Megatest

;; Move me elsewhere ...
;; RADT => Why do we meed the version check here, this is called only if version misma
;;
(define (common:cleanup-db dbstruct)
  (db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'dejunk
   'adj-target
   ;; 'old2new
   'new2old
   'schema)
   )
  (if (common:version-changed?)
      (common:set-last-run-version)))

;; Rotate logs, logic: 
;;                 if > 500k and older than 1 week:
;;                     remove previous compressed log and compress this log
;; WARNING: This proc operates assuming that it is in the directory above the

;;
(define (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:version-changed?)
	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
                (dbfile (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
                (read-only (not (file-write-access? dbfile)))
                (dbstruct (db:setup)))
                (dbstruct (db:setup #t)))
	    (debug:print 0 *default-log-port*
			 "WARNING: Version mismatch!\n"
			 "   expected: " (common:version-signature) "\n"
			 "   got:      " (common:get-last-run-version))
            (cond
             ((get-environment-variable "MT_SKIP_DB_MIGRATE") #t)
             ((and (file-exists? mtconf) (file-exists? dbfile) (not read-only)

	    (if (common:low-noise-print 30)
		(debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" this-wd-num="this-wd-num)))))))

;; TODO: for multiple areas, we will have multiple watchdogs; and multiple threads to manage
(define (common:watchdog)
  (debug:print-info 13 *default-log-port* "common:watchdog entered.")
  (if (common:on-homehost?)
      (let ((dbstruct (db:setup)))
      (let ((dbstruct (db:setup #t)))
	(debug:print-info 13 *default-log-port* "after db:setup with dbstruct="dbstruct)
	(cond
	 ((dbr:dbstruct-read-only dbstruct)
	  (debug:print-info 13 *default-log-port* "loading read-only watchdog")
	  (common:readonly-watchdog dbstruct))
	 (else
	  (debug:print-info 13 *default-log-port* "loading writable-watchdog.")

			"-guimonitor"
			"-main"
			"-v"
			"-q"
			"-use-db-cache"
			"-skip-version-check"
			"-repl"
                        "-rh5.11" ;; fix to allow running on rh5.11
			)
		 args:arg-hash
		 0))

(if (not (null? remargs))
    (begin
      (print "Unrecognised arguments: " (string-intersperse remargs " "))
      (exit)))

(if (args:get-arg "-h")
    (begin
      (print help)
      (exit)))

;; TODO: Move this inside (main)
;;
(if (not (launch:setup))
    (begin
      (print "Failed to find megatest.config, exiting") 
      (exit 1)))

;; deal with RH 5.11 gtk lib or iup lib missing detachbox feature
;; first check for the switch
;;
(if (or (args:get-arg "-rh5.11")
	(configf:lookup *configdat* "dashboard" "no-detachbox"))
    (set! iup:detachbox iup:vbox))

(if (not (common:on-homehost?))
    (begin
      (debug:print 0 *default-log-port* "WARNING: Current policy requires running dashboard on homehost: " (common:get-homehost))))
    
;; RA => Might require revert for filters 
;; create a watch dog to move changes from lt/.db/*.db to megatest.db

;;     (dbr:dbstruct-olddb-set!  dbstruct olddb)
;;     ;;; (mutex-unlock! *rundb-mutex*) ;;; why did we need a mutex on opening db's?
;;     (db:sync-tables db:sync-tests-only *megatest-db* db)
;;     db))

;; This routine creates the db if not already present. It is only called if the db is not already opened
;;
(define (db:open-db dbstruct #!key (areapath #f)) ;; TODO: actually use areapath
(define (db:open-db dbstruct #!key (areapath #f)(do-sync #t)) ;; TODO: actually use areapath
  (let ((tmpdb-stack (dbr:dbstruct-dbstack dbstruct))) ;; RA => Returns the first reference in dbstruct
    (if (stack? tmpdb-stack)
	(db:get-db tmpdb-stack) ;; get previously opened db (will create new db handle if all in the stack are already used
        (let* ((dbpath       (db:dbfile-path ))      ;; path to tmp db area
               (dbexists     (file-exists? dbpath))
	       (tmpdbfname   (conc dbpath "/megatest.db"))
	       (dbfexists    (file-exists? tmpdbfname))  ;; (conc dbpath "/megatest.db")))

