Megatest

(define *pkts-info*    (make-hash-table)) ;; store stuff like the last parent here
(define *configinfo*   #f)   ;; raw results from setup, includes toppath and table from megatest.config
(define *runconfigdat* #f)   ;; run configs data
(define *configdat*    #f)   ;; megatest.config data
(define *configstatus* #f)   ;; status of data; 'fulldata : all processing done, #f : no data yet, 'partialdata : partial read done
(define *toppath*      #f)
(define *dbdir* ".mtdb")
(define *already-seen-runconfig-info* #f)

(define *test-meta-updated* (make-hash-table))
(define *globalexitstatus*  0) ;; attempt to work around possible thread issues
(define *passnum*           0) ;; when running track calls to run-tests or similar
;; (define *alt-log-file* #f)  ;; used by -log
;; (define *common:denoise*    (make-hash-table)) ;; for low noise printing
(define *default-log-port*  (current-error-port))
(define *time-zero* (current-seconds)) ;; for the watchdog
(define *default-area-tag* "local")

;; DATABASE
;; (define *dbstruct-dbs*         #f) ;; used to cache the dbstruct in db:setup. Goal is to remove this.
;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))
;; db access
(define *db-last-access*      (current-seconds)) ;; last db access, used in server
;; (define *db-write-access*     #t)
;; db sync
;; (define *db-last-sync*        0)                 ;; last time the sync to megatest.db happened
;; (define *db-last-sync*        0)                 ;; last time the sync to nfs db happened
(define *db-sync-in-progress* #f)                ;; if there is a sync in progress do not try to start another
;; (define *db-multi-sync-mutex* (make-mutex))      ;; protect access to *db-sync-in-progress*, *db-last-sync*
;; task db
(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
;; (define *db-access-mutex*     (make-mutex)) ;; moved to dbfile
(define *db-transaction-mutex* (make-mutex))

		     (handle-exceptions
		      exn
		      (debug:print-error 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 (common:exit-on-version-changed)
  (if (common:on-homehost?)
      (if (common:api-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 #t))) ;; (db:setup-db *dbstruct-dbs* *toppath* #f))) ;;  #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 (common:file-exists? mtconf) (common:file-exists? dbfile) (not read-only)
                   (eq? (current-user-id)(file-owner mtconf))) ;; safe to run -cleanup-db
              (debug:print 0 *default-log-port* "   I see you are the owner of megatest.config, attempting to cleanup and reset to new version")
              (handle-exceptions
               exn
               (begin
                 (debug:print 0 *default-log-port* "Failed to switch versions. exn=" exn)
                 (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
                 (print-call-chain (current-error-port))
                 (exit 1))
               (common:cleanup-db dbstruct)))
             ((not (common:file-exists? mtconf))
              (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (common:file-exists? dbfile))
              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             ((not (eq? (current-user-id)(file-owner mtconf)))
              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (read-only
              (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
              (exit 1))
             (else
              (debug:print 0 *default-log-port* " to switch versions you can run: \"megatest -cleanup-db\"")
              (exit 1)))))))
;;======================================================================
;;      (begin
;;	(debug:print 0 *default-log-port* "ERROR: cannot migrate version unless on homehost. Exiting.")
;;	(exit 1))))

;;======================================================================
;; S P A R S E   A R R A Y S

					  (string-translate *toppath* "/" "."))
				    (conc "/tmp/" (current-process-id) ;; just in case we have an issue with the dir by own user name
					  "/megatest_localdb/"
					  tsname
					  (string-translate *toppath* "/" "."))
				    ))))
		(set! *db-cache-path* dbpath)
		;; ensure megatest area has .megatest
		(let ((dbarea (conc *toppath* "/.megatest")))
		;; ensure megatest area has dbdir
		(let ((dbarea (conc *toppath* "/" *dbdir*)))
		  (if (not (file-exists? dbarea))
		      (create-directory dbarea)))
		;; ensure tmp area has .megatest
		(let ((dbarea (conc dbpath "/.megatest")))
		;; ensure tmp area has dbdir
		(let ((dbarea (conc dbpath "/" *dbdir*)))
		  (if (not (file-exists? dbarea))
		      (create-directory dbarea)))
		dbpath))
	  #f)))

