Megatest

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

TODO
====

WW38
. Add test_rundat to no-sync ==> correction, put in <testdir>/.meta/test-run.dat
. Add STATE/STATUS transitions to .meta/test-run.dat or similar
. Swizzle update-test-rundat to operate on no-sync
. Swizzle update-run-duration, -uname-host and cpuload-diskfree to no-sync
. On state/status change update tests table with duration

WW15
. fill newview matrix with data, filter pipeline gui elements
. improve [script], especially indent handling

WW16
. split db into megatest.db (runs etc.) db/<something>.db
. release basic newview implementation

WW18
. release split db implementation
. mtutil calls from dashboard (for remote control)
. logs browser (esp. for surfacing mtutil related activities)

WW19
. break command line into sections; all, run control, queries, utilities etc.
. pull in ftfplan (not integrated, just code pulled in)

WW20
. ./configure => ubuntu, sles11, sles12, rh7
. Jenkins junit XML support
. Add output flushing in teamcity support
. Switch to using simple runs query everywhere
. Add end_time to runs and add a rollup call that sets state, status and end_time

Future
. Switch to scsh-process pipeline management for job execution/control
. Use call-with-environment-variables more.

				   (lambda (a b)
				     (< (hash-table-ref all-files a)(hash-table-ref all-files b))))
			     (- num-logs max-allowed))))
	    (for-each
	     (lambda (file)
	       (let* ((fullname (conc "logs/" file)))
		 (if (directory? fullname)
		     (debug:print-error 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
		     (debug:print-info 0 *default-log-port* fullname " in logs directory is a directory! Cannot rotate it, it is best to not put subdirectories in the logs dir.")
		     (handle-exceptions
		      exn
		      (debug:print-error 0 *default-log-port* "failed to remove " fullname ", exn=" exn)
		      (debug:print-info 0 *default-log-port* "failed to remove " fullname ", exn=" exn)
		      (delete-file* fullname)))))
	     files)
	    (debug:print-info 0 *default-log-port* "Deleted " (length files) " files from logs, keeping " max-allowed " files."))))))
  
;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
;; Do NOT check if not on homehost!
;;

;;
(define (dboard:common-set-tabdat! commondat tabnum tabdat)
  (hash-table-set!
   (dboard:commondat-tabdats commondat)
   tabnum
   tabdat))

(define *updater-running* #f) ;; move this into one of the stucts
;; gets and calls updater list based on curr-tab-num
;;
(define (dboard:common-run-curr-updaters commondat #!key (tab-num #f))
  (if (dboard:common-get-tabdat commondat tab-num: tab-num) ;; only update if there is a tabdat
      (let* ((tnum     (or tab-num (dboard:commondat-curr-tab-num commondat)))
	     (updaters (hash-table-ref/default (dboard:commondat-updaters commondat)
					       tnum

;;
;;    NOTE: Yes, this is used
;;
(define (dboard:get-tests-for-run-duplicate tabdat run-id run testnamepatt key-vals)
  (let* ((start-time   (current-seconds))
	 (access-mode  (dboard:tabdat-access-mode tabdat))
         (num-to-get   (string->number (or (configf:lookup *configdat* "setup" "num-tests-to-get")
                                           "200")))
                                           "50"))) ;; was 200, which is fine in a normal run area.
	 (states       (hash-table-keys (dboard:tabdat-state-ignore-hash tabdat)))
	 (statuses     (hash-table-keys (dboard:tabdat-status-ignore-hash tabdat)))
         (do-not-use-db-file-timestamps #t) ;; (configf:lookup *configdat* "setup" "do-not-use-db-file-timestamps")) ;; this still hosts runs-summary-tab
         (do-not-use-query-timestamps   #t) ;; (configf:lookup *configdat* "setup" "do-not-use-query-timestamps")) ;; this no longer troubles runs-summary-tab
	 (sort-info    (get-curr-sort))
	 (sort-by      (vector-ref sort-info 1))
	 (sort-order   (vector-ref sort-info 2))

		   (elapsed-time (- (current-seconds) start-time)))
	      (if (null? all-test-ids)
		  (hash-table-delete! (dboard:tabdat-allruns-by-id tabdat) run-id)
		  (hash-table-set!    (dboard:tabdat-allruns-by-id tabdat) run-id run-struct))
	      (if (or (null? tal)
		      (> elapsed-time 2)) ;; stop loading data after 5 seconds, on the next call more data *should* be loaded since get-tests-for-run uses last update
		  (begin
		    (when (> elapsed-time 2)   
		    #;(when (> elapsed-time 2)   
                      (debug:print 0 *default-log-port* "NOTE: updates are taking a long time, " elapsed-time "s elapsed.")
                      (let* ((old-val (iup:attribute *tim* "TIME"))
                             (new-val (number->string (inexact->exact (floor (* 2  (string->number old-val)))))))
                        (if (< (string->number new-val) 5000)
                            (begin
                            ((debug:print 0 *default-log-port* "NOTE: increasing poll interval from "old-val" to "new-val)
                            (iup:attribute-set! *tim* "TIME" new-val))))
			      (debug:print 0 *default-log-port* "NOTE: increasing poll interval from "old-val" to "new-val)
			      (iup:attribute-set! *tim* "TIME" new-val)))))


