Megatest

			     (alist-ref "keep-going" params equal?)
			     #f)))
	(debug:print 0 *default-log-port* "keep-going=" keep-going)
	(and keep-going (equal? (car keep-going) "yes")))))

;; if handed a string, process it, else look for MT_CMDINFO
(define (launch:get-cmdinfo-assoc-list #!key (encoded-cmd #f))
  (let ((enccmd (if encoded-cmd encoded-cmd (get-environment-variable "MT_CMDINFO"))))
    (if enccmd
	(common:read-encoded-string enccmd)
	'())))

(define (launch:manage-steps run-id test-id item-path fullrunscript ezsteps subrun test-name tconfigreg exit-info m)
  ;; (let-values
  ;;  (((pid exit-status exit-code)

			     (prevstep #f))
                    (debug:print-info 0 *default-log-port* "Processing ezstep \"" (string-intersperse ezstep " ") "\"")
		    ;; check exit-info (vector-ref exit-info 1)
		    (if (launch:einf-exit-status exit-info) ;; (vector-ref exit-info 1)
			(let* ((logpro-used (launch:runstep ezstep run-id test-id exit-info m tal testconfig all-steps-dat))
			       (stepname    (car ezstep))
			       (stepparms   (hash-table-ref all-steps-dat stepname)))
			  (set-environment-variable! "MT_STEP_NAME" stepname)
			  (pp (hash-table->alist all-steps-dat))
			  ;; if logpro-used read in the stepname.dat file
			  (if (and logpro-used (common:file-exists? (conc stepname ".dat")))
			      (launch:load-logpro-dat run-id test-id stepname))
			  (if (steprun-good? logpro-used (launch:einf-exit-code exit-info) stepparms)
			      (if (not (null? tal))
				  (loop (car tal) (cdr tal) stepname))

          (rmt:set-state-status-and-roll-up-items run-id test-id 'foo "RUNNING" "n/a" "was marked dead; really still running.")
          ;;(set! kill-reason "KILLING TEST because it was marked as DEAD by launch:handle-zombie-tests (might indicate really overloaded server or else overzealous setup.deadtime)") ;; MARK RUNNING
          (set! kill-job? #f)))

        (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
        (launch:handle-zombie-tests run-id)
        (when do-sync
          ;;(with-output-to-file (conc (get-environment-variable "MT_TEST_RUN_DIR") "/last-loadinfo.log" #:append)
          ;;  (lambda () (pp (list (current-seconds) new-cpu-load new-disk-free (calc-minutes)))))
          ;; (common:telemetry-log "zombie" (conc  "launch:monitor-job - dosync started at "(current-seconds)))
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          ;; (common:telemetry-log "zombie" (conc "launch:monitor-job - dosync finished at "(current-seconds)))
	  )
        
	(if kill-job? 

                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional


(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (set-environment-variable! "MT_CMDINFO" encoded-cmd)
    ;;(bb-check-path msg: "launch:execute incoming")
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)
	;; (test-name sqlitespeed) (runscript runscript.rb) (db-host localhost) (run-id 1))
	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; testpath is the test spec area
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))  ;; work-area is the test run area
	       (test-name (assoc/default 'test-name cmdinfo))

                                                    (debug:print 0 *default-log-port* "INFO: test data from " testpath " not copied properly or filesystem problems causing data to not be found. Re-running the copy command.")
                                                    (debug:print 0 *default-log-port* "INFO: missing files from " work-area ": " (string-intersperse bad-files ", "))
                                                    (launch:test-copy testpath work-area))))
                                            ;; one more time, change to the work-area directory
                                            (change-directory work-area)))
	       ) ;; let*

	  (if contour (set-environment-variable! "MT_CONTOUR" contour))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (set-environment-variable! "MT_TESTSUITENAME" areaname)
	  (set-environment-variable! "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (set-environment-variable! "MT_TEST_RUN_DIR"  work-area)

	  (launch:setup) ;; should be properly in the run area home now

	  (if contour (set-environment-variable! "MT_CONTOUR" contour))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (set-environment-variable! "MT_TESTSUITENAME" areaname)
	  (set-environment-variable! "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
          (change-directory *toppath*) ;; temporarily switch to the run area home
	  (set-environment-variable! "MT_TEST_RUN_DIR"  work-area)

	  (launch:setup) ;; should be properly in the run area home now
          
	  (set! tconfigreg (tests:get-all)) ;; mapping of testname => test source path
	  (let ((sighand (lambda (signum)
			   ;; (signal-mask! signum) ;; to mask or not? seems to cause issues in exiting
			   (if (eq? signum signal/stop)