                (dbr:dbstruct-read-only-set! dbstruct #t)))
          (dbr:dbstruct-mtdb-set!   dbstruct mtdb)
          (dbr:dbstruct-tmpdb-set!  dbstruct tmpdb)
          (dbr:dbstruct-dbstack-set! dbstruct (make-stack)) ;; BB: why a stack?  Why would the number of db's be indeterminate?  Is this a legacy of 1.db 2.db .. ?
          (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb) ;; olddb is already a (cons db path)
          (dbr:dbstruct-refndb-set! dbstruct refndb)
          ;;	    (mutex-unlock! *rundb-mutex*)
          (if (or (not dbfexists)
                  (and modtimedelta
                       (> modtimedelta 10))) ;; if db in tmp is over ten seconds older than the file in MTRA then do a sync back
          (if (and  (or (not dbfexists)
			(and modtimedelta
			     (> modtimedelta 10))) ;; if db in tmp is over ten seconds older than the file in MTRA then do a sync back
		    do-sync)
	      (begin
		(debug:print 4 *default-log-port* "filling db " (db:dbdat-get-path tmpdb) " with data \n    from " (db:dbdat-get-path mtdb) " mod time delta: " modtimedelta)
		(db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb refndb tmpdb)
                (debug:print-info 13 *default-log-port* "db:sync-all-tables-list done.")
                )
	      (debug:print 4 *default-log-port* " db, " (db:dbdat-get-path tmpdb) " already exists or fresh enough, not propogating data from\n     " (db:dbdat-get-path mtdb) " mod time delta: " modtimedelta) )
	  ;; (db:multi-db-sync dbstruct 'old2new))  ;; migrate data from megatest.db automatically
          tmpdb))))

;; Make the dbstruct, setup up auxillary db's and call for main db at least once
;;
;; called in http-transport and replicated in rmt.scm for *local* access. 
;;
(define (db:setup #!key (areapath #f))
(define (db:setup do-sync #!key (areapath #f))
  ;;

  (cond
   (*dbstruct-db* *dbstruct-db*);; TODO: when multiple areas are supported, this optimization will be a hazard
   (else ;;(common:on-homehost?)
    (debug:print-info 13 *default-log-port* "db:setup entered (first time, not cached.)")
    (let* ((dbstruct (make-dbr:dbstruct)))
      (when (not *toppath*)
        (debug:print-info 13 *default-log-port* "in db:setup, *toppath* not set; calling launch:setup")
        (launch:setup areapath: areapath))
      (debug:print-info 13 *default-log-port* "Begin db:open-db")
      (db:open-db dbstruct areapath: areapath)
      (db:open-db dbstruct areapath: areapath do-sync: do-sync)
      (debug:print-info 13 *default-log-port* "Done db:open-db")
      (set! *dbstruct-db* dbstruct)
      ;;(debug:print-info 13 *default-log-port* "new dbstruct = "(dbr:dbstruct->alist dbstruct))
      dbstruct))))
   ;; (else
   ;;  (debug:print 0 *default-log-port* "ERROR: attempt to open database when not on homehost. Exiting. Homehost: " (common:get-homehost))
   ;;  (exit 1))))

                            (all-curr-statuses    (common:special-sort  ;; worst -> best
                                                   (delete-duplicates
                                                    (cons status (map dbr:counts-status state-status-counts)))
                                                   *common:std-statuses* >))
			    (non-completes     (filter (lambda (x)
							 (not (equal? x "COMPLETED")))
						       all-curr-states))
			    (num-non-completes (length non-completes))
                            (newstate          (cond
						((> running 0)
						 "RUNNING") ;; anything running, call the situation running
						((> bad-not-started 0)  ;; we have an ugly situation, it is completed in the sense we cannot do more.
						 "COMPLETED") 
						((> (length non-completes) 0) ;;
						((> num-non-completes 0) ;;
						 (car non-completes))  ;;  (remove (lambda (x)(equal? "COMPLETED" x)) all-curr-states)))
						(else
						 (car all-curr-states))))
			                       ;; (if (> running 0)
                                               ;;     "RUNNING"
                                               ;;     (if (> bad-not-started 0)
                                               ;;         "COMPLETED"
                                               ;;         (car all-curr-states))))
                            (newstatus            (if (> bad-not-started 0)
                                                      "CHECK"
                            (newstatus            (if (or (> bad-not-started 0)
							  (and (equal? newstate "NOT_STARTED")
							       (> num-non-completes 0)))
						      "CHECK"
                                                      (car all-curr-statuses))))
                       ;; (print "bad-not-supported: " bad-not-support " all-curr-states: " all-curr-states " all-curr-statuses: " all-curr-states)
                       ;;      " newstate: " newstate " newstatus: " newstatus)
                       ;; NB// Pass the db so it is part of the transaction
                       (db:test-set-state-status db run-id tl-test-id newstate newstatus #f)))))))
         (mutex-unlock! *db-transaction-mutex*)
         (if (and test-id state status (equal? status "AUTO")) 

