Megatest

;; 	(mutex-lock! *rundb-mutex*)
;; 	(if (eq? mod-read 'mod)
;; 	    (dbr:dbstruct-mtime-set! dbstruct (current-milliseconds))
;; 	    (dbr:dbstruct-rtime-set! dbstruct (current-milliseconds)))
;; 	(dbr:dbstruct-inuse-set! dbstruct #f)
;; 	(mutex-unlock! *rundb-mutex*))))

(define-inline (db:generic-error-printout exn . message)
  (print-call-chain (current-error-port))
  (apply debug:print-error 0 *default-log-port* message)
  (debug:print-error 0 *default-log-port* "   params: " params
		     ", error: "     ((condition-property-accessor 'exn 'message)   exn)
		     ", arguments: " ((condition-property-accessor 'exn 'arguments) exn)
		     ", location: "  ((condition-property-accessor 'exn 'location)  exn)
		     ))

(print-call-chain (current-error-port))
;; (db:with-db dbstruct run-id sqlite3:exec "select blah fgrom blaz;")
;; r/w is a flag to indicate if the db is modified by this query #t = yes, #f = no
;;
(define (db:with-db dbstruct run-id r/w proc . params)
  (let* ((have-struct (dbr:dbstruct? dbstruct))
         (dbdat     (if have-struct 
			(db:get-db dbstruct)
			#f))
	 (db        (if have-struct
			(db:dbdat-get-db dbdat)
			dbstruct))
	 (fname     (db:dbdat-get-path dbdat))
	 (use-mutex (> *api-process-request-count* 25))) ;; was 25
    (if (and use-mutex
	     (common:low-noise-print 120 "over-50-parallel-api-requests"))
	(debug:print-info 0 *default-log-port* *api-process-request-count* " parallel api requests being processed in process " (current-process-id) ", throttling access"))
    (if (common:low-noise-print 600 (conc "parallel-api-requests" *max-api-process-requests*))
	(debug:print-info 2 *default-log-port* "Parallel api request count: " *api-process-request-count* " max parallel requests: " *max-api-process-requests*))
    (condition-case
     (begin
    ;;;;;;;;; (handle-exceptions
    ;;;;;;;;;  exn
    ;;;;;;;;;  (begin
    ;;;;;;;;;    (print-call-chain (current-error-port))
    ;;;;;;;;;    (debug:print-error 0 *default-log-port* "sqlite3 issue in db:with-db, dbstruct=" dbstruct ", run-id=" run-id ", proc=" proc ", params=" params " error: " ((condition-property-accessor 'exn 'message) exn))
    ;;;;;;;;;    ;; there is no recovering at this time. exit
    ;;;;;;;;;    (exit 50))
       (if use-mutex (mutex-lock! *db-with-db-mutex*))
       (let ((res (apply proc db params)))
	 (if use-mutex (mutex-unlock! *db-with-db-mutex*))
	 ;; (if (vector? dbstruct)(db:done-with dbstruct run-id r/w))
	 (if dbdat (stack-push! (dbr:dbstruct-dbstack dbstruct) dbdat))
	 res))
     (exn (io-error)
	  (db:generic-error-printout exn "ERROR: i/o error with " fname ". Check permissions, disk space etc. and try again."))
     (exn (corrupt)
	  (db:generic-error-printout exn "ERROR: database " fname " is corrupt. Repair it to proceed."))
     (exn (busy)
	  (db:generic-error-printout exn "ERROR: database " fname
				     " is locked. Try copying to another location, remove original and copy back."))
     (exn (permission)(db:generic-error-printout exn "ERROR: database " fname " has some permissions problem."))
     (exn ()
	  (db:generic-error-printout exn "ERROR: Unknown error with database " fname " message: "
		       ((condition-property-accessor 'exn 'message) exn))))))
      
;;======================================================================
;; K E E P   F I L E D B   I N   dbstruct
;;======================================================================

;; (define (db:get-filedb dbstruct run-id)
;;   (let ((db (vector-ref dbstruct 2)))
;;     (if db

     (let ((stmts       (make-hash-table)) ;; table-field => stmt
	   (all-stmts   '())              ;; ( ( stmt1 value1 ) ( stml2 value2 ))
	   (numrecs     (make-hash-table))
	   (start-time  (current-milliseconds))
	   (tot-count   0))
       (for-each ;; table
	(lambda (tabledat)