		(debug:print 4 *default-log-port* "varpairs: " varpairs)
		(map (lambda (varpair)
		       (let ((varval (string-split varpair "=")))
			 (if (eq? (length varval) 2)
			     (let ((var (car varval))
				   (val (cadr varval)))
			       (debug:print 1 *default-log-port* "Adding pre-var/val " var " = " val " to the environment")
			       (set-environment-variable! var val)))))
		     varpairs)))
          ;;(bb-check-path msg: "launch:execute post block 2")
	  (for-each
	   (lambda (varval)
	     (let ((var (car varval))
		   (val (cadr varval)))
	       (if val
		   (set-environment-variable! var val)
		   (begin
		     (debug:print-error 0 *default-log-port* "required variable " var " does not have a valid value. Exiting")
		     (exit)))))
	     (list 
	      (list  "MT_TEST_RUN_DIR" work-area)
	      (list  "MT_TEST_NAME" test-name)
	      (list  "MT_ITEM_INFO" (conc itemdat))
	      (list  "MT_ITEMPATH"  item-path)
	      (list  "MT_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
	      (list  "MT_TESTSUITENAME" (common:get-area-name))))
          ;;(bb-check-path msg: "launch:execute post block 3")

	  (if mt-bindir-path (set-environment-variable! "PATH" (conc (get-environment-variable "PATH") ":" mt-bindir-path)))
          ;;(bb-check-path msg: "launch:execute post block 4")
	  ;; (change-directory top-path)
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

	  
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
          ;;(bb-check-path msg: "launch:execute post block 41")
	  (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals)
          ;;(bb-check-path msg: "launch:execute post block 42")
	  (set-item-env-vars itemdat)
          ;;(bb-check-path msg: "launch:execute post block 43")
          (let ((blacklist (configf:lookup *configdat* "setup" "blacklistvars")))
            (if blacklist
		(let ((vars (string-split blacklist)))
		  (save-environment-as-files "megatest" ignorevars: vars)
		  (for-each (lambda (var)
			      (unset-environment-variable! var))
			    vars))
                (save-environment-as-files "megatest")))
          ;;(bb-check-path msg: "launch:execute post block 44")
	  ;; open-run-close not needed for test-set-meta-info
	  ;; (tests:set-full-meta-info #f test-id run-id 0 work-area)
	  ;; (tests:set-full-meta-info test-id run-id 0 work-area)
	  (tests:set-full-meta-info #f test-id run-id 0 work-area 10)

	   (or (args:get-arg "-run")
	       (args:get-arg "-runtests")
	       (args:get-arg "-execute")))
      (let* ((linktree (common:get-linktree)) ;; (get-environment-variable "MT_LINKTREE"))
	     (target   (common:args-get-target exit-if-bad: #t))
	     (runname  (or (args:get-arg "-runname")
			   (args:get-arg ":runname")
			   (get-environment-variable "MT_RUNNAME")))
	     (fulldir  (conc linktree "/"
			     target "/"
			     runname)))
	(if (and linktree (common:file-exists? linktree)) ;; can't proceed without linktree
	    (begin
	      (debug:print-info 0 *default-log-port* "Have -run with target=" target ", runname=" runname ", fulldir=" fulldir ", testpatt=" (or (args:get-arg "-testpatt") "%"))
	      (if (not (common:file-exists? fulldir))