(define (gutils:colors-similar? color1 color2)
  (let* ((c1 (map string->number (string-split color1)))
	 (c2 (map string->number (string-split color2)))
	 (delta (map (lambda (a b)(abs (- a b))) c1 c2)))
    (null? (filter (lambda (x)(> x 3)) delta))))

(define gutils:colors
  '((PASS . "70 249 73")
    (FAIL . "253 33 49")
    (SKIP . "230 230 0")))

(define (gutils:get-color-spec effective-state)
  (or (alist-ref effective-state gutils:colors)
      (alist-ref 'FAIL gutils:colors)))
 
(define (gutils:get-color-for-state-status state status);; #!key (get-label #f))
  ;; ((if get-label cadr car)
  (case (string->symbol state)
    ((COMPLETED) ;; ARCHIVED)
     (case (string->symbol status)
       ((PASS)        (list "70  249 73" status))
       ((WARN WAIVED) (list "255 172 13" status))
       ((SKIP)        (list "230 230 0" status))
       ((SKIP)        (list (gutils:get-color-spec 'SKIP) status))
       ((ABORT)       (list "198 36 166" status))
       (else (list "253 33 49" status))))
    ((ARCHIVED)
     (case (string->symbol status)
       ((PASS)        (list "70  170 73" status))
       ((WARN WAIVED) (list "200 130 13" status))
       ((SKIP)        (list "180 180 0" status))
       ((SKIP)        (list (gutils:get-color-spec 'SKIP) status))
       (else (list "180 33 49" status))))
    ;;      (if (equal? status "PASS")
    ;;	  '("70 249 73" "PASS")
    ;;	  (if (or (equal? status "WARN")
    ;;		  (equal? status "WAIVED"))
    ;;	      (list "255 172 13" status)
    ;;	      (list "223 33 49"  status)))) ;; greenish orangeish redish
    ((LAUNCHED)         (list "101 123 142"  state))
    ((CHECK)            (list "255 100 50"   state))
    ((REMOTEHOSTSTART)  (list "50 130 195"   state))
    ((RUNNING)          (list "9 131 232"    state))
    ((KILLREQ)          (list "39 82 206"    state))
    ((KILLED)           (list "234 101 17"   state))
    ((NOT_STARTED)      (case (string->symbol status)
			  ((CHECK)(list (gutils:get-color-spec 'SKIP) state))
    ((NOT_STARTED)      (list "240 240 240"  state))
			  (else   (list "240 240 240"                 state))))
    ;; for xor mode below
    ;;
    ((CLEAN)
     (case (string->symbol status)
       ((CLEAN-FAIL CLEAN-CHECK CLEAN-ABORT)  (list "200 130 13" status)) ;; orange requested for these
       (else  (list "60  235 63" status))))
    ((DIRTY-BETTER)     (list "160  255 153" status))

						  headers: '((content-type text/plain))))
				  (else (continue))))))))
    (http-transport:try-start-server ipaddrstr start-port)))

;; This is recursively run by http-transport:run until sucessful
;;
(define (http-transport:try-start-server ipaddrstr portnum)
  (let ((config-hostname (configf:lookup *configdat* "server" "hostname")))
  (let ((config-hostname (configf:lookup *configdat* "server" "hostname"))
	(config-use-proxy (equal? (configf:lookup *configdat* "client" "use-http_proxy") "yes")))
    (if (not config-use-proxy)
	(determine-proxy (constantly #f)))
    (debug:print-info 0 *default-log-port* "http-transport:try-start-server time=" (seconds->time-string (current-seconds)) " ipaddrsstr=" ipaddrstr " portnum=" portnum " config-hostname=" config-hostname)
    (handle-exceptions
	exn
	(begin
	  (print-error-message exn)
	  (if (< portnum 64000)
	      (begin 

