Megatest

(define *state-ignore-hash*  (make-hash-table))

(define *tests-sort-options* (vector (vector "Sort +a" 'testname   "ASC")
				     (vector "Sort -a" 'testname   "DESC")
				     (vector "Sort +t" 'event_time "ASC")
				     (vector "Sort -t" 'event_time "DESC")
				     (vector "Sort +s" 'statestatus "ASC")
				     (vector "Sort -s" 'statestatus "DESC")
				     (vector "Sort +a" 'testname   "ASC")))

(define *tests-sort-type-index* '(("+testname" 0)
				  ("-testname" 1)
				  ("+event_time" 2)
				  ("-event_time" 3)
				  ("+statestatus" 4)
				  ("-statestatus" 5)))

			   (teststatus (db:test-get-status   test))
			   (teststate  (db:test-get-state    test))
			   ;;(teststart  (db:test-get-event_time test))
			   (runtime    (db:test-get-run_duration test))
			   (buttontxt  (cond
					((equal? teststate "COMPLETED") teststatus)
					((and (equal? teststate "NOT_STARTED")
					      (member teststatus '("ZERO_ITEMS" "BLOCKED" "PREQ_FAIL" "PREQ_DISCARDED" "TIMED_OUT" "KEEP_TRYING" "TEN_STRIKES")))
					 teststatus)
					(else
					 teststate)))
			   (button     (vector-ref columndat rown))
			   (color      (car (gutils:get-color-for-state-status teststate teststatus)))
			   (curr-color (vector-ref buttondat 1)) ;; (iup:attribute button "BGCOLOR"))
			   (curr-title (vector-ref buttondat 2))) ;; (iup:attribute button "TITLE")))

digraph megatest_state_status {
  ranksep=0.05
  // rankdir=LR

node [shape=box,style=filled];

// subgraph cluster_notstarted {
//   label="Not started";

  "NOT_STARTED FAILS" [
  label = "{ NOT_STARTED/FAILS |{ NO_ITEMS |<here> FAIL_PREREQ |<here> FAIL_TIMEOUT }}";
  shape= "record";
  ]

"NOT_STARTED n/a" -> "LAUNCHED n/a" [label=" launch"];
"NOT_STARTED WAIT" -> "LAUNCHED n/a"

  "NOT_STARTED n/a";
  "NOT_STARTED WAIT" [
  label = "{NOT_STARTED WAIT|{ NO_SLOTS | <here> WAIT_PREREQ}}";
  shape = "record";
]

// struct3 [shape=record,label="hello\nworld |{ b |{c|<here> d|e}| f}| g | h"];

  "NOT_STARTED n/a" -> "NOT_STARTED FAILS";
  "NOT_STARTED n/a" -> "NOT_STARTED WAIT";

  "RUNNING" [
     shape="record";
     label="{RUNNING|{n/a|<here> PASS |<here> FAIL}}";
  ]

  "COMPLETED" [
      shape="record";
      label = "{COMPLETED|{PASS | <here> FAIL |<here> CHECK|<here> SKIP}}";
  ]


"RUNNING" -> "COMPLETED";
"RUNNING" -> "INCOMPLETE" [label="test dead for > 24hrs"];


"LAUNCHED n/a" -> "REMOTEHOSTSTART n/a" -> "RUNNING";

}

							     (if p-id
								 (begin
								   (debug:print 0 "Killing " (cadr parts) "; kill -9  " p-id)
								   ;; (process-signal pid signal/kill))))) ;; 
								   (system (conc "kill -9 " p-id))))))
							 (car processes)))
						      (system (conc "kill -9 -" pid))
						      (tests:test-set-status! test-id "KILLED"  "KILLED" (args:get-arg "-m") #f)))
						   (begin
						     (debug:print 0 "WARNING: Request received to kill job but problem with process, attempting to kill manager process")
						     (tests:test-set-status! test-id "KILLED"  "KILLED" (args:get-arg "-m") #f)
						     (tests:test-set-status! run-id test-id "KILLED"  "FAIL" (args:get-arg "-m") #f)
						     (exit 1) ;; IS THIS NECESSARY OR WISE???
						     )))
					     (set! kill-tries (+ 1 kill-tries))
					     (mutex-unlock! m)))
				       (if keep-going
					   (begin

