Megatest

   ;;  (debug:print 0 *default-log-port* "ERROR: attempt to open database when not on homehost. Exiting. Homehost: " (common:get-homehost))
   ;;  (exit 1))))

;; Open the classic megatest.db file (defaults to open in toppath)
;;
;;   NOTE: returns a dbdat not a dbstruct!
;;

;;(define (db:reopen-megatest-db

(define (db:open-megatest-db #!key (path #f)(name #f))
  (let* ((dbdir        (or path *toppath*))
         (dbpath       (conc  dbdir "/" (or name "megatest.db")))
	 (dbexists     (common:file-exists? dbpath))
	 (db           (db:lock-create-open dbpath
					    (lambda (db)
                                              (db:initialize-main-db db)

;;                          end_time,strftime('%s','now') as now from tests where state in
;;      ('RUNNING','REMOTEHOSTSTART','LAUNCHED'));

(define (db:find-and-mark-incomplete dbstruct run-id ovr-deadtime)
  (let* ((incompleted '())
	 (oldlaunched '())
	 (toplevels   '())
         (deadtime-trim (or ovr-deadtime (configf:lookup-number *configdat* "setup" "deadtime")))
         (server-start-allowance 200)
         (server-overloaded-budget 200)
         (launch-monitor-off-time (or (configf:lookup-number *configdat* "setup" "test-stats-update-period") 30))
         (launch-monitor-on-time-budget 30)
         (launch-monitor-period (+ launch-monitor-off-time launch-monitor-on-time-budget server-overloaded-budget))
         (remotehoststart-deadtime-default (+ server-start-allowance server-overloaded-budget 30))
         (remotehoststart-deadtime (or deadtime-trim remotehoststart-deadtime-default))
         (running-deadtime-default (+ server-start-allowance (* 2 launch-monitor-period)))
         (running-deadtime (or deadtime-trim running-deadtime-default)) ;; two minutes (30 seconds between updates, this leaves 3x grace period)
         )
    (db:with-db 
     dbstruct #f #f
     (lambda (db)


       ;; in RUNNING or REMOTEHOSTSTART for more than 10 minutes
       ;;
       ;; HOWEVER: this code in run:test seems to work fine
       ;;              (> (- (current-seconds)(+ (db:test-get-event_time testdat)
       ;;                     (db:test-get-run_duration testdat)))
       ;;                    600) 
       ;; (db:delay-if-busy dbdat)
       (sqlite3:for-each-row 
        (lambda (test-id run-dir uname testname item-path event-time run-duration)
          (if (and (equal? uname "n/a")
                   (equal? item-path "")) ;; this is a toplevel test
              ;; what to do with toplevel? call rollup?
              (begin
                (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
                (debug:print-info 0 *default-log-port* "Found old toplevel test in RUNNING state, test-id=" test-id))
              (begin
                (set! incompleted (cons (list test-id run-dir uname testname item-path run-id) incompleted))
                (debug:print-info 0 *default-log-port* "Found old test in RUNNING state, test-id=" test-id" exceeded running-deadtime "running-deadtime" now="(current-seconds)" event-time="event-time" run-duration="run-duration))))
        db
        "SELECT id,rundir,uname,testname,item_path,event_time,run_duration FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?) AND state IN ('RUNNING');"
        run-id running-deadtime)

       
       (sqlite3:for-each-row 
        (lambda (test-id run-dir uname testname item-path event-time run-duration)
          (if (and (equal? uname "n/a")
                   (equal? item-path "")) ;; this is a toplevel test
              ;; what to do with toplevel? call rollup?
              (begin
                (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
                (debug:print-info 0 *default-log-port* "Found old toplevel test in RUNNING state, test-id=" test-id))
              (begin
                (debug:print-info 0 *default-log-port* "Found old test in REMOTEHOSTSTART state, test-id=" test-id" exceeded running-deadtime "running-deadtime" now="(current-seconds)" event-time="event-time" run-duration="run-duration)
                (set! incompleted (cons (list test-id run-dir uname testname item-path run-id) incompleted)))))
        db
        "SELECT id,rundir,uname,testname,item_path,event_time,run_duration FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?) AND state IN ('REMOTEHOSTSTART');"
        run-id remotehoststart-deadtime)