		  (set! *configinfo* first-pass)
		  (set! *toppath*    (or toppath (cadr first-pass))) ;; use the gathered data unless already have it
		  (set! toppath      *toppath*)
		  (if (not *toppath*)
		      (begin
			(debug:print-error 0 *default-log-port* "you are not in a megatest area!")
			(exit 1)))
		  (set-environment-variable! "MT_RUN_AREA_HOME" *toppath*)
		  ;; the seed read is done, now read runconfigs, cache it then read megatest.config one more time and cache it
		  (let* ((keys         (common:list-or-null (rmt:get-keys)
							    message: "Failed to retrieve keys in launch.scm. Please report this to the developers."))
			 (key-vals     (keys:target->keyval keys target))
			 (linktree     (common:get-linktree)) ;; (or (get-environment-variable "MT_LINKTREE")(if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))
					;     (if *configdat*
					; 	   (configf:lookup *configdat* "setup" "linktree")
					; 	   (conc *toppath* "/lt"))))
			 (second-pass  (configf:find-and-read-config
					mtconfig
					environ-patt: "env-override"
					given-toppath: toppath
					pathenvvar: "MT_RUN_AREA_HOME"
					env-to-use: (module-environment 'bigmod)))
			 (runconfigdat (begin     ;; this read of the runconfigs will see any adjustments made by re-reading megatest.config
					 (for-each (lambda (kt)
						     (set-environment-variable! (car kt) (cadr kt)))
						   key-vals)
					 (configf:read-config (conc toppath "/runconfigs.config") *runconfigdat* #t ;; consider using runconfig:read some day ...
						      sections: sections)))
                         (cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
                         (mtcachef     (car cachefiles))
                         (rccachef     (cdr cachefiles)))
                    ;;  trap exception due to stale NFS handle -- Error: (open-output-file) cannot open file - Stale NFS file handle: "/p/fdk/gwa/lefkowit/mtTesting/qa/primbeqa/links/p1222/11/PDK_r1.1.1/prim/clean/pcell_testgen/.runconfigs.cfg-1.6427-7d1e789cb3f62f9cde719a4865bb51b3c17ea853" - ticket 220546342

			(create-symbolic-link linktree tlink)))))
	      (begin
		(debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
		)))
	(if (and *toppath*
		 (directory-exists? *toppath*))
	    (begin
	      (set-environment-variable! "MT_RUN_AREA_HOME" *toppath*)
	      (set-environment-variable! "MT_TESTSUITENAME" (common:get-area-name)))
	    (begin
	      (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
	      (set! *toppath* #f) ;; force it to be false so we return #f
	      #f))
	
        ;; one more attempt to cache the configs for future reading
        (let* ((cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))

	    (list #f #f)))))


(define (launch:handle-zombie-tests run-id)
  (let* ((key (conc "zombiescan-runid-"run-id))
         (now (current-seconds))
         (threshold (- (current-seconds)  (* 2 (or (configf:lookup-number *configdat* "setup" "deadtime") 120))))
         (val (rmt:get-var run-id key))
         (do-scan?
          (cond
           ((not val)
            #t)
           ((< val threshold)
            #t)
           (else #f))))
    (when do-scan?
      (debug:print 1 *default-log-port* "INFO: search and mark zombie tests")
      (rmt:set-var run-id key (current-seconds))
      (runs:find-and-mark-incomplete-and-check-end-of-run run-id #f))))





;; 1. look though disks list for disk with most space

;; > 3 RUNNING with not test_dead do nothing (run should already be RUNNING/ na
;; > 0 RUNNING and test_dead then send KILLREQ ==> COMPLETED
;; 0 RUNNING ==> this is actually the first condition, should not get here

(define (runs:end-of-run-check run-id )
  (let* ((not-completed-cnt (rmt:get-not-completed-cnt run-id))  
	 (running-cnt (rmt:get-count-tests-running-for-run-id run-id))
	 (all-test-launched (rmt:get-var run-id (conc "lunch-complete-" run-id)))
	 (current-state (rmt:get-run-state run-id))
	 (current-status (rmt:get-run-status run-id)))
    ;;get-vars run-id to query metadata table to check if all completed. if all-test-launched = yes then only not-completed-cnt = 0 means everyting is completed if no entry found in the table do nothing 
    (debug:print 0 *default-log-port* "Running test cnt :" running-cnt)                      
    (rmt:set-state-status-and-roll-up-run  run-id current-state current-status)
    (runs:update-junit-test-reporter-xml run-id) 
    (cond 
     ((and all-test-launched (eq? not-completed-cnt 0) (equal? all-test-launched "yes" ))
      (if (and (equal? (rmt:get-var run-id (conc "end-of-run-" run-id)) "no") (common:simple-lock (conc "endOfRun" run-id)))
	  (begin
	    (debug:print 4 *default-log-port* "look for  post hook. currseconds: " (current-seconds) " EOR " (rmt:get-var run-id (conc "end-of-run-" run-id)))
	    (debug:print 0 *default-log-port* "End of Run Detected.")
	    (rmt:set-var (conc "end-of-run-" run-id) "yes")
					;(thread-sleep! 10)
	    (runs:run-post-hook run-id)
	    (debug:print 4 *default-log-port* "currseconds: " (current-seconds)" eor: " (rmt:get-var run-id (conc "end-of-run-" run-id)))
	    (common:simple-unlock (conc "endOfRun" run-id)))
	  (debug:print 0 *default-log-port* "End of Run Detected but not running post hook. This should happen when eor is set to yes. This will happen only when 2 tests exit at smae time. eor= " (rmt:get-var run-id (conc "end-of-run-" run-id)))))
     ((> running-cnt 3) 
      (debug:print 0 *default-log-port* "There are " running-cnt " tests running." ))
     ((> running-cnt 0)
      (debug:print 0 *default-log-port* "running cnt > 0 but <= 3 kill-running-tests-if-dead" )
      (let ((kill-cnt (launch:kill-tests-if-dead run-id)))
	(if (and all-test-launched  (equal? all-test-launched "yes") (eq? kill-cnt running-cnt))
	    (runs:end-of-run-check run-id)))) ;;todo