(declare (uses common))

;;======================================================================
;; attempt to prevent overlapping updates of rollup files by queueing
;; update requests in an sqlite db
;;======================================================================

(define (lock-queue:open-db fname #!key (count 10))
  (let* ((actualfname (conc fname ".lockdb"))
	 (dbexists (file-exists? actualfname))
	 (db       (sqlite3:open-database actualfname))
	 (handler  (make-busy-timeout 136000)))
    (if dbexists
	db
	(begin
	  (handle-exceptions
	   exn
	   (begin
	     (thread-sleep! 10)
	     (if (> count 0)
		 (lock-queue:open-db fname count: (- count 1))
		 db))
	   (sqlite3:execute 
	    db
	    "CREATE TABLE IF NOT EXISTS queue (
  	      id         INTEGER PRIMARY KEY,
              test_id    INTEGER,
              start_time INTEGER,
              state      TEXT,
              CONSTRAINT queue_constraint UNIQUE (test_id));")
	   (sqlite3:execute
	    db
	    "CREATE TABLE IF NOT EXISTS runlocks (
              id         INTEGER PRIMARY KEY,
              test_id    INTEGER,
              run_lock   TEXT,
              CONSTRAINT runlock_constraint UNIQUE (run_lock));"))))
    (sqlite3:set-busy-handler! db handler)
    db))

(define (lock-queue:set-state db test-id newstate #!key (remtries 10))
  (handle-exceptions
   exn
   (if (> remtries 0)

	;; Actually this should not be needed as mystart cannot be simultaneously less than and test-id same as 
	(if (not (equal? tid test-id)) 
	    (set! res tid)))
      db
      "SELECT test_id FROM queue WHERE start_time > ?;" mystart)
     res)))

(define (lock-queue:get-lock db test-id #!key (count 10))
  (let ((res       #f)
	(lckqry    (sqlite3:prepare db "SELECT test_id,run_lock FROM runlocks WHERE run_lock='locked';"))
	(mklckqry  (sqlite3:prepare db "INSERT INTO runlocks (test_id,run_lock) VALUES (?,'locked');")))
    (let ((result 
	   (handle-exceptions
	    exn
	    (begin
	      (thread-sleep! 10)
	      (if (> count 0)
		  (lock-queue:get-lock db test-id count: (- count 1)))
	      #f)
	    (sqlite3:with-transaction
	     db
	     (lambda ()
	       (sqlite3:for-each-row (lambda (tid lockstate)
				       (set! res (list tid lockstate)))
				     lckqry)
	       (if res
		   (if (equal? (car res) test-id)
		       #t ;; already have the lock
		       #f)
		   (begin
		     (sqlite3:execute mklckqry test-id)
		     ;; if no error handled then return #t for got the lock
		     #t)))))))
      (sqlite3:finalize! lckqry)
      (sqlite3:finalize! mklckqry)
      result)))

(define (lock-queue:release-lock fname test-id #!key (count 10))
  (let ((db (lock-queue:open-db fname)))
    (handle-exceptions
     exn
     (begin
       (thread-sleep! 10)
       (if (> count 0)
	   (lock-queue:release-lock fname test-id count: (- count 1))
	   #f))
     (sqlite3:execute db "DELETE FROM runlocks WHERE test_id=?;" test-id)
     (sqlite3:finalize! db))))

(define (lock-queue:steal-lock db test-id #!key (count 10))
  (handle-exceptions
   exn
   (begin
     (thread-sleep! 10)
     (if (> count 0)
	 (lock-queue:steal-lock db test-id count: (- count 1))
	 #f))
   (sqlite3:execute db "DELETE FROM runlocks WHERE run_lock='locked';"))
  (lock-queue:get-lock db test-it))

