Megatest

(declare (uses configfmod))
(declare (uses configf))
(declare (uses db))
(declare (uses rmtmod))
(declare (uses ezsteps))
;; (declare (uses dbmod))
(declare (uses dbfile))
(declare (uses dbmod))
(declare (uses mtargs))
(declare (uses mtmod))
(declare (uses megatestmod))
(declare (uses tasksmod))

(use regex regex-case base64 sqlite3 srfi-18 directory-utils posix posix-extras z3
     call-with-environment-variables csv hostinfo 
     typed-records pathname-expand matchable)

(import (prefix base64 base64:)
	(prefix sqlite3 sqlite3:)
	(prefix mtargs args:)
)

(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")
(include "megatest-fossil-hash.scm")

(import commonmod
	processmod
	configfmod
	rmtmod
	debugprint
	dbmod
	dbfile
	mtmod
	megatestmod
	tasksmod
	)

;;======================================================================
;; ezsteps
;;======================================================================

;; ezsteps were going to be coded as
;; stepname[,predstep1,predstep2 ...] [{VAR1=first,second,third}] command to execute

		     (read-symbolic-link (conc "/proc/" pid "/cwd"))
		     #f)))
    ;; now wait on that process if all is correct
    ;; periodically update the db with runtime
    ;; when the process exits look at the db, if still RUNNING after 10 seconds set
    ;; state/status appropriately
    (process-wait pid)))

;;======================================================================
;; Maintenance
;;======================================================================

(define (rmt:find-and-mark-incomplete run-id ovr-deadtime)
  (let* ((cfg-deadtime             (configf:lookup-number *configdat* "setup" "deadtime"))
	 (test-stats-update-period (configf:lookup-number *configdat* "setup" "test-stats-update-period")))
   (rmt:find-and-mark-incomplete-engine run-id ovr-deadtime cfg-deadtime test-stats-update-period)
   ;;call end of eud of run detection for posthook
   (launch:end-of-run-check run-id)))

;;  select end_time-now from
;;      (select testname,item_path,event_time+run_duration as
;;                          end_time,strftime('%s','now') as now from tests where state in
;;      ('RUNNING','REMOTEHOSTSTART','LAUNCHED'));
;;
;; NOT EASY TO MIGRATE TO db{file,mod}
;;
(define (rmt:find-and-mark-incomplete-engine run-id ovr-deadtime cfg-deadtime test-stats-update-period)
  (let* ((incompleted '())
	 (oldlaunched '())
	 (toplevels   '())
          ;; The default running-deadtime is 720 seconds = 12 minutes.
          ;; "(running-deadtime-default (+ server-start-allowance (* 2 launch-monitor-period)))" = 200 + (2 * (200 + 30 + 30))
         (deadtime-trim (or ovr-deadtime cfg-deadtime))
         (server-start-allowance 200)
         (server-overloaded-budget 200)
         (launch-monitor-off-time (or 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)

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

    (let* ((dat (rmt:get-toplevels-and-incompletes run-id running-deadtime remotehoststart-deadtime)))
      (set! oldlaunched (list-ref dat 1))
      (set! toplevels   (list-ref dat 2))
      (set! incompleted (list-ref dat 0)))

    (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-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* ((tinfo   (rmt:get-test-info-by-id 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 (rmt: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")
		       (rmt:set-state-status-and-roll-up-items
			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.
					  (commonmod: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.")
			   (begin
			     (debug:print 0 *default-log-port* "INFO: test " test-id
					  " final state/status is not COMPLETED/PASS. It is " result)
			     (rmt:set-state-status-and-roll-up-items
			      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)
	    )))))