Megatest

(declare (uses megatestmod))
(declare (uses tasksmod))
(declare (uses subrunmod))
(declare (uses testsmod))
(declare (uses runsmod))
(declare (uses fsmod))

(use srfi-69)



(module launchmod
	*



(import scheme)
(cond-expand
 (chicken-4
  
  (import chicken
	  ports

	tasksmod
	subrunmod
	testsmod
	runsmod
	fsmod
	)

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

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

;; ezsteps were going to be coded as

		     #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)
	    )))))

;; 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    (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 (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))

    ;; get the info from the db and put it in the cache
    (if link-tree
	(setenv "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"))(setenv "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)))
	    (if (< count 5)
		(begin ;; this call is colliding, do some crude stuff to fix it.
		  (debug:print 0 *default-log-port* "ERROR: *configdat* was inaccessible! This should never happen. Retry #" count
			       ", exn=" exn)
		  (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.")
              ;;(BB> "ERROR: *configdat* was inaccessible! This should never happen.  Brute force reread.")
              (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
	  (setenv "MT_RUNNAME" runname)
	  (debug:print-error 0 *default-log-port* "no value for runname for id " run-id)))
    (setenv "MT_RUN_AREA_HOME" *toppath*)
    ;; if a testname and itempath are available set the remaining appropriate variables
    (if testname (setenv "MT_TEST_NAME" testname))
    (if itempath (setenv "MT_ITEMPATH"  itempath))
    ;;(bb-check-path msg: "runs:set-megatest-env-vars block 3")
    (if (and testname link-tree)
	(setenv "MT_TEST_RUN_DIR" (conc (getenv "MT_LINKTREE")  "/"
					(getenv "MT_TARGET")    "/"
					(getenv "MT_RUNNAME")   "/"
					(getenv "MT_TEST_NAME")
					(if (and itempath
						 (not (equal? itempath "")))
					    (conc "/" itempath)
					    ""))))))

;; TODO: deprecate me in favor of ezsteps.scm
;;
(define (ezsteps:runstep ezstep run-id test-id exit-info m tal testconfig all-steps-dat)
  (let* ((stepname       (car ezstep))  ;; do stuff to run the step
	 (stepinfo       (cadr ezstep))
	;; (let ((info (cadr ezstep)))

	 (html-file      (conc stepname ".html"))
	 (dat-file       (conc stepname ".dat"))
	 (tconfig-logpro (configf:lookup testconfig "logpro" stepname))
	 (logpro-used    (common:file-exists? logpro-file))
	 (mtexepath      (common:get-megatest-exe-path)))
    (setenv "MT_STEP_NAME" stepname)
    (hash-table-set! all-steps-dat stepname `((params . ,paramparts)))
    (debug:print 0 *default-log-port* "stepparts: " stepparts ", stepparams: " stepparams
                 ", paramparts: " paramparts ", subrun: " subrun ", stepcmd: " stepcmd)
    
    (if (and tconfig-logpro
	     (not logpro-used)) ;; no logpro file found but have a defn in the testconfig
	(begin
	  (with-output-to-file logpro-file
	    (lambda ()