;; returns #f if ok to skip the task
;; returns #t if ok to proceed with task
;; otherwise waits
;;
(define (lock-queue:wait-turn fname test-id #!key (count 10))
  (let ((db      (lock-queue:open-db fname))
	(mystart (current-seconds)))
    (handle-exceptions
     exn
     (begin
       (thread-sleep! 10)
       (if (> count 0)
	   (lock-queue:wait-turn fname test-id count: (- count 1))
	   #f))
     (sqlite3:execute
      db
      "INSERT OR REPLACE INTO queue (test_id,start_time,state) VALUES (?,?,'waiting');"
      test-id mystart)
     (thread-sleep! 1) ;; give other tests a chance to register
     (let ((result 
	    (let loop ((younger-waiting (lock-queue:any-younger? db mystart test-id)))
	      (if younger-waiting
		  (begin
		    ;; no need for us to wait. mark in the lock queue db as skipping
		    (lock-queue:set-state db test-id "skipping")
		    #f) ;; let the calling process know that nothing needs to be done
		  (if (lock-queue:get-lock db test-id)
		      #t
		      (if (> (- (current-seconds) mystart) 36000) ;; waited too long, steal the lock
			  (lock-queue:steal-lock db test-id)
			  (begin
			    (thread-sleep! 1)
			    (loop (lock-queue:any-younger? db mystart test-id)))))))))
       (sqlite3:finalize! db)
       result))))
	  
            
;; (use trace)
;; (trace lock-queue:get-lock lock-queue:release-lock lock-queue:wait-turn lock-queue:any-younger? lock-queue:set-state)

(define runs:nothing-left-in-queue-count 0)

(define (runs:expand-items hed tal reg reruns regfull newtal jobgroup max-concurrent-jobs run-id waitons item-path testmode test-record can-run-more items runname tconfig reglen test-registry test-records itemmap)
  (let* ((loop-list       (list hed tal reg reruns))
	 (prereqs-not-met (rmt:get-prereqs-not-met run-id waitons item-path testmode itemmap: itemmap))
	 ;; (prereqs-not-met (mt:lazy-get-prereqs-not-met run-id waitons item-path mode: testmode itemmap: itemmap))
	 (fails           (runs:calc-fails prereqs-not-met))
	 (prereq-fails    (runs:calc-prereq-fail prereqs-not-met))
	 (non-completed   (runs:calc-not-completed prereqs-not-met))
	 (runnables       (runs:calc-runnable prereqs-not-met)))
    (debug:print-info 4 "START OF INNER COND #2 "
		      "\n can-run-more:    " can-run-more
		      "\n testname:        " hed
		      "\n prereqs-not-met: " (runs:pretty-string prereqs-not-met)
		      "\n non-completed:   " (runs:pretty-string non-completed) 
		      "\n prereq-fails:    " (runs:pretty-string prereq-fails)
		      "\n fails:           " (runs:pretty-string fails)
		      "\n testmode:        " testmode
		      "\n (member 'toplevel testmode): " (member 'toplevel testmode)
		      "\n (null? non-completed):    " (null? non-completed)
		      "\n reruns:          " reruns
		      "\n items:           " items
		      "\n can-run-more:    " can-run-more)

		(tests:testqueue-set-items! test-record items-list)
		(list hed tal reg reruns))
	      (begin
		(debug:print 0 "ERROR: The proc from reading the items table did not yield a list - please report this")
		(exit 1))))))

     ((and (null? fails)
	   (null? prereq-fails)
	   (not (null? non-completed)))
      (let* ((allinqueue (map (lambda (x)(if (string? x) x (db:test-get-testname x)))
        		      (append newtal reruns)))
	     ;; prereqstrs is a list of test names as strings that are prereqs for hed
             (prereqstrs (delete-duplicates (map (lambda (x)(if (string? x) x (db:test-get-testname x)))
						 prereqs-not-met)))
	     ;; a prereq that is not found in allinqueue will be put in the notinqueue list

		  #f
		  (list (runs:queue-next-hed trimmed-tal trimmed-reg reglen regfull)
			(runs:queue-next-tal trimmed-tal trimmed-reg reglen regfull)
			(runs:queue-next-reg trimmed-tal trimmed-reg reglen regfull)
			reruns)))
	      (list (car newtal)(append (cdr newtal) reg) '() reruns))))























































