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.

	  (debug:print 0 *default-log-port* "ERROR: cannot read " infile)
          (debug:print 0 *default-log-port* "ERROR: run-dir is " run-dir)
          #f
          )
        (with-input-from-file infile read-lines)
	)))

;; check duration against test-run.dat file if it exists and update the value in
;; the db if necessary
;;  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'));

;;
(define (db:adjust-run-duration dbstruct test-id run-dir event-time run-duration)
  (let* ((datf             (conc run-dir ".mt_data/test-run.dat"))
	 (modt             (if (and (file-exists? datf)
				    (file-read-access? datf))
			       (file-modification-time datf)
			       #f)) ;; (+ event-time run-duration))))
	 (alt-run-duration (if modt
			       (- modt event-time)
			       #f)))
    (if (and alt-run-duration
	     (> alt-run-duration run-duration))
	(begin
	  (debug:print 0 *default-log-port* "Test " test-id " run duration mismatch. Setting to " alt-run-duration)
	  (db:with-db
	   dbstruct #f #f
	   (lambda (db)
	     (sqlite3:execute db "UPDATE tests SET run_duration=? WHERE id=?;" alt-run-duration test-id)
	     #t)))
	#f))) ;; #f = we did NOT adjust the time
	      
(define (db:find-and-mark-incomplete dbstruct run-id ovr-deadtime)
  (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 (configf:lookup-number *configdat* "setup" "deadtime")))

	 ;; 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 (not (db:adjust-run-duration dbstruct test-id run-dir 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))))
		(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 (not (db:adjust-run-duration dbstruct test-id run-dir 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)))))
		(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))))))
	  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

  (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.
	     ;;((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"))))
	     ((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
	      (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

(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)))
             (trundir  (vector-ref testdat 10))
	     (trundatf (conc trundir"/.mt_data/test-run.dat")))
	;; now we can update a couple fields from the filesystem
	(handle-exceptions
	    exn
	    (begin
	      (debug:print-info 0 *default-log-port* "Could not update testdat record from "trundatf", exn=" exn)
	      #f)
	  (if (and trundir
		   (file-exists? trundatf))
	      (let* ((duration     (vector-ref testdat 12)) ;; (db:test-get-run_duration testdat))
		     (event-time   (vector-ref testdat 5))   ;; (db:test-get-event_time   testdat))
		     (last-touch   (file-modification-time trundatf))
		     (new-duration (max duration (- last-touch event-time))))
		(vector-set! testdat 12 new-duration))))
	      #;(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