                      )
		    (dboard:tabdat-allruns-set! tabdat new-res)
		    maxtests)
		  (if (> (dboard:rundat-run-data-offset run-struct) 0)
		      (loop run tal new-res newmaxtests) ;; not done getting data for this run
		      (loop (car tal)(cdr tal) new-res newmaxtests)))))))
    (dboard:tabdat-filters-changed-set! tabdat #f)
    (dboard:update-tree tabdat runs-hash header tb)))

            sort-lb))) 
	)

        ;; insert extra widget here
        (if extra-widget
            extra-widget
            (iup:hbox)) ;; empty widget

        

        
        )))

     (let* ((status-toggles (map (lambda (status)
				   (iup:toggle (conc status)
					       #:fontsize 8 ;; btn-fontsz ;; "10"
					       ;; #:expand "HORIZONTAL"
					       #:action   (lambda (obj val)

	 commondat 
	 (lambda ()
	   (dashboard:runs-tab-updater commondat 1))
	 tab-num: 2)
	(iup:callback-set! *tim*
			   "ACTION_CB"
			   (lambda (time-obj)
			     (let ((update-is-running #f))
			     (mutex-lock! (dboard:commondat-update-mutex commondat))
			     (set! update-is-running (dboard:commondat-updating commondat))
			     (if (not update-is-running)
			     (dboard:commondat-updating-set! commondat #t))
			     (mutex-unlock! (dboard:commondat-update-mutex commondat))
			     (if (not update-is-running) ;; we know that the update was not running and we now have a lock on doing an update
			     (begin
			     (if (not *updater-running*)
				 (begin
				   ;; (mutex-lock! (dboard:commondat-update-mutex commondat))
				   ;; (set! update-is-running (dboard:commondat-updating commondat))
				   ;;(if (not update-is-running)
				   ;;	   (dboard:commondat-updating-set! commondat #t))
				   ;;(mutex-unlock! (dboard:commondat-update-mutex commondat))
				   ;;(if (not update-is-running) ;; we know that the update was not running and we now have a lock on doing an update
				   ;;	   (begin
				   (set! *updater-running* #t)
			     (dboard:common-run-curr-updaters commondat) ;; (dashboard:run-update commondat)
				   (set! *updater-running* #f)
			     (mutex-lock! (dboard:commondat-update-mutex commondat))
			     (dboard:commondat-updating-set! commondat #f)
			     (mutex-unlock! (dboard:commondat-update-mutex commondat)))
				   ;; (mutex-lock! (dboard:commondat-update-mutex commondat))
				   ;;  (dboard:commondat-updating-set! commondat #f)
				   ;;   (mutex-unlock! (dboard:commondat-update-mutex commondat)))
				   ))
			     1))))
      
      (let ((th1 (make-thread (lambda ()
				(thread-sleep! 1)
				(dboard:common-run-curr-updaters commondat 0) ;; force update of summary tab 
				) "update buttons once"))

                              id           INTEGER PRIMARY KEY,
                              test_id      INTEGER,
                              state        TEXT DEFAULT 'new',
                              status       TEXT DEFAULT 'n/a',
                              archive_type TEXT DEFAULT 'bup',
                              du           INTEGER,
                              archive_path TEXT);")))
         (print "creating trigges from init") 
        (db:create-triggers db)    
     db)) ;; )
    (db:create-triggers db)    
    db)) ;; )

;;======================================================================
;; A R C H I V E S
;;======================================================================

;; dneeded is minimum space needed, scan for existing archives that 
;; are on disks with adequate space and already have this test/itempath

  (let* ((run-ids (db:get-all-run-ids mtdb)))
    (for-each 
     (lambda (run-id)
       (let ((testrecs (db:get-all-tests-info-by-run-id mtdb run-id)))
	 (db:prep-megatest.db-adj-test-ids (db:dbdat-get-db mtdb) run-id testrecs)))
     run-ids)))

;; Get test data using test_id, run-id is not used
;; Get test data using test_id, run-id is not used - but it will be!
;; 
(define (db:get-test-info-by-id dbstruct run-id test-id)
  (db:with-db
   dbstruct
   #f ;; run-id
   #f
   (lambda (db)

