Megatest

;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;======================================================================
;; launch a task - this runs on the originating host, tests themselves
;;
;;======================================================================

(declare (unit launch))
(declare (uses subrun))
(declare (uses common))
(declare (uses debugprint))
(declare (uses commonmod))
(declare (uses configf))
(declare (uses db))
(declare (uses rmtmod))
(declare (uses ezsteps))
;; (declare (uses dbmod))
(declare (uses dbfile))
(declare (uses mtargs))

(use regex regex-case base64 sqlite3 srfi-18 directory-utils posix-extras z3
     call-with-environment-variables csv)
(use 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
	rmtmod
	debugprint
	;; dbmod
	dbfile)

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

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

		   (setenv "MT_STEP_NAMES" (string-intersperse all-step-names " "))
		   (let loop ((ezstep (car ezstepslst))
			     (tal    (cdr ezstepslst))
			     (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 (ezsteps: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)))
			  (setenv "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))

			(debug:print 0 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))
                        (file-close status-file)
                        )

                        ))))))

(define (launch:monitor-job run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
  (let* ((update-period (string->number (or (configf:lookup *configdat* "setup" "test-stats-update-period") "60")))
         (start-seconds (current-seconds))
	 (calc-minutes  (lambda ()
			  (inexact->exact 
			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))

                                   (delta (abs (- df disk-free))))
                              (if (and (> df 0)
                                       (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                  df
                                  #f)))
             (do-sync       (or new-cpu-load new-disk-free over-time))

             (test-info   (rmt:get-test-state-status-by-id run-id test-id))
             (state       (car test-info));; (db:test-get-state test-info))
             (status      (cdr test-info));; (db:test-get-status test-info))
             (kill-reason  "no kill reason specified")
             (kill-job?    #f))
        ;; (common:telemetry-log "zombie" (conc "launch:monitor-job - decision time encountered at "(current-seconds)" with last-sync="last-sync" do-sync="do-sync" over-time="over-time" update-period="update-period))
        (cond
         ((test-get-kill-request run-id test-id)
          (set! kill-reason "KILLING TEST since received kill request (KILLREQ)")
          (set! kill-job? #t))
         ((and runtlim (> (- (current-seconds) start-seconds) runtlim))
          (set! kill-reason (conc "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" (- (current-seconds) start-seconds) " seconds, limit=" runtlim))
          (set! kill-job? #t))
         ((equal? status "DEAD")
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f)
          (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)
        (if (common:low-noise-print 600 "run zombie") ;; every five minutes is plenty
	    (launch:handle-zombie-tests run-id))
        (when do-sync
          ;;(with-output-to-file (conc (getenv "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)))
	  )

		      (tests:test-set-status! run-id test-id "KILLED"  "KILLED" (conc (args:get-arg "-m")" "kill-reason) #f)) ;; BB ADDED kill-reason -- confirm OK with Matt
		    (begin
		      (debug:print-error 0 *default-log-port* "Nothing to kill, pid1=" pid1 ", pid2=" pid2)
		      (tests:test-set-status! run-id test-id "KILLED"  "FAILED TO KILL" (conc (args:get-arg "-m")" "kill-reason) #f) ;; BB ADDED kill-reason -- confirm OK with Matt
		      )))
	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now. But run end of run before exiting?
              (launch:end-of-run-check run-id)
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes)
                        (or new-cpu-load cpu-load)

	      (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-testsuite-name))))
          ;;(bb-check-path msg: "launch:execute post block 3")

	  (let ((tmppath (getenv "PATH")))
	    (if (string-search tmppath " ")
		(debug:print 0 *default-log-port* "WARNING: spaces in PATH are not supported."))
	    (if mt-bindir-path (setenv "PATH" (conc tmppath":"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.

		    ;; NO NEED TO CALL set-state-status-and-roll-up-items HERE, THIS IS DONE IN set-state-status-and-roll-up-items called by tests:test-set-status!
		 )
              )


	      ;; for automated creation of the rollup html file this is a good place...
	      (if (not (equal? item-path ""))
		  (tests:summarize-items run-id test-id test-name #f))
	      ;; BUG was this meant to be the antecnt of the if above?
	      ;; BUG was this meant to be the antecnt of the if above?
	      (tests:summarize-test run-id test-id)  ;; don't force - just update if no
              ;; Leave a .final-status file for the top level test
              (tests:save-final-status run-id test-id)
	      (rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id))) ;; end of let*

