Megatest

;; 	(mutex-lock! *rundb-mutex*)
;; 	(if (eq? mod-read 'mod)
;; 	    (dbr:dbstruct-mtime-set! dbstruct (current-milliseconds))
;; 	    (dbr:dbstruct-rtime-set! dbstruct (current-milliseconds)))
;; 	(dbr:dbstruct-inuse-set! dbstruct #f)
;; 	(mutex-unlock! *rundb-mutex*))))

(define-inline (db:generic-error-printout exn . message)
  (print-call-chain (current-error-port))
  (apply debug:print-error 0 *default-log-port* message)
  (debug:print-error 0 *default-log-port* "   params: " params
		     ", error: "     ((condition-property-accessor 'exn 'message)   exn)
		     ", arguments: " ((condition-property-accessor 'exn 'arguments) exn)
		     ", location: "  ((condition-property-accessor 'exn 'location)  exn)
		     ))

(print-call-chain (current-error-port))
;; (db:with-db dbstruct run-id sqlite3:exec "select blah fgrom blaz;")
;; r/w is a flag to indicate if the db is modified by this query #t = yes, #f = no
;;
(define (db:with-db dbstruct run-id r/w proc . params)
  (let* ((have-struct (dbr:dbstruct? dbstruct))
         (dbdat (if have-struct 
                    (db:get-db dbstruct)
                    #f))
	 (db    (if have-struct
		    (db:dbdat-get-db dbdat)
		    dbstruct))
	 (use-mutex (> *api-process-request-count* 25)))
         (dbdat     (if have-struct 
			(db:get-db dbstruct)
			#f))
	 (db        (if have-struct
			(db:dbdat-get-db dbdat)
			dbstruct))
	 (fname     (db:dbdat-get-path dbdat))
	 (use-mutex (> *api-process-request-count* 25))) ;; was 25
    (if (and use-mutex
	     (common:low-noise-print 120 "over-50-parallel-api-requests"))
	(debug:print-info 0 *default-log-port* *api-process-request-count* " parallel api requests being processed in process " (current-process-id) ", throttling access"))
    (if (common:low-noise-print 600 (conc "parallel-api-requests" *max-api-process-requests*))
	(debug:print-info 2 *default-log-port* "Parallel api request count: " *api-process-request-count* " max parallel requests: " *max-api-process-requests*))
    (condition-case
     (begin
    (handle-exceptions
     exn
     (begin
       (print-call-chain (current-error-port))
       (debug:print-error 0 *default-log-port* "sqlite3 issue in db:with-db, dbstruct=" dbstruct ", run-id=" run-id ", proc=" proc ", params=" params " error: " ((condition-property-accessor 'exn 'message) exn))
       ;; there is no recovering at this time. exit
       (exit 50))
     (if use-mutex (mutex-lock! *db-with-db-mutex*))
     (let ((res (apply proc db params)))
       (if use-mutex (mutex-unlock! *db-with-db-mutex*))
       ;; (if (vector? dbstruct)(db:done-with dbstruct run-id r/w))
       (if dbdat (stack-push! (dbr:dbstruct-dbstack dbstruct) dbdat))
       res))))

    ;;;;;;;;; (handle-exceptions
    ;;;;;;;;;  exn
    ;;;;;;;;;  (begin
    ;;;;;;;;;    (print-call-chain (current-error-port))
    ;;;;;;;;;    (debug:print-error 0 *default-log-port* "sqlite3 issue in db:with-db, dbstruct=" dbstruct ", run-id=" run-id ", proc=" proc ", params=" params " error: " ((condition-property-accessor 'exn 'message) exn))
    ;;;;;;;;;    ;; there is no recovering at this time. exit
    ;;;;;;;;;    (exit 50))
       (if use-mutex (mutex-lock! *db-with-db-mutex*))
       (let ((res (apply proc db params)))
	 (if use-mutex (mutex-unlock! *db-with-db-mutex*))
	 ;; (if (vector? dbstruct)(db:done-with dbstruct run-id r/w))
	 (if dbdat (stack-push! (dbr:dbstruct-dbstack dbstruct) dbdat))
	 res))
     (exn (io-error)
	  (db:generic-error-printout exn "ERROR: i/o error with " fname ". Check permissions, disk space etc. and try again."))
     (exn (corrupt)
	  (db:generic-error-printout exn "ERROR: database " fname " is corrupt. Repair it to proceed."))
     (exn (busy)
	  (db:generic-error-printout exn "ERROR: database " fname
				     " is locked. Try copying to another location, remove original and copy back."))
     (exn (permission)(db:generic-error-printout exn "ERROR: database " fname " has some permissions problem."))
     (exn ()
	  (db:generic-error-printout exn "ERROR: Unknown error with database " fname " message: "
		       ((condition-property-accessor 'exn 'message) exn))))))
      
;;======================================================================
;; K E E P   F I L E D B   I N   dbstruct
;;======================================================================

;; (define (db:get-filedb dbstruct run-id)
;;   (let ((db (vector-ref dbstruct 2)))
;;     (if db

         )
    (debug:print-info 4  *default-log-port* "running-deadtime = " running-deadtime)
    (debug:print-info 4  *default-log-port* "deadtime-trim = " deadtime-trim)