	       (server-state 'available)
	       (bad-sync-count 0)
	       (start-time     (current-milliseconds)))
      ;; Use this opportunity to sync the tmp db to megatest.db
      (if (not server-going) ;; *dbstruct-db* 
	  (begin
	    (debug:print 0 *default-log-port* "SERVER: dbprep")
	    (set! *dbstruct-db*  (db:setup)) ;;  run-id))
	    (set! *dbstruct-db*  (db:setup #t)) ;;  run-id))
	    (set! server-going #t)
	    (debug:print 0 *default-log-port* "SERVER: running, megatest version: " (common:get-full-version)) ;; NOTE: the server is NOT yet marked as running in the log. We do that in the keep-running routine.
	    (thread-start! *watchdog*)))
      
      ;; when things go wrong we don't want to be doing the various queries too often
      ;; so we strive to run this stuff only every four seconds or so.
      (let* ((sync-time (- (current-milliseconds) start-time))

;;    - 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)
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ((item-path       (item-list->path itemdat))
	 (contour         #f)) ;; NOT READY FOR THIS (args:get-arg "-contour")))
    (let loop ((delta        (- (current-seconds) *last-launch*))
	       (launch-delay (string->number (or (configf:lookup *configdat* "setup" "launch-delay") "5"))))
	       (launch-delay (string->number (or (configf:lookup *configdat* "setup" "launch-delay") "0"))))
      (if (> launch-delay delta)
	  (begin
	    (debug:print-info 0 *default-log-port* "Delaying launch of " test-name " for " (- launch-delay delta) " seconds")
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))
    (change-directory *toppath*)
    (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute", *maybe* - the longer they are set the longer each launch takes (must be non-overlapping with the vars)

         "-server"
         "-list-disks"
         "-list-targets"
         "-show-runconfig"
         ;;"-list-db-targets"
         "-show-runconfig"
         "-show-config"
         "-show-cmdinfo"))
         "-show-cmdinfo"
	 "-cleanup-db"))
       (no-watchdog-args-vals (filter (lambda (x) x)
                                      (map args:get-arg no-watchdog-args)))
       (start-watchdog (null? no-watchdog-args-vals)))
  ;;(BB> "no-watchdog-args="no-watchdog-args "no-watchdog-args-vals="no-watchdog-args-vals) 
  (if start-watchdog
      (thread-start! *watchdog*)))

;; manually started and thus should never be started in a non-megatest area. Thus no need to handle situation
;; where (launch:setup) returns #f?
;;
(if (or (args:get-arg "-log")(args:get-arg "-server")) ;; redirect the log always when a server
    (handle-exceptions
	exn
	(begin
	  (print "ERROR: Failed to switch to log output. " ((conition-property-accessor 'exn 'message) exn))
	  (print "ERROR: Failed to switch to log output. " ((condition-property-accessor 'exn 'message) exn))
	  )
      (let* ((tl   (or (args:get-arg "-log")(launch:setup)))   ;; run launch:setup if -server, ensure we do NOT run launch:setup if -log specified
	     (logf (or (args:get-arg "-log") ;; use -log unless we are a server, then craft a logfile name
		       (conc tl "/logs/server-" (current-process-id) "-" (get-host-name) ".log")))
	     (oup  (open-logfile logf)))
	(if (not (args:get-arg "-log"))
	    (hash-table-set! args:arg-hash "-log" logf)) ;; fake out future queries of -log

(if (args:get-arg "-cleanup-db")
    (begin
      (if (not (launch:setup))
	  (begin
	    (debug:print 0 *default-log-port* "Failed to setup, exiting") 
	    (exit 1)))
      (let ((dbstruct (db:setup *toppath*)))
      (let ((dbstruct (db:setup #f areapath: *toppath*)))
        (common:cleanup-db dbstruct))
      (set! *didsomething* #t)))

(if (args:get-arg "-mark-incompletes")
    (begin
      (if (not (launch:setup))
	  (begin