(define (common:get-area-path-signature)
  (message-digest-string (md5-primitive) *toppath*))

                        0
                        (dboard:rundat-last-update run-dat)))
	 (last-db-time (if do-not-use-db-file-timestamps
			   0
			   (dboard:rundat-last-db-time run-dat)))
	 (db-path      (or (dboard:rundat-db-path run-dat)
			   (let* ((db-dir (common:get-db-tmp-area))
				  (db-pth (conc db-dir "/.megatest/main.db")))
				  (db-pth (conc db-dir "/" *dbdir* "/main.db")))
			     (dboard:rundat-db-path-set! run-dat db-pth)
			     db-pth)))
	 (db-mod-time  (common:lazy-sqlite-db-modification-time db-path))
	 (db-modified  (>= db-mod-time last-db-time))
	 (multi-get    (> (dboard:rundat-run-data-offset run-dat) 0))  ;; multi-get in progress
	 (tmptests     (if (or do-not-use-db-file-timestamps
			       (dboard:tabdat-filters-changed tabdat)

;;======================================================================

(stop-the-train)

(define (main)
  ;; (print "Starting dashboard main")
    
  (let* ((mtdb-path (conc *toppath* "/.megatest/main.db"))
  (let* ((mtdb-path (conc *toppath* "/" *dbdir* "/main.db"))
         (target (args:get-arg "-target"))
         (commondat       (dboard:commondat-make)))
    (if target
        (begin
          (args:remove-arg-from-ht "-target")
          (dboard:commondat-target-set! commondat target)
        )

    (if (not (common:on-homehost?))
    (begin
      (debug:print 0 *default-log-port* "WARNING: You are starting the dashboard on a machine that is not the homehost:" (common:get-homehost))
      (debug:print 0 *default-log-port* "It will be slower.")
      ))


    (if (and (common:file-exists? mtdb-path)
	     (file-write-access? mtdb-path))
	(if (not (args:get-arg "-skip-version-check"))
            (common:exit-on-version-changed)))

    (let* ()
      ;; Move this stuff to db.scm? I'm not sure that is the right thing to do...
      (cond 
       ((args:get-arg "-test") ;; run-id,test-id
	(let* ((dat     (let ((d (map string->number (string-split (args:get-arg "-test") ",")))) 
			  (if (> (length d) 1)
			      d

(define last-copy-time 0)


;; Sync to tmp only if in read-only mode.

(define (sync-db-to-tmp tabdat)
  (let* ((db-file "./.megatest/main.db"))
  (let* ((db-file (conc "./" *dbdir* "/main.db")))
    (if (and (not (file-write-access? db-file)) ( > (current-seconds) (+ last-copy-time 5)))
      (begin
        (db:multi-db-sync (db:setup #f) 'old2new)
        (set! last-copy-time (current-seconds))
      )
    )
  )

(define (db:all-db-sync dbstruct)
  (let* ((dbdat (db:open-db dbstruct #f db:initialize-main-db))
	 (data-synced       0) ;; count of changed records
    (tmp-area       (common:get-db-tmp-area))
    (dbfiles        (glob (conc tmp-area"/.megatest/*.db")))
    (dbfiles        (glob (conc tmp-area"/" *dbdir* "/*.db")))
    (sync-durations (make-hash-table))
    (no-sync-db        (db:open-no-sync-db)))
    (for-each
     (lambda (file) ;; tmp db file
       (debug:print-info 3 *default-log-port* "file: " file)
       (let* ((fname       (conc (pathname-file file) ".db")) ;; fname is tmp db file
              (wal-file (conc file "-wal"))
              (shm-file (conc file "-shm"))
	      (fulln       (conc *toppath*"/.megatest/"fname)) ;; fulln is nfs db name
	      (fulln       (conc *toppath*"/" *dbdir* "/"fname)) ;; fulln is nfs db name
              (wal-time     (if (file-exists? wal-file)             
			       (file-modification-time wal-file)
                               0))
              (shm-time     (if (file-exists? shm-file)             
			       (file-modification-time shm-file)
                               0))