			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10 update-db: #t)

    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory)))
               (last-sync (current-seconds)))
      (common:telemetry-log "zombie" (conc "launch:monitor-job - top of loop encountered at "(current-seconds)" with last-sync="last-sync))
      ;; (common:telemetry-log "zombie" (conc "launch:monitor-job -
      ;; top of loop encountered at "(current-seconds)" with
      ;; last-sync="last-sync))
      (let* ((over-time     (> (current-seconds) (+ last-sync update-period)))
             (new-cpu-load  (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
                                   (delta (abs (- load cpu-load))))
                              (if (> delta 0.1) ;; don't bother updating with small changes
                                  load
                                  #f)))
             (new-disk-free (let* ((df    (if over-time ;; only get df every 30 seconds

             (do-sync       (or new-cpu-load new-disk-free over-time))

             (test-info   (rmt:get-test-info-by-id run-id test-id))
             (state       (db:test-get-state test-info))
             (status      (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))
        #;(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)
          (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f update-db: #t)
          (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)
        (launch:handle-zombie-tests run-id)
        (when do-sync
        (if do-sync ;; save meta data about the running of this test
          ;;(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)
	    (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))))
        
	(if kill-job? 
	    (begin
              (debug:print-info 0 *default-log-port* "proceeding to kill test: "kill-reason)
	      (mutex-lock! m)
	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
	      ;;       between tries?

	    (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)
                        (or new-disk-free disk-free)
                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f update-db: #t))) ;; NOTE: Checking twice for keep-going is intentional


(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)
    ;;(bb-check-path msg: "launch:execute incoming")

				(db:test-get-host        test-info)
				(begin
				  (debug:print 0 *default-log-port* "ERROR: failed to find a record for test-id " test-id ", exiting.")
				  (exit))))
		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond
             ;; -mrw- I'm removing KILLREQ from this list so that a test in KILLREQ state is treated as a "do not run" flag.
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
	     ((or (member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
		  (and (equal? (db:test-get-state test-info) "COMPLETED")                           ;; completed/abort => rerun if asked
		       (member (db:test-get-status test-info) '("ABORT"))))
	      (debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")

              (rmt:general-call 'set-test-start-time #f test-id)
              (rmt:test-set-state-status run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      ) ;; prime it for running
	     ((member (db:test-get-state test-info) '("RUNNING" "REMOTEHOSTSTART"))

;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;; Always use two or four digit decimal
;; 1.01, 1.02...1.10,1.11,1.1101 ... 1.99,2.00..

;; (declare (unit megatest-version))

(define megatest-version 1.65661)
(define megatest-version 1.6574)

(use format typed-records) ;; RADT => purpose of json format??

(declare (unit rmt))
(declare (uses api))
(declare (uses http-transport))
(include "common_records.scm")
(include "db_records.scm")

;; (declare (uses rmtmod))

;; (import rmtmod)

;;
;; THESE ARE ALL CALLED ON THE CLIENT SIDE!!!
;;

;; Just some syntatic sugar
(define (rmt:register-test run-id test-name item-path)
  (rmt:general-call 'register-test run-id run-id test-name item-path))

(define (rmt:get-test-id run-id testname item-path)
  (rmt:send-receive 'get-test-id run-id (list run-id testname item-path)))

;; run-id is NOT used
;; run-id is NOT used - but it will be! 
;;
(define (rmt:get-test-info-by-id run-id test-id)
  (if (number? test-id)
      (rmt:send-receive 'get-test-info-by-id run-id (list run-id test-id))
      (let* ((testdat  (rmt:send-receive 'get-test-info-by-id run-id (list run-id test-id)))
	     (trundatf (conc (db:test-get-rundir testdat) "/.mt_data/test-run.dat")))
	;; now we can update a couple fields from the filesystem
	(if (and (db:test-get-rundir testdat)
		 (file-exists? trundatf))
	    (let* ((duration   (db:test-get-run_duration testdat))
		   (event-time (db:test-get-event_time   testdat))
		   (last-touch (file-modification-time trundatf)))
	      (db:test-set-run_duration! testdat (max duration (- last-touch event-time)))))
	testdat)
      (begin
	(debug:print 0 *default-log-port* "WARNING: Bad data handed to rmt:get-test-info-by-id run-id=" run-id ", test-id=" test-id)
	(print-call-chain (current-error-port))
	#f)))

(define (rmt:test-get-rundir-from-test-id run-id test-id)
  (rmt:send-receive 'test-get-rundir-from-test-id run-id (list run-id test-id)))

	 (lambda (count)
	   (set! res count))
	 tdb
	 "SELECT count(id) FROM test_rundat;")
	res))
  0)