     ((and (null? fails)
	   (null? prereq-fails)
	   (null? non-completed))
      (if  (runs:can-keep-running? hed 5)
	  (begin
	    (runs:inc-cant-run-tests hed)
	    (debug:print-info 1 "no fails in prerequisites for " hed " but also none running, keeping " hed " for now. Try count: " (hash-table-ref/default *seen-cant-run-tests* hed 0))
	    ;; num-retries code was here
	    ;; we use this opportunity to move contents of reg to tal
	    (list (car newtal)(append (cdr newtal) reg) '() reruns)) ;; an issue with prereqs not yet met?
	  (begin
	    (debug:print-info 1 "no fails in prerequisites for " hed " but nothing seen running in a while, dropping test " hed " from the run queue")
	    (let ((test-id (rmt:get-test-id run-id hed "")))
	      (mt:test-set-state-status-by-id test-id "NOT_STARTED" "TIMED_OUT" "Nothing seen running in a while."))
	    (list (runs:queue-next-hed tal reg reglen regfull)
		  (runs:queue-next-tal tal reg reglen regfull)
		  (runs:queue-next-reg tal reg reglen regfull)
		  reruns))))

     ((and 
       (or (not (null? fails))
	   (not (null? prereq-fails)))
       (member 'normal testmode))
      (debug:print-info 1 "test "  hed " (mode=" testmode ") has failed prerequisite(s); "
			(string-intersperse (map (lambda (t)(conc (db:test-get-testname t) ":" (db:test-get-state t)"/"(db:test-get-status t))) fails) ", ")
			", removing it from to-do list")
      (let ((test-id (rmt:get-test-id run-id hed "")))
	(if (not (null? prereq-fails))
	    (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_DISCARDED" "Failed to run due to prior failed prerequisites")
	    (mt:test-set-state-status-by-id run-id test-id "NOT_STARTED" "PREQ_FAIL"      "Failed to run due to failed prerequisites")))
      (if (or (not (null? reg))(not (null? tal)))
	  (begin
	    (hash-table-set! test-registry hed 'CANNOTRUN)
	    (list (runs:queue-next-hed tal reg reglen regfull)
		  (runs:queue-next-tal tal reg reglen regfull)
		  (runs:queue-next-reg tal reg reglen regfull)
		  (cons hed reruns)))
	  #f)) ;; #f flags do not loop

     ((and (not (null? fails))(member 'toplevel testmode))
      (if (or (not (null? reg))(not (null? tal)))
	   (list (car newtal)(append (cdr newtal) reg) '() reruns)
	  #f)) 
     ((null? runnables) #f) ;; if we get here and non-completed is null the it's all over.
     (else
      (debug:print 0 "WARNING: FAILS or incomplete tests maybe preventing completion of this run. Watch for issues with test " hed ", continuing for now")
      ;; (list (runs:queue-next-hed tal reg reglen regfull)
      ;;   	(runs:queue-next-tal tal reg reglen regfull)
      ;;   	(runs:queue-next-reg tal reg reglen regfull)
      ;;   	reruns)
      (list (car newtal)(cdr newtal) reg reruns)))))

	    (if (not (vector? t))
		t
		(let ((state  (db:test-get-state t))
		      (status (db:test-get-status t)))
		  (case (string->symbol state)
		    ((COMPLETED) #f)
		    ((NOT_STARTED)
		     (if (member status '("TEN_STRIKES" "BLOCKED" "PREQ_FAIL" "ZERO_ITEMS" "PREQ_DISCARDED" "TIMED_OUT" ))
			 #f
			 t))
		    ((DELETED) #f)
		    (else t)))))
	  tests))

;; every time though the loop increment the test/itempatt val.