    (tmp-area       (common:get-db-tmp-area))
    (old2new (member 'old2new options))
    (dejunk (member 'dejunk options))
    (killservers (member 'killservers options))
    (servers (server:get-list *toppath*))
    (src-area (if old2new *toppath* tmp-area))
    (dest-area (if old2new tmp-area *toppath*))
    (dbfiles        (if old2new (glob (conc *toppath* "/.megatest/*.db")) (glob (conc tmp-area "/.megatest/*.db"))))
    (dbfiles        (if old2new (glob (conc *toppath* "/" *dbdir* "/*.db")) (glob (conc tmp-area "/" *dbdir* "/*.db"))))
    (keys (db:get-keys dbstruct))
    (sync-durations (make-hash-table)))


    (if killservers
      (begin
       	  (for-each

    )
    (for-each
     (lambda (srcfile)
       (debug:print-info 3 *default-log-port* "file: " srcfile)
       (let* ((fname (conc (pathname-file srcfile) ".db"))
              (basename (pathname-file srcfile))
              (run-id (if (string= basename "main") #f (string->number basename)))
	      (destfile (conc dest-area "/.megatest/" fname))
              (dest-directory  (conc dest-area "/.megatest/"))
	      (destfile (conc dest-area "/" *dbdir* "/" fname))
              (dest-directory  (conc dest-area "/" *dbdir* "/"))
              (dummy (debug:print-info 2 *default-log-port* "destfile = " destfile))
              (dummy2 (debug:print-info 2 *default-log-port* "dejunk = " dejunk))
              ;; TODO: time1 and time2 need to take into account -wal and -shm files
	      (time1 (file-modification-time srcfile))
              (time2 (if (file-exists? destfile)
                         (begin
                            (debug:print-info 2 *default-log-port* "destfile " destfile " exists")

 (define (db:get-changed-run-ids since-time)
  (let* ((dbdir      (db:dbfile-path)) ;; (configf:lookup *configdat* "setup" "dbdir"))
        (alldbs     (glob (conc dbdir "/.megatest/[0-9]*.db*")))
        (alldbs     (glob (conc dbdir "/" *dbdir* "/[0-9]*.db*")))
        (changed    (filter (lambda (dbfile)
                              (> (file-modification-time dbfile) since-time))
                            alldbs)))
    (delete-duplicates
     (map (lambda (dbfile)
           (let* ((res (string-match ".*\\/(\\d\\d)\\.db*" dbfile)))
             (if res

         (changed_run_dbs (db:get-changed-run-ids since-time)) ;; gets the rundb numbers
         (all_run_ids 
          (db:with-db dbstruct #f #f 
            (lambda (dbdat db)
              (sqlite3:fold-row backcons '() db "SELECT id FROM runs"))
          )
         )
         (changed_run_ids (filter (lambda (run) (member (modulo run 100) changed_run_dbs)) all_run_ids))
         (changed_run_ids (filter (lambda (run) (member (modulo run (num-run-dbs)) changed_run_dbs)) all_run_ids))
         ;; TODO: couldn't we just use changed_run_ids for run_ids?
         (run_ids 
          (db:with-db dbstruct #f #f 
            (lambda (dbdat db)
              (sqlite3:fold-row backcons '() db "SELECT id FROM runs  WHERE last_update>=?" since-time))
          )
         )

	  #f
        ))))