       ;; in LAUNCHED for more than one day. Could be long due to job queues TODO/BUG: Need override for this in config
       ;;
       ;; (db:delay-if-busy dbdat)
       (sqlite3:for-each-row
        (lambda (test-id run-dir uname testname item-path)
          (if (and (equal? uname "n/a")
                   (equal? item-path "")) ;; this is a toplevel test
              ;; what to do with toplevel? call rollup?
              (set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
              (begin
                (debug:print-info 0 *default-log-port* "Found old test in LAUNCHED state, test-id=" test-id" 1 day since event_time marked")
                (set! oldlaunched (cons (list test-id run-dir uname testname item-path run-id) oldlaunched)))))
        db
        "SELECT id,rundir,uname,testname,item_path FROM tests WHERE run_id=? AND (strftime('%s','now') - event_time) > 86400 AND state IN ('LAUNCHED');"
        run-id)
       
       (debug:print-info 18 *default-log-port* "Found " (length oldlaunched) " old LAUNCHED items, " (length toplevels) " old LAUNCHED toplevel tests and " (length incompleted) " tests marked RUNNING but apparently dead.")

       ;; These are defunct tests, do not do all the overhead of set-state-status. Force them to INCOMPLETE.
       ;;
       ;; (db:delay-if-busy dbdat)
       (let* (;; (min-incompleted (filter (lambda (x)
              ;;      		      (let* ((testpath (cadr x))
              ;;      			     (tdatpath (conc testpath "/testdat.db"))
              ;;      			     (dbexists (common:file-exists? tdatpath)))
              ;;      			(or (not dbexists) ;; if no file then something wrong - mark as incomplete
              ;;      			    (> (- (current-seconds)(file-modification-time tdatpath)) 600)))) ;; no change in 10 minutes to testdat.db - she's dead Jim
              ;;      		    incompleted))
              (min-incompleted-ids (map car incompleted)) ;; do 'em all
              (all-ids             (append min-incompleted-ids (map car oldlaunched))))
         (if (> (length all-ids) 0)
             (begin
               (debug:print 0 *default-log-port* "WARNING: Marking test(s); " (string-intersperse (map conc all-ids) ", ") " as DEAD")
               (for-each
                (lambda (test-id)
                  (db:set-state-status-and-roll-up-items dbstruct run-id test-id 'foo "COMPLETED" "DEAD" "Test stopped responding while in RUNNING or REMOTEHOSTSTART; presumed dead."))
                  ;;(db:test-set-state-status dbstruct run-id test-id "COMPLETED" "DEAD" "Test stopped responding while in RUNNING or REMOTEHOSTSTART; presumed dead.")) ;; fix for one aspect of Randy's ticket 1405717332 ;; TODO - fix problem where test goes to COMPLETED/DEAD while in progress, only later to go to COMPLETED/PASS.  ref ticket 220546828
                all-ids))))))))

;; ALL REPLACED BY THE BLOCK ABOVE
;;
;; 	    (sqlite3:execute 
;; 	     db
;; 	     (conc "UPDATE tests SET state='INCOMPLETE' WHERE run_id=? AND id IN (" 

			   (db:test-get-testname testdat)
			   test-name))
	 (item-path    (db:test-get-item-path testdat))
         (tl-testdat   (db:get-test-info dbstruct run-id test-name ""))
         (tl-test-id   (if tl-testdat
			   (db:test-get-id tl-testdat)
			   #f)))
    (if (member state '("LAUNCHED" "REMOTEHOSTSTART")) 
	(db:general-call dbstruct 'set-test-start-time (list test-id)))
    (mutex-lock! *db-transaction-mutex*)
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (let ((tr-res
              (sqlite3:with-transaction