(define (runs:calc-fails prereqs-not-met)
  (filter (lambda (test)
	    (and (vector? test) ;; not (string? test))
		 (equal? (db:test-get-state test) "COMPLETED")
		 (not (member (db:test-get-status test)
			      '("PASS" "WARN" "CHECK" "WAIVED" "SKIP")))))
	  prereqs-not-met))

(define (runs:calc-prereq-fail prereqs-not-met)
  (filter (lambda (test)
	    (and (vector? test) ;; not (string? test))
		 (equal? (db:test-get-state test) "NOT_STARTED")
		 (not (member (db:test-get-status test)
			      '("n/a" "KEEP_TRYING")))))
	  prereqs-not-met))

(define (runs:calc-not-completed prereqs-not-met)
  (filter
   (lambda (t)
     (or (not (vector? t))
	 (not (equal? "COMPLETED" (db:test-get-state t)))))
   prereqs-not-met))

(define (runs:calc-not-completed prereqs-not-met)
  (filter
   (lambda (t)
     (or (not (vector? t))
	 (not (equal? "COMPLETED" (db:test-get-state t)))))
   prereqs-not-met))

(define (runs:calc-runnable prereqs-not-met)
  (filter 
   (lambda (t)
     (or (not (vector? t))
	 (and (equal? "NOT_STARTED" (db:test-get-state t))
	      (member (db:test-get-status t)
			      '("n/a" "KEEP_TRYING")))))
   prereqs-not-met))

(define (runs:pretty-string lst)
  (map (lambda (t)
	 (if (not (vector? t))
	     (conc t)
	     (conc (db:test-get-testname t) ":" (db:test-get-state t) "/" (db:test-get-status t))))
       lst))

      (case (if force ;; (args:get-arg "-force")
		'NOT_STARTED
		(if testdat
		    (string->symbol (test:get-state testdat))
		    'failed-to-insert))
	((failed-to-insert)
	 (debug:print 0 "ERROR: Failed to insert the record into the db"))
	((NOT_STARTED COMPLETED DELETED INCOMPLETE)
	 (let ((runflag #f))
	   (cond
	    ;; -force, run no matter what
	    (force (set! runflag #t))
	    ;; NOT_STARTED, run no matter what
	    ((member (test:get-state testdat) '("DELETED" "NOT_STARTED" "INCOMPLETE"))(set! runflag #t))
	    ;; not -rerun and PASS, WARN or CHECK, do no run
	    ((and (or (not rerun)
		      keepgoing)
		  ;; Require to force re-run for COMPLETED or *anything* + PASS,WARN or CHECK
		  (or (member (test:get-status testdat) '("PASS" "WARN" "CHECK" "SKIP" "WAIVED"))
		      (member (test:get-state  testdat) '("COMPLETED")))) 
	     (debug:print-info 2 "running test " test-name "/" item-path " suppressed as it is " (test:get-state testdat) " and " (test:get-status testdat))

(define (runs:remove-test-directory db test remove-data-only)
  (let* ((run-dir       (db:test-get-rundir test))    ;; run dir is from the link tree
	 (real-dir      (if (file-exists? run-dir)
			    (resolve-pathname run-dir)
			    #f)))
    (if (not remove-data-only)
	(mt:test-set-state-status-by-id (db:test-get-run_id test)(db:test-get-id test) "REMOVING" "LOCKED" #f))
    (debug:print-info 1 "Attempting to remove " (if real-dir (conc " dir " real-dir " and ") "") " link " run-dir)
    (if (and real-dir 
	     (> (string-length real-dir) 5)
	     (file-exists? real-dir)) ;; bad heuristic but should prevent /tmp /home etc.
	(begin ;; let* ((realpath (resolve-pathname run-dir)))
	  (debug:print-info 1 "Recursively removing " real-dir)
	  (if (file-exists? real-dir)

test6: fullprep
	cd fullrun;$(MEGATEST) -preclean -runtests runfirst -testpatt %/1 -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_itempatt -v
	cd fullrun;$(MEGATEST) -preclean -runtests runfirst -testpatt %blahha% -reqtarg ubuntu/nfs/none :runname $(RUNNAME)_itempatt -debug 10
	cd fullrun;$(MEGATEST) -rollup :runname newrun -target ubuntu/nfs/none -debug 10

test7: 
	@echo Only a/c testname c should remain. If there is a run a/b/c then there is a cache issue.
	cd simplerun;$(DASHBOARD) &
	(cd simplerun; \
	 $(MEGATEST) -server - -daemonize; \
         $(MEGATEST) -remove-runs -target %/% :runname % -testpatt %; \
         $(MEGATEST) -preclean -runtests %  -target a/b :runname c; sleep 5; \
	 $(MEGATEST) -remove-runs -target a/c :runname c; \
	 $(MEGATEST) -preclean -runtests %  -target a/c :runname c; \
	 $(MEGATEST) -remove-runs -target a/b :runname c -testpatt % ; \