;; sync for filesystem local db writes
;;
(define (db:run-lock-and-sync no-sync-db)
  (let* ((tmp-area       (common:get-db-tmp-area))
	 (dbfiles        (glob (conc tmp-area"/.megatest/*.db")))
	 (dbfiles        (glob (conc tmp-area"/" *dbdir* "/*.db")))
	 (sync-durations (make-hash-table)))
    ;; (debug:print-info 0 *default-log-port* "lock-and-sync, dbfiles: "dbfiles)
    (for-each
     (lambda (file)
       (let* ((fname (conc (pathname-file file) ".db"))
	      (fulln (conc *toppath*"/.megatest/"fname))
	      (fulln (conc *toppath*"/" *dbdir* "/"fname))
	      (time1 (if (file-exists? file)
			 (file-modification-time file)
			 (begin
			   (debug:print-info 0 *default-log-port* "Sync - I do not see file "file)
			   1)))
	      (time2 (if (file-exists? fulln)
			 (file-modification-time fulln)

	ports

	commonmod
	)

;; (import debugprint)

;; Parameters

(define num-run-dbs           (make-parameter 10))

;;======================================================================
;;  R E C O R D S
;;======================================================================

;; a single Megatest area with it's multiple dbs is
;; managed in a dbstruct
;;
(defstruct dbr:dbstruct
  (areapath  #f)
  (homehost  #f)
  (tmppath   #f)
  (read-only #f)
  (subdbs (make-hash-table))
  )

;; NOTE: Need one dbr:subdb per main.db, 1.db ...
;;
(defstruct dbr:subdb
  (dbname      #f) ;; .megatest/1.db
  (mtdbfile    #f) ;; mtrah/.megatest/1.db
  (dbname      #f) ;; " *dbdir* "/1.db
  (mtdbfile    #f) ;; mtrah/" *dbdir* "/1.db
  (mtdbdat     #f) ;; only need one of these for syncing
  ;; (dbdats      (make-hash-table))  ;; id => dbdat 
  (tmpdbfile   #f) ;; /tmp/.../.megatest/1.db
  ;; (refndbfile  #f) ;; /tmp/.../.megatest/1.db_ref
  (tmpdbfile   #f) ;; /tmp/.../" *dbdir* "/1.db
  ;; (refndbfile  #f) ;; /tmp/.../" *dbdir* "/1.db_ref
  (dbstack     (make-stack)) ;; stack for tmp dbr:dbdat,
  (homehost    #f) ;; not used yet
  (on-homehost #f) ;; not used yet
  (read-only   #f)
  (last-sync   0)
  (last-write  (current-seconds))
  )                ;; goal is to converge on one struct for an area but for now it is too confusing

(define *db-sync-in-progress* #f)
(define *db-with-db-mutex*    (make-mutex))
(define *max-api-process-requests* 0)
(define *api-process-request-count* 0)
(define *db-write-access*     #t)
(define *db-last-sync*        0)                 ;; last time the sync to megatest.db happened
(define *db-multi-sync-mutex* (make-mutex))      ;; protect access to *db-sync-in-progress*, *db-last-sync*
(define *dbdir* ".mtdb")

(define (db:generic-error-printout exn . message)
  (print-call-chain (current-error-port))
  (apply dbfile:print-err message)
  (dbfile:print-err
    ", error: "     ((condition-property-accessor 'exn 'message)   exn)
    ", arguments: " ((condition-property-accessor 'exn 'arguments) exn)

(define (db:dbname->path apath dbname)
  (conc apath"/"dbname))

;; POTENTIAL BUG: this implementation could produce a db file if run-id is neither #f or a number
(define (dbfile:run-id->dbname run-id)
  (cond
   ((number? run-id) (conc ".megatest/" (modulo run-id 100) ".db"))
   ((not run-id)     (conc ".megatest/main.db"))
   ((number? run-id) (conc  *dbdir* "/" (modulo run-id (num-run-dbs)) ".db"))
   ((not run-id)     (conc  *dbdir* "/main.db"))
   (else             run-id)))

;; 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 (dbfile:setup do-sync areapath tmppath)

	        (sqlite3:for-each-row
	          (lambda (a . b)
	            (set! fromdat (cons (apply vector a b) fromdat))
	            (if (> (length fromdat) batch-len)
		      (begin
		        (set! fromdats (cons fromdat fromdats))
		        (set! fromdat  '())
		        (set! totrecords (+ totrecords 1)))
		        (set! totrecords (+ totrecords 1))
                        (thread-sleep! 2)
                      )
                    )
                 )
	         (dbr:dbdat-dbh fromdb)
	         full-sel)
              )
            )