    (db:with-db 
     dbstruct #f #f
     (lambda (db)
       (let* ((stmth1 (db:get-cache-stmth
		       dbstruct 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');"))
	      (stmth2 (db:get-cache-stmth
		       dbstruct 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');"))
	      (stmth3 (db:get-cache-stmth
		       dbstruct db
		       "SELECT id,rundir,uname,testname,item_path FROM tests
                           WHERE run_id=? AND (strftime('%s','now') - event_time) > 86400
                                          AND state IN ('LAUNCHED');")))
       ;; 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) ;; default time 720 seconds

	 ;; 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))))
	  stmth1
	  run-id running-deadtime) ;; default time 720 seconds
       
       
       (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)))))
	 (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) ;; default time 230 seconds

       ;; 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")
	  stmth2
	  run-id remotehoststart-deadtime) ;; default time 230 seconds
	 
	 ;; 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)
	  stmth3
	  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.")
	 (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.
	 ;; These are defunct tests, do not do all the overhead of set-state-status. Force them to INCOMPLETE.
       ;;
       ;; (db:delay-if-busy dbdat)
	 ;; (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
	       ;; (launch:is-test-alive "localhost" 435)
               (debug:print 0 *default-log-port* "WARNING: Marking test(s); " (string-intersperse (map conc all-ids) ", ") " as DEAD")
               (for-each
	 (let* ((min-incompleted-ids (map car incompleted)) ;; do 'em all
		(all-ids             (append min-incompleted-ids (map car oldlaunched))))
	   (if (> (length all-ids) 0)
	       (begin
		 ;; (launch:is-test-alive "localhost" 435)
		 (debug:print 0 *default-log-port* "WARNING: Marking test(s); " (string-intersperse (map conc all-ids) ", ")
			      " as DEAD")
		 (for-each
                  (lambda (test-id)
                    (let* (;; (run-dir (db:test-get-rundir-from-test-id dbstruct run-id test-id))
			   (tinfo   (db:get-test-info-by-id dbstruct run-id test-id))
			   (run-dir (db:test-get-rundir     tinfo))
			   (host    (db:test-get-host       tinfo))
			   (pid     (db:test-get-process_id tinfo))
			   (result (db:get-status-from-final-status-file run-dir)))
		      (if (and (list? result) (> (length result) 1) (equal? "PASS" (cadr result)) (equal? "COMPLETED" (car result))) 
			  (begin
			    (debug:print 0 *default-log-port* "INFO: test " test-id " actually passed, so marking PASS not DEAD")
			    (db:set-state-status-and-roll-up-items dbstruct run-id test-id 'foo "COMPLETED" "PASS"
								   "Test stopped responding but it has PASSED; marking it PASS in the DB."))
			  (let ((is-alive (launch:is-test-alive host pid)))
			    (db:set-state-status-and-roll-up-items
			     dbstruct run-id test-id 'foo "COMPLETED" "PASS"
			     "Test stopped responding but it has PASSED; marking it PASS in the DB."))
			  (let ((is-alive (and (not (eq? pid 0))  ;; 0 is default in re-used field "attemptnum" where pid stored.
					       (launch:is-test-alive host pid))))
			    (if is-alive
				(debug:print 0 *default-log-port* "INFO: test " test-id " on host " host " has a process on pid " pid ", NOT setting to DEAD.")
				(debug:print 0 *default-log-port* "INFO: test " test-id " on host " host
					     " has a process on pid " pid ", NOT setting to DEAD.")
				(begin
				  (debug:print 0 *default-log-port* "INFO: test " test-id " final state/status is not COMPLETED/PASS. It is " result)
				  (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.")))))))
                     ;; call end of eud of run detection for posthook - from merge, is it needed?
                     ;; (launch:end-of-run-check run-id)
				  (debug:print 0 *default-log-port* "INFO: test " test-id
					       " final state/status is not COMPLETED/PASS. It is " result)
				  (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.")))))))
		  ;; call end of eud of run detection for posthook - from merge, is it needed?
		  ;; (launch:end-of-run-check run-id)
		  all-ids)
	       ;;call end of eud of run detection for posthook
	       (launch:end-of-run-check run-id)
	       )))))))
		 ;;call end of eud of run detection for posthook
		 (launch:end-of-run-check run-id)
		 )))))))


;; ALL REPLACED BY THE BLOCK ABOVE
;;
;; 	    (sqlite3:execute 
;; 	     db
;; 	     (conc "UPDATE tests SET state='INCOMPLETE' WHERE run_id=? AND id IN (" 
;; 		   (string-intersperse (map conc all-ids) ",")
;; 		   ");")
;;              run-id))))
;; 
;;     ;; Now do rollups for the toplevel tests
;;     ;;
;;     ;; (db:delay-if-busy dbdat)
;;     (for-each
;;      (lambda (toptest)
;;        (let ((test-name (list-ref toptest 3)))
;; ;;	     (run-id    (list-ref toptest 5)))
;; 	 (db:top-test-set-per-pf-counts dbstruct run-id test-name)))
;;      toplevels)))

;; BUG: Probably broken - does not explicitly use run-id in the query
;;
(define (db:top-test-set-per-pf-counts dbstruct run-id test-name)
  (db:general-call dbstruct 'top-test-set-per-pf-counts (list test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name)))

;; Clean out old junk and vacuum the database

     #f
     (lambda (db)
       (let* ((stmth (db:get-cache-stmth dbstruct db qry)))
	 (sqlite3:first-result stmth run-id))))))

;; For a given testname how many items are running? Used to determine
;; probability for regenerating html
;; 
;;
(define (db:get-count-tests-running-for-testname dbstruct run-id testname)
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (db)
     (let* ((stmt "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id=? AND NOT (uname = 'n/a' AND item_path = '') AND testname=?;")