# Some simple checks for bootstrapping and run loop logic 

test9 : minsetup test9a test9b test9c test9d test9e

test9a :
	@echo Run super-simple mintest e, no waitons.
	cd mintest;$(DASHBOARD)&
	cd mintest;$(MEGATEST) -preclean -runtests e -target $(VER) :runname `date +%H.%M.%S` -debug $(DEBUG)

test9b :
	@echo Run simple mintest d with one waiton c
	cd mintest;$(MEGATEST) -preclean -runtests d -target $(VER) :runname `date +%H.%M.%S` -debug $(DEBUG)

test9c :

[fields]
SYSTEM TEXT
RELEASE TEXT

[setup]
# Adjust max_concurrent_jobs to limit how much you load your machines
# max_concurrent_jobs 150
max_concurrent_jobs 500

# This is your link path, you can move it but it is generally better to keep it stable
linktree #{shell readlink -f #{getenv PWD}/../simplelinks}

[include testqa/configs/megatest.abc.config]

# The empty var should have a definition with null string
EMPTY_VAR

WRAPPEDVAR This var should have the work blah thrice: \
blah \
blah

MAX_ALLOWED_LOAD 200
# XTERM   [system xterm]
# RUNDEAD [system exit 56]

[server]

# If the server can't be started on this port it will try the next port until
# it succeeds

#
function launchjob () {
  # Can't always trust $PWD
  CURRWD=`pwd`
  if [[ $TARGETHOST_LOGF == "" ]]; then
      TARGETHOST_LOGF=NBFAKE-`date +%GWW%V.%u_%T`
  fi
  # echo "#======================================================================"
  # echo "# NBFAKE Running command:"
  # echo "#     \"$*\""
  # echo "#======================================================================"
  
  if [[ $TARGETHOST == ""  ]]; then
    sh -c "cd $CURRWD;export DISPLAY=$DISPLAY; export PATH=$PATH; nohup $* > $TARGETHOST_LOGF 2>&1 &"
  else
    ssh -n -f $TARGETHOST "sh -c \"cd $CURRWD;export DISPLAY=$DISPLAY; export PATH=$PATH; nohup $* > $TARGETHOST_LOGF 2>&1 &\""
  fi
}

    numcpu=2
fi

# NB// max_load is in units of percent.
#
lperc=$(echo "100 * $load / $numcpu"|bc)
lperc2=$(echo "100 * $load2 / $numcpu"|bc)
let lperc2adj="$lperc2 + 0.5"
if [[ "x$MAX_ALLOWED_LOAD" == "x" ]]; then
  max_load=100
else
  max_load=$MAX_ALLOWED_LOAD
fi

lfile=/tmp/loadrunner-$USER.lockfile
lockfile -r 5 -l 60 $lfile &> /dev/null

if [[  $lperc -lt $max_load ]];then
  if [[ $lperc -le $lperc2adj ]];then
    # echo "Load acceptable: lperc=$lperc %, max_load=$max_load %, load=$load, numcpu=$numcpu, MAX_ALLOWED_LOAD=$MAX_ALLOWED_LOAD % and $lperc2 < $lperc"
    # echo "Starting command: \"$@\""
    launchjob "$@"
    # we sleep ten seconds here to keep the lock a little longer and give time for
    # the uptime to show a response
    sleep 10
  else
   echo "$LOADRUNNER $@" | at now + 2 minutes &> /dev/null
  fi
else
  # echo "Load too high: lperc=$lperc, max_load=$max_load, waiting two minutes before trying to run command: \"$@\""
  echo "$LOADRUNNER $@" | at now + 2 minutes &> /dev/null
fi

sleep $(get_delay_time 10)
rm -f $lfile