;;     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 1.7014)
(define megatest-version 1.7101)

     (else
      (if (not (car *configinfo*))
	  (begin
	    (debug:print-error 0 *default-log-port* "Attempted " action "on test(s) but run area config file not found")
	    (exit 1))
	  ;; put test parameters into convenient variables
	  (begin
	    ;; check for correct version, exit with message if not correct
	    (common:exit-on-version-changed)
	    (runs:operate-on  action
			      target
			      runname
			      testpatt
			      state:  (common:args-get-state)
			      status: (common:args-get-status)
			      new-state-status: (args:get-arg "-set-state-status")

		;; 				exn
		;; 				(begin
		;; 				  (print-call-chain)
		;; 				  (print " message: " ((condition-property-accessor 'exn 'message) exn)))
		;; 			      (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests
		;; 						    (any->number reglen) all-tests-registry)))
		;; 			  "runs:run-tests-queue"))
		 (th2        (make-thread (lambda ()			 ;; BBQ: why are we visiting ALL runs here?	    
		 #;(th2        (make-thread (lambda ()			 ;; BBQ: why are we visiting ALL runs here?	    
					    ;; (rmt:find-and-mark-incomplete-all-runs))))) CAN'T INTERRUPT IT ...
					    (let ((run-ids (rmt:get-all-run-ids)))
					      (for-each (lambda (run-id)
							  (if keep-going
							      (handle-exceptions
							       exn
							       (debug:print 0 *default-log-port* "error in calling find-and-mark-incomplete for run-id " run-id ", exn=" exn)
							       (rmt:find-and-mark-incomplete run-id #f)))) ;; ovr-deadtime))) ;; could be root of https://hsdes.intel.com/appstore/article/#/220546828/main -- Title: Megatest jobs show DEAD even though they are still running (1.64/27)
							run-ids)))
					  "runs: mark-incompletes")))
	    ;; (thread-start! th1)
	    (thread-start! th2)
	    ;; (thread-start! th2)
	    ;; (thread-join! th1)
	    ;; just do the main stuff in the main thread
	    (runs:run-tests-queue run-id runname test-records keyvals flags test-patts required-tests
								    (any->number reglen) all-tests-registry)
	    (set! keep-going #f)
	    (thread-join! th2)
	    #;(thread-join! th2)
	    ;; if run-count > 0 call, set -preclean and -rerun STUCK/DEAD
	    (if (> run-count 0) ;; handle reruns
		(begin
		  (if (not (hash-table-ref/default flags "-preclean" #f))
		      (hash-table-set! flags "-preclean" #t))
		  (if (not (hash-table-ref/default flags "-rerun" #f))
		      (hash-table-set! flags "-rerun" "ABORT,STUCK/DEAD,n/a,ZERO_ITEMS"))

	 (state-status (if (string? new-state-status) (string-split new-state-status ",") '(#f #f)))
	 (rp-mutex     (make-mutex))
	 (bup-mutex    (make-mutex))
         (keep-records (args:get-arg "-keep-records")) ;; used in conjunction with -remove-runs to keep the records, TODO: consolidate this with "mode".
	 (test-records '())) ;; for tasks that we wish to operate on all tests in one fell swoop

    (let* ((write-access-actions '(remove-runs set-state-status archive run-wait kill-runs))
           (dbfile             (conc  *toppath* "/.megatest/main.db"))
           (dbfile             (conc  *toppath* "/" *dbdir* "/main.db"))
           (readonly-mode      (not (file-write-access? dbfile))))
      (when (and readonly-mode
                 (member action write-access-actions))
        (debug:print-error 0 *default-log-port* dbfile " is readonly.  Cannot proceed with action ["action"] in which write-access isrequired .")
        (exit 1)))
    
    (debug:print-info 4 *default-log-port* "runs:operate-on => Header: " header " action: " action " new-state-status: " new-state-status)