht)))
      actions)
    runs-ht))

(define (remove-last-path-directory path-in)
  (let* ((dparts  (string-split path-in "/"))

	 (state-status (if (string? new-state-status) (string-split new-state-status ",") '(#f #f)))
	 (rp-mutex     (make-mutex))
	 (bup-mutex    (make-mutex))
         (keep-records (args:get-arg "-keep-records")) ;; used in conjunction with -remove-runs to keep the records, TODO: consolidate this with "mode".
	 (test-records '())) ;; for tasks that we wish to operate on all tests in one fell swoop

    (let* ((write-access-actions '(remove-runs set-state-status archive run-wait kill-runs))
           (dbfile             (conc  *toppath* "/.mtdb/main.db"))
           (readonly-mode      (not (file-write-access? dbfile))))
      (when (and readonly-mode
                 (member action write-access-actions))
        (debug:print-error 0 *default-log-port* dbfile " is readonly.  Cannot proceed with action ["action"] in which write-access isrequired .")
        (exit 1)))
    
    (debug:print-info 4 *default-log-port* "runs:operate-on => Header: " header " action: " action " new-state-status: " new-state-status)
    (if (> 2 (length state-status))
	(begin
	  (debug:print-error 0 *default-log-port* "the parameter to -set-state-status is a comma delimited string. E.g. COMPLETED,FAIL")
	  (exit)))

	       (begin
		 (case action
                   ((kill-runs)
                    (tasks:kill-runner target run-name "%")
                    (debug:print 1 *default-log-port* "Killing tests for run: " runkey " " (db:get-value-by-header run header "runname"))
                    )
		   ((remove-runs)
		    ;; use this location to cleanup old DELETED records? No. See below for same call
		    ;; (rmt:delete-old-deleted-test-records run-id)
		    ;; (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 testpatt)
		    ;; (debug:print 0 *default-log-port* "not attempting to kill any run launcher processes as testpatt is " testpatt))
		    (debug:print 1 *default-log-port* "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 2 *default-log-port* "Modifying state and status for tests for run: " runkey " " (db:get-value-by-header run header "runname")))
		   ((print-run)
		    (debug:print 1 *default-log-port* "Printing info for run " runkey ", run=" run ", tests=" tests ", header=" header)
		    action)
		   ((run-wait)
		    (debug:print 1 *default-log-port* "Waiting for run " runkey ", run=" runnamepatt " to complete"))
		   ((archive)
		    (debug:print 1 *default-log-port* "Archiving/restoring (" (args:get-arg "-archive") ") data for run: " runkey " " (db:get-value-by-header run header "runname"))

				       )
				      ) ; end case rem-status
                                    ) ; end let
                                  ); end cond has-subrun

                                 (else
                                  ;; BB - TODO - consider backgrounding to threads to delete tests (work below) 
                                  (debug:print-info 2 *default-log-port* "test: " test-name " itest-state: " test-state)
                                  (if (member test-state (list "RUNNING" "LAUNCHED" "REMOTEHOSTSTART" "KILLREQ"))
                                      (begin
                                        (if (not (hash-table-ref/default test-retry-time test-fulln #f))
                                            (begin
                                              ;; want to set to REMOVING BUT CANNOT do it here?
                                              (hash-table-set! test-retry-time test-fulln (current-seconds))))
                                        (if (> (- (current-seconds)(hash-table-ref test-retry-time test-fulln)) allow-run-time)

                                            (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)))))))





                                      (if (not (null? tal))
                                         (loop (car tal)(cdr tal)))))))
				(rmt:update-run-stats run-id 'run)) ;; (rmt:get-raw-run-stats run-id)))
                               ((kill-runs)
                                ;; RUNNING -> KILLREQ
                                ;; LAUNCHED,RUNNING,REMOTEHOSTSTART -> NOT STARTED
                                (cond
                                 ((and has-subrun (member test-state (list "RUNNING" "LAUNCHED" "REMOTEHOSTSTART" "KILLREQ")))
                                  (common:send-thunk-to-background-thread
                                   (lambda ()

                    (debug:print 2 *default-log-port* "run-paths-hash: " (hash-table-keys run-paths-hash))

                    (debug:print 1 *default-log-port* "Removing target " target "run: " run-name)
                    (if (not keep-records)
                      (begin
                        (debug:print 1 *default-log-port* "Removing DB records for the run.")
                        (rmt:delete-run run-id)
                        (rmt:delete-old-deleted-test-records run-id))
                    )
                    (if (not (equal?  linkspath "/does/not/exist/I"))
	               (begin 
                         (debug:print 1 *default-log-port* "Recursively removing links dir " linkspath)
                         (runs:recursive-delete-with-error-msg linkspath)))

                   (for-each (lambda(runpath)
                       (debug:print 1 *default-log-port* "Recursively removing runs dir " runpath)
                       (runs:recursive-delete-with-error-msg runpath)
                     )
                     runpaths
                   )
		       )))))
	 ))
     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 
             (debug:print 0 *default-log-port* "db archive started")  
             (archive:megatest-db target runnamepatt)
             (debug:print 0 *default-log-port* "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)

	;; 			      (thread-sleep! 1)
	 ;; 			      (loop (rmt:no-sync-get-lock lock-key) expire-time)))))))
	 )
    (case clean-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 2 *default-log-port* "Attempting to remove " (if real-dir (conc " dir " real-dir " and ") "") " link " run-dir)
    (if (and real-dir 
	     (> (string-length real-dir) 5)
	     (common:file-exists? real-dir)) ;; bad heuristic but should prevent /tmp /home etc.
	(let* ((realpath (resolve-pathname run-dir)))
	  (debug:print-info 1 *default-log-port* "Recursively removing " realpath)
	  (if (common:file-exists? realpath)
	      (runs:safe-delete-test-dir realpath)

     (lambda (key)
       (let* ((idx (cadr key))
	      (fld (car  key))
	      (val (configf:lookup test-conf "test_meta" fld)))
	 ;; (debug:print 5 *default-log-port* "idx: " idx " fld: " fld " val: " val)
	 (if (and val (not (equal? (vector-ref currrecord idx) val)))
	     (begin
	       (debug:print 2 *default-log-port* "Updating " test-name " " fld " to " val)
	       (rmt:testmeta-update-field test-name fld val)))))
     '(("author" 2)("owner" 3)("description" 4)("reviewed" 5)("tags" 9)("jobgroup" 10)))))

;; find tests with matching tags, tagpatt is a string "tagpatt1,tagpatt2%, ..."
;;
(define (runs:get-tests-matching-tags tagpatt)
  (let* ((tagdata (rmt:get-tests-tags))

	  (let* ((linktree (common:get-linktree)) ;; (if toppath (configf:lookup *configdat* "setup" "linktree")))
		 (runtop   (conc linktree "/" target "/" runname))
		 (files    (if (common:file-exists? runtop)
			       (append (glob (conc runtop "/.megatest*"))
				       (glob (conc runtop "/.runconfig*")))
			       '())))
	    (if (null? files)
		(debug:print-info 2 *default-log-port* "No cached megatest or runconfigs files found. None removed.")
		(begin
		  (debug:print-info 2 *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 ", 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")))