	    (mutex-unlock! m)
            (launch:end-of-run-check run-id )

;; if dead safe to mark the test as killed in the db
;; State/status table
;; new
;; 100% COMPLETED/ (PASS,FAIL,ABORT etc.) ==> COMPLETED / X where X is same as itemized rollup
;; > 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 *last-rollup* 0)
(define (launch: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 (conc "lunch-complete-" run-id)))
	   (current-state-status (rmt:get-run-state-status run-id))
           (current-state        (car current-state-status))  ;; (rmt:get-run-state run-id))
           (current-status       (cdr current-state-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)
      ;;
      ;; TODO: add a final rollup when run is done (if there isn't one already)
      ;;
      (if (or (< running-cnt 3)                              ;; have only few running
	      (> (- (current-seconds) *last-rollup*) 10))    ;; or haven't rolled up in past ten seconds
	  (begin
	    (rmt:set-state-status-and-roll-up-run  run-id current-state current-status)
	    (set! *last-rollup* (current-seconds))))
     (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 (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 (conc "end-of-run-" run-id)))
                (debug:print 0 *default-log-port* "End of Run Detected.")

			     (tal    (cdr not-completed-tests)))
		       (let* ((test-name (vector-ref running-test 2))
                 (item-path (vector-ref running-test 11)))
			       	(debug:print 0 *default-log-port* "test " test-name "/" item-path " not completed")
              (if (not (null? tal))
		  (loop (car tal) (cdr tal)))))))))))

































(define (launch:kill-tests-if-dead run-id)
  (let* ((running-tests (rmt:get-tests-for-run run-id "%" `("RUNNING" "LAUNCHED" "REMOTEHOSTSTART") `() #f #f #f #f #f #f #f #f)))
       (let loop ((running-test (car running-tests))
			     (tal    (cdr running-tests))
			     (kill-cnt 0))
		       (let* ((test-name (vector-ref running-test 2))
                 (item-path (vector-ref running-test 11))
		 (test-id (vector-ref running-test 0))
                 (host (vector-ref running-test 6))
                 (pid  (rmt:test-get-top-process-pid run-id test-id))   
                 (event-time (vector-ref running-test 5))
                 (duration (vector-ref running-test 12))
                 (flag 0)   
                 (curr-time (current-seconds)))
       (if (and (< (+ event-time duration 600) curr-time) (not (commonmod:is-test-alive host pid))) ;;test has not updated duration in last 10 min then likely its not running but confirm before marking it as killed
           (begin    
			       	(debug:print 0 *default-log-port* "test " test-name "/" item-path " needs to be killed")
              (set! flag 1) 
              (rmt:set-state-status-and-roll-up-items run-id test-name item-path "KILLREQ" "n/a" #f)))
               (if (not (null? tal))
				  (loop (car tal) (cdr tal) (+ kill-cnt flag))
                 (+ kill-cnt flag))))))

	    (begin
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (setenv "MT_TESTSUITENAME" (common:get-testsuite-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))

	;; needed by various transport and db modules
	(dbfile:testsuite-name (common:get-testsuite-name)) ;; (get-testsuite-name *toppath* *configdat*))

        ;; one more attempt to cache the configs for future reading
        (let* ((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: "...somepath.../.runconfigs.cfg-1.6427-7d1e789cb3f62f9cde719a4865bb51b3c17ea853" - ticket 220546342
          ;; TODO - consider 1) using simple-lock to bracket cache write

              (set! *configstatus* 'fulldata)))

	;; if have -append-config then read and append here
	(let ((cfname (args:get-arg "-append-config")))
	  (if (and cfname
		   (file-read-access? cfname))
	      (read-config cfname *configdat* #t))) ;; values are added to the hash, no need to do anything special.

	;; have config at this time, this is a good place to set params based on config file settings
	(let*  ((dbmode   (configf:lookup *configdat* "setup" "dbcache-mode"))
		(syncmode (configf:lookup *configdat* "setup" "sync-mode"))
		(srvdebug (configf:lookup *configdat* "server" "debug-parameter")))
	  (if dbmode
	      (begin
		(debug:print-info 0 *default-log-port* "Overriding dbmode to "dbmode)
		(dbcache-mode (string->symbol dbmode))))
	  (if syncmode
	      (begin
		(debug:print-info 0 *default-log-port* "Overriding syncmode to "syncmode)
		(dbfile:sync-method (string->symbol syncmode))))
	  (if srvdebug
	      (begin
		(debug:print-info 0 *default-log-port* "Overriding server debug parameter to "srvdebug)
		(tt-server-profile-string srvdebug)))
	  )
		