                                                      #f ;; sort-by
                                                      #f ;; sort-order
                                                      #f ;; get full data (not 'shortlist)
                                                      0 ;; (runs:gendat-inc-results-last-update *runs:general-data*) ;; last update time
                                                      'dashboard)
                               '()))
           (log-dir         (conc *toppath* "/logs"))
           (log-file        (conc "post-hook-" (string-translate (get-environment-variable "MT_TARGET") "/" "-") "-" (get-environment-variable "MT_RUNNAME") ".log"))
           (full-log-fname  (conc log-dir "/" log-file)))
      (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

	      (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."))))))


(define (runs:rerun-hook test-id new-test-path testdat rerunlst)
    (let* ((rerun-hook   (configf:lookup *configdat* "runs" "rerun-hook"))
           (log-dir         (conc *toppath* "/reruns/logs"))
           (target (get-environment-variable "MT_TARGET"))
           (runname (common:args-get-runname))
           (rundir (db:test-get-rundir testdat))
           (tarfiledir (conc *toppath* "/reruns"))
           (status (db:test-get-status testdat))
           (comment (conc "\"" (db:test-get-comment testdat) "\"" ))
           (testname (db:test-get-testname testdat))
           (itempath (db:test-get-item-path testdat))

(define (runs:update-junit-test-reporter-xml run-id)
  (let*	((junit-test-reporter	(configf:lookup *configdat* "runs" "junit-test-reporter-xml"))
	 (junit-test-report-dir  (configf:lookup *configdat* "runs" "junit-test-report-dir"))
	 (xml-dir		(if (and junit-test-reporter (equal? junit-test-reporter "yes" ))
				    (if junit-test-report-dir
					junit-test-report-dir
					(conc (get-environment-variable "MT_LINKTREE") "/" (get-environment-variable "MT_TARGET") "/" (get-environment-variable "MT_RUNNAME")))
				    #f))
	 (xml-ts-name		(if xml-dir
				    (conc (get-environment-variable "MT_TESTSUITENAME")"."(string-translate (get-environment-variable "MT_TARGET") "/" ".") "." (get-environment-variable "MT_RUNNAME"))
				    #f))
         (keyname               (if xml-ts-name (common:get-signature xml-ts-name) #f))
	 (xml-path		(if xml-dir
				    (conc xml-dir "/" keyname ".xml")
				    #f))

	 (test-data		(if xml-dir

  '(*TOP*
    (*PI* xml "version='1.0'")
    (testsuite)))

(define (set-item-env-vars itemdat)
  (for-each (lambda (item)
	      (debug:print 2 *default-log-port* "setenv " (car item) " " (cadr item))
	      (set-environment-variable! (car item) (cadr item)))
	    itemdat))

;; set up needed environment variables given a run-id and optionally a target, itempath etc.
;;
(define (runs:set-megatest-env-vars run-id #!key (inkeys #f)(inrunname #f)(inkeyvals #f)(intarget #f)(testname #f)(itempath #f))
  ;;(bb-check-path msg: "runs:set-megatest-env-vars entry")
  (let* ((target    (or intarget 
			(common:args-get-target)
			(get-environment-variable "MT_TARGET")))
	 (keys      (if inkeys    inkeys    (common:get-fields *configdat*) #;(rmt:get-keys)))
	 (keyvals   (if inkeyvals inkeyvals (keys:target->keyval keys target)))
	 (vals      (hash-table-ref/default *env-vars-by-run-id* run-id #f))
	 (link-tree (common:get-linktree))) ;; (configf:lookup *configdat* "setup" "linktree")))
    (if testname (set-environment-variable! "MT_TEST_NAME" testname))
    (if itempath (set-environment-variable! "MT_ITEMPATH"  itempath))