;; 
(define (tests:update-central-meta-info run-id test-id cpuload diskfree minutes uname hostname)
  (rmt:general-call 'update-test-rundat run-id test-id (current-seconds) (or cpuload -1)(or diskfree -1) -1 (or minutes -1))
  (if (and cpuload diskfree)
      (rmt:general-call 'update-cpuload-diskfree run-id cpuload diskfree test-id))
  (if minutes 
      (rmt:general-call 'update-run-duration run-id minutes test-id))
  (if (and uname hostname)
      (rmt:general-call 'update-uname-host run-id uname hostname test-id)))
(define (tests:update-central-meta-info run-id test-id cpuload diskfree minutes uname hostname #!key (update-db #f)(tmpfree #f))
  (if (get-environment-variable "MT_TEST_RUN_DIR")
      (let* ((dest-dir (conc (get-environment-variable "MT_TEST_RUN_DIR") "/.mt_data"))
	     (or-dash  (lambda (instr)
			 (cond
			  ((not instr) "") ;; #f -> blank, indicates value unchanged since last measurement taken
			  ((string? instr)(if (string-search " " instr) (conc "\"" instr "\"") instr))
			  (else instr))))
	     (file-new (not (directory-exists? dest-dir))))
	(if file-new (create-directory dest-dir #t))
	(let* ((outp (open-output-file (conc dest-dir "/test-run.dat") #:append)))
	  (with-output-to-port outp
	    (lambda ()
	      (if file-new
		  (print "epoch_time,run_id,test_id,cpuload,diskfree,tmpfree,run_minutes,hostname,uname"))
	      (print (current-seconds) "," (or-dash run-id)   "," (or-dash test-id)  ","
		     (or-dash cpuload) "," (or-dash diskfree) "," (or-dash tmpfree)  ","
		     (or-dash minutes) "," (or-dash hostname) ","
		     (or-dash uname)))) ;; put uname last as it has spaces in it
	  (close-output-port outp)))
      (begin
	(rmt:general-call 'update-test-rundat run-id test-id (current-seconds) (or cpuload -1)(or diskfree -1) -1 (or minutes -1))))
  (if update-db
      (begin
	(if (and cpuload diskfree)
	    (rmt:general-call 'update-cpuload-diskfree run-id cpuload diskfree test-id))
	(if minutes 
	    (rmt:general-call 'update-run-duration run-id minutes test-id))
	(if (and uname hostname)
	    (rmt:general-call 'update-uname-host run-id uname hostname test-id)))))
  
;; This one is for running with no db access (i.e. via rmt: internally)
(define (tests:set-full-meta-info db test-id run-id minutes work-area remtries)
(define (tests:set-full-meta-info db test-id run-id minutes work-area remtries #!key (update-db #f))
;; (define (tests:set-full-meta-info test-id run-id minutes work-area)
;;  (let ((remtries 10))
  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (tmpfree  (get-df "/tmp"))
	 (uname    (get-uname "-srvpio"))
	 (hostname (get-host-name)))
    (tests:update-central-meta-info run-id test-id cpuload diskfree minutes uname hostname)))
    (tests:update-central-meta-info run-id test-id cpuload diskfree minutes uname hostname update-db: update-db tmpfree: tmpfree)))
    
;; (define (tests:set-partial-meta-info test-id run-id minutes work-area)
#;(define (tests:set-partial-meta-info test-id run-id minutes work-area remtries)
  (let* ((cpuload  (get-cpu-load))
	 (diskfree (get-df (current-directory)))
	 (remtries 10))
    (handle-exceptions
     exn
     (if (> remtries 0)
	 (begin
	   (print-call-chain (current-error-port))
	   (debug:print-info 0 *default-log-port* "WARNING: failed to set meta info. Will try " remtries " more times")
	   (set! remtries (- remtries 1))
	   (thread-sleep! 10)
	   (tests:set-full-meta-info db test-id run-id minutes work-area (- remtries 1)))
	 (let ((err-status ((condition-property-accessor 'sqlite3 'status #f) exn)))
	   (debug:print-error 0 *default-log-port* "tried for over a minute to update meta info and failed. Giving up")
	   (debug:print 0 *default-log-port* "EXCEPTION: database probably overloaded or unreadable.")
	   (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
	   (debug:print 5 *default-log-port* "exn=" (condition->list exn))
	   (debug:print 0 *default-log-port* " status:  " ((condition-property-accessor 'sqlite3 'status) exn))
	   (print-call-chain (current-error-port))))
     (tests:update-testdat-meta-info db test-id work-area cpuload diskfree minutes)
  )))
	 
;;======================================================================
;; A R C H I V I N G
;;======================================================================

(define (test:archive db test-id)
  #f)