	*toppath*)))


(define (get-best-disk confdat testconfig)
  (let* ((disks   (or (and testconfig (hash-table-ref/default testconfig "disks" #f))
		      (hash-table-ref/default confdat "disks" #f)))
	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))

	  ;; (rmt:general-call 'test-set-rundir run-id lnkpath testname "") ;; toptest-path)
	  (if (or (not curr-test-path)
		  (not (directory-exists? toptest-path)))
	      (begin
		(debug:print-info 2 *default-log-port* "Creating " toptest-path " and link " lnkpath)
		(handle-exceptions
		    exn
		  (if (directory-exists? toptest-path) ;; it was likely created in parallel
		      #t
		      (begin
			(debug:print-info 0 *default-log-port* "failed to create directory " toptest-path ", exn=" exn)
			#f))
		 (create-directory toptest-path #t))
		(hash-table-set! *toptest-paths* testname toptest-path)))))

    ;; The toptest path has been created, the link to the test in the linktree has
    ;; been created. Now, if this is an iterated test the real test dir must be created
    (if (not not-iterated) ;; this is an iterated test
	(begin ;; (let ((lnktarget (conc lnkpath "/" item-path)))
	  (debug:print 2 *default-log-port* "Setting up sub test run area")
	  (debug:print 2 *default-log-port* " - creating run area in " test-path)
	  (handle-exceptions
	   exn
	   (if (directory-exists? test-path)
	       #t
	       (begin
		 (debug:print-error 0 *default-log-port* " Failed to create directory " test-path ((condition-property-accessor 'exn 'message) exn)
				    ", exiting, exn=" exn)
		 (exit 1)))
	   (create-directory test-path #t))
	  (debug:print 2 *default-log-port* 
		       " - creating link from: " test-path "\n"
		       "                   to: " lnktarget)

	  ;; If there is already a symlink delete it and recreate it.
	  (handle-exceptions

;; 2. create run dir on disk, path name is meaningful
;; 3. create link from run dir to megatest runs area 
;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat params)
  (assert runname "FATAL: launch-test called with no runname")
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ( ;; (lock-key        (conc "test-" test-id))
	;; (got-lock        (let loop ((lock        (rmt:no-sync-get-lock lock-key))
	;; 			     (expire-time (+ (current-seconds) 15))) ;; give up on getting the lock and steal it after 15 seconds
	;; 		    (if (car lock)
	;; 			#t
	;; 			(if (> (current-seconds) expire-time)

      ;; the following call handles waiver propogation. cannot yet condense into roll-up-pass-fail
      (tests:test-set-status! run-id test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
      (rmt:set-state-status-and-roll-up-items run-id test-name item-path #f "LAUNCHED" #f)
      ;; (pp (hash-table->alist tconfig))
      (set! diskpath (get-best-disk *configdat* tconfig))
      (debug:print 2 *default-log-port* "best disk path = " diskpath)
      (if diskpath
	  (let ((dat (create-work-area run-id run-info keyvals test-id test-path diskpath test-name itemdat)))
	    (set! work-area (car dat))
	    (set! toptest-work-area (cadr dat))
	    (debug:print-info 2 *default-log-port* "Using work area " work-area))
	  (begin
	    (set! work-area (conc test-path "/tmp_run"))
	    (create-directory work-area #t)
	    (debug:print 0 *default-log-port* "WARNING: No disk work area specified - running in the test directory under tmp_run")))
      (set! cmdparms (base64:base64-encode 
		      (z3:encode-buffer 
		       (with-output-to-string
			 (lambda () ;; (list 'hosts     hosts)
			   (write (list (list 'testpath  test-path)
					;; (list 'transport (conc *transport-type*))
					;; (list 'serverinf *server-info*)
					#;(list 'homehost  (let* ((hhdat (server:get-homehost)))
							   (if hhdat
							       (car hhdat)
							       #f)))
					#;(list 'serverurl (if *runremote* ;; would like to add this back ... WORK NEEDED
							     (remote-server-url *runremote*)
							     #f)) ;;
					(list 'areaname  (common:get-testsuite-name))
					(list 'toppath   *toppath*)
					(list 'work-area work-area)
					(list 'test-name test-name) 
					(list 'runscript runscript)