    ;; get the info from the db and put it in the cache
    (if link-tree
	(set-environment-variable! "MT_LINKTREE" link-tree)
	(debug:print-error 0 *default-log-port* "linktree not set, should be set in megatest.config in [setup] section."))
    (if (not vals)
	(let ((ht (make-hash-table)))
	  (hash-table-set! *env-vars-by-run-id* run-id ht)
	  (set! vals ht)
	  (for-each
	   (lambda (key)
	     (hash-table-set! vals (car key) (cadr key)))
	   keyvals)))
    ;; from the cached data set the vars
    
    (hash-table-for-each
     vals
     (lambda (key val)
       (debug:print 2 *default-log-port* "setenv " key " " val)
       (safe-setenv key val)))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 1")
    ;;(BB> "*env-vars-by-run-id*/runid("run-id" vals="(hash-table->alist vals))

    (if (not (get-environment-variable "MT_TARGET"))(set-environment-variable! "MT_TARGET" target))
    ;; we had a case where there was an exception generated by the hash-table-ref
    ;; due to *configdat* being #f Adding a handle and exit
    (let fatal-loop ((count 0)) 
      (handle-exceptions
	  exn
	  (let ((call-chain (get-call-chain))
		(msg        ((condition-property-accessor 'exn 'message) exn)))

              (thread-sleep! 2) ;; assuming nfs lag.
              (launch:setup force-reread: #t))
          (alist->env-vars (hash-table-ref/default *configdat* "env-override" '())))) ;;;; environment is tainted HERE in this let block.
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 2")
    ;; Lets use this as an opportunity to put MT_RUNNAME in the environment
    (let ((runname  (if inrunname inrunname (rmt:get-run-name-from-id run-id))))
      (if runname
	  (set-environment-variable! "MT_RUNNAME" runname)
	  (debug:print-error 0 *default-log-port* "no value for runname for id " run-id)))
    (set-environment-variable! "MT_RUN_AREA_HOME" *toppath*)
    ;; if a testname and itempath are available set the remaining appropriate variables
    (if testname (set-environment-variable! "MT_TEST_NAME" testname))
    (if itempath (set-environment-variable! "MT_ITEMPATH"  itempath))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 3")
    (if (and testname link-tree)
	(set-environment-variable! "MT_TEST_RUN_DIR" (conc (get-environment-variable "MT_LINKTREE")  "/"
					(get-environment-variable "MT_TARGET")    "/"
					(get-environment-variable "MT_RUNNAME")   "/"
					(get-environment-variable "MT_TEST_NAME")
					(if (and itempath
						 (not (equal? itempath "")))
					    (conc "/" itempath)
					    ""))))))

  
;; cache the runconfigs in $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME/.runconfig
;;
(define (full-runconfigs-read)
  ;; in the envprocessing branch the below code replaces the further below code
  ;;  (if (eq? *configstatus* 'fulldata)
  ;;      *runconfigdat*
  ;;      (begin
  ;;	(launch:setup)
  ;;	*runconfigdat*)))
  (let* ((rundir (if (and (get-environment-variable "MT_LINKTREE")(get-environment-variable "MT_TARGET")(get-environment-variable "MT_RUNNAME"))
		     (conc (get-environment-variable "MT_LINKTREE") "/" (get-environment-variable "MT_TARGET") "/" (get-environment-variable "MT_RUNNAME"))
		     #f))
	 (cfgf   (if rundir (conc rundir "/.runconfig." megatest-version "-" megatest-fossil-hash) #f)))
    (if (and cfgf
	     (file-exists? cfgf)
	     (file-writable? cfgf)
	     (common:use-cache?))
	(configf:read-alist cfgf) ;; use the cached file
	(let* ((keys   (common:get-fields *configdat*))     ;; (rmt:get-keys))
	       (target (common:args-get-target))
	       (key-vals (if target (keys:target->keyval keys target) #f))
	       (sections (if target (list "default" target) #f))
	       (data     (begin
			   (set-environment-variable! "MT_RUN_AREA_HOME" *toppath*)
			   (if key-vals
			       (for-each (lambda (kt)
					   (set-environment-variable! (car kt) (cadr kt)))
					 key-vals))
			   ;; (configf:read-config (conc *toppath* "/runconfigs.config") #f #t sections: sections))))
			   (runconfig:read (conc *toppath* "/runconfigs.config") target #f))))
	  (if (and rundir ;; have all needed variabless
		   (directory-exists? rundir)
		   (file-writable? rundir))
	      (begin