(if (or (getenv "MT_RUNSCRIPT")
	(args:get-arg "-repl")
	(args:get-arg "-load"))
    (let* ((toppath (launch:setup))
	   (dbstruct (if (and toppath
                              (common:on-homehost?))
                         (db:setup)
                         (db:setup #t)
                         #f))) ;; make-dbr:dbstruct path: toppath local: (args:get-arg "-local")) #f)))
      (if *toppath*
	  (cond
	   ((getenv "MT_RUNSCRIPT")
	    ;; How to run megatest scripts
	    ;;
	    ;; #!/bin/bash

;; ;; ;; redo me        (db:close-all dbstruct)
;; ;; ;; redo me        (list "uname" "rundir" "final_logf" "comment"))
;; ;; ;; redo me       (set! *didsomething* #t)))

(if (args:get-arg "-import-megatest.db")
    (begin
      (db:multi-db-sync 
       (db:setup)
       (db:setup #f)
       'killservers
       'dejunk
       'adj-testids
       'old2new
       ;; 'new2old
       )
      (set! *didsomething* #t)))

(if (args:get-arg "-sync-to-megatest.db")
    (begin
      (db:multi-db-sync 
       (db:setup)
       (db:setup #f)
       'new2old
       )
      (set! *didsomething* #t)))

(if (args:get-arg "-sync-to")
    (let ((toppath (launch:setup)))
      (tasks:sync-to-postgres *configdat* (args:get-arg "-sync-to"))

			     (loop (car tal)(cdr tal) newmax-cmd currmax)))))))
    (mutex-unlock! *db-stats-mutex*)
    res))

(define (rmt:open-qry-close-locally cmd run-id params #!key (remretries 5))
  (let* ((qry-is-write   (not (member cmd api:read-only-queries)))
	 (db-file-path   (db:dbfile-path)) ;;  0))
	 (dbstruct-local (db:setup))  ;; make-dbr:dbstruct path:  dbdir local: #t)))
	 (dbstruct-local (db:setup #t))  ;; make-dbr:dbstruct path:  dbdir local: #t)))
	 (read-only      (not (file-write-access? db-file-path)))
	 (start          (current-milliseconds))
	 (resdat         (if (not (and read-only qry-is-write))
			     (let ((v (api:execute-requests dbstruct-local (vector (symbol->string cmd) params))))
			       (handle-exceptions ;; there has been a long history of receiving strange errors from values returned by the client when things go wrong..
				exn               ;;  This is an attempt to detect that situation and recover gracefully
				(begin

#!/bin/bash

prefix=$1
cmd=$2
target=$3
cfgfile="$prefix/bin/.$(lsb_release -sr)/cfg.sh"

if [ "$LD_LIBRARY_PATH" != "" ];then
  cfgfile="$prefix/bin/.$(lsb_release -sr)/cfg.sh"
  echo "INFO: Using LD_LIBRARY_PATH=$LD_LIBRARY_PATH" >&2
( cat << __EOF
if [ "\$LD_LIBRARY_PATH" != "" ];then
  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:\$LD_LIBRARY_PATH
else
  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH
fi
__EOF
) > $cfgfile
  echo 
else
  echo "INFO: LD_LIBRARY_PATH not set" >&2
fi

# echo "#!/bin/bash" > $target
# if [ "$LD_LIBRARY_PATH" != "" ];then
#   echo "source $prefix/bin/.\$(lsb_release -sr)/cfg.sh" >> $target
# fi
# echo "exec $prefix/bin/.\$(lsb_release -sr)/$cmd \"\$@\"" >> $target
echo "#!/bin/bash" > $target

if [[ $cmd =~ dboard ]]; then
    cat >> $target <<'EOF'

# # disable if not running on homehost
# if [[ -e .homehost ]]; then

    exit 1
  fi
fi
EOF

fi

# echo "#!/bin/bash" > $target
# echo "exec $prefix/bin/.\$(lsb_release -sr)/$cmd \"\$@\"" >> $target

echo "lsbr=\$(lsb_release -sr)" >> $target
if [ "$LD_LIBRARY_PATH" != "" ];then
  echo "source $prefix/bin/.\$lsbr/cfg.sh" >> $target
fi

echo "if [[ -e \$lsbr ]];then source \$lsbr;fi" >> $target
# echo "if [[ -e \$lsbr ]];then source \$lsbr;fi" >> $target
echo "exec $prefix/bin/.\$lsbr/$cmd \"\$@\"" >> $target