	  (let* ((tablename        (car tabledat))
		 (fields           (cdr tabledat))
		 (has-last-update  (member "last_update" fields))
		 (use-last-update  (cond
				    ((and has-last-update
					  (member "last_update" fields))
				     #t) ;; if given a number, just use it for all fields

              (thread-sleep! delay-handicap)
              (debug:print-info 0 *default-log-port* "synthetic sync delay of "delay-handicap" seconds completed")
              )
            
	    ;; first pass implementation, just insert all changed rows
	    (for-each 
	     (lambda (targdb)

	       (let* ((db     (db:dbdat-get-db targdb))
                      (drp-trigger (if (member "last_update" field-names)
                                      (db:drop-trigger db tablename) 
                                       #f))
                       (is-trigger-dropped (if (member "last_update" field-names)
                                              (db:is-trigger-dropped db tablename) #f)) 
		      (stmth  (sqlite3:prepare db full-ins)))

	;; archive disk areas, cached info from [archive-disks]
	(sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_disks (
                                id INTEGER PRIMARY KEY,
                                archive_area_name TEXT,
                                disk_path TEXT,
                                last_df INTEGER DEFAULT -1,
                                last_df_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                creation_time TIMESTAMP DEFAULT (strftime('%s','now')));")
	;; individual bup (or tar) data chunks
	(sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_blocks (
                                id INTEGER PRIMARY KEY,
                                archive_disk_id INTEGER,
                                disk_path TEXT,
                                last_du INTEGER DEFAULT -1,
                                last_du_time TIMESTAMP DEFAULT (strftime('%s','now')),
                                creation_time TIMESTAMP DEFAULT (strftime('%s','now')));")
	;; tests allocated to what chunks. reusing a chunk for a test/item_path is very efficient
	;; NB// the per run/test recording of where the archive is stored is done in the test
	;;      record. 
	(sqlite3:execute db "CREATE TABLE IF NOT EXISTS archive_allocations (
                                id INTEGER PRIMARY KEY,
                                archive_block_id INTEGER,
                                testname TEXT,
                                item_path TEXT,
                                creation_time TIMESTAMP DEFAULT (strftime('%s','now')));")
	;; move this clean up call somewhere else
	(sqlite3:execute db "DELETE FROM tasks_queue WHERE state='done' AND creation_time < ?;" (- (current-seconds)(* 24 60 60))) ;; remove older than 24 hrs
	(sqlite3:execute db (conc "CREATE INDEX IF NOT EXISTS runs_index ON runs (runname" (if havekeys "," "") keystr ");"))
	;; (sqlite3:execute db "CREATE VIEW runs_tests AS SELECT * FROM runs INNER JOIN tests ON runs.id=tests.run_id;")
	(sqlite3:execute db "CREATE TABLE IF NOT EXISTS extradat (id INTEGER PRIMARY KEY, run_id INTEGER, key TEXT, val TEXT);")
	(sqlite3:execute db "CREATE TABLE IF NOT EXISTS metadat (id INTEGER PRIMARY KEY, var TEXT, val TEXT,
                                  CONSTRAINT metadat_constraint UNIQUE (var));")

         )
    (debug:print-info 4  *default-log-port* "running-deadtime = " running-deadtime)
    (debug:print-info 4  *default-log-port* "deadtime-trim = " deadtime-trim)

    (db:with-db 
     dbstruct #f #f
     (lambda (db)
       (let* ((stmth1 (db:get-cache-stmth
		       dbstruct db
		       "SELECT id,rundir,uname,testname,item_path,event_time,run_duration FROM tests 
                           WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?)
                                          AND state IN ('RUNNING');"))
	      (stmth2 (db:get-cache-stmth
		       dbstruct db
		       "SELECT id,rundir,uname,testname,item_path,event_time,run_duration FROM tests 
                           WHERE run_id=? AND (strftime('%s','now') - event_time) > (run_duration + ?)
                                          AND state IN ('REMOTEHOSTSTART');"))
	      (stmth3 (db:get-cache-stmth
		       dbstruct db
		       "SELECT id,rundir,uname,testname,item_path FROM tests
                           WHERE run_id=? AND (strftime('%s','now') - event_time) > 86400
                                          AND state IN ('LAUNCHED');")))
	 ;; in RUNNING or REMOTEHOSTSTART for more than 10 minutes
	 ;;
	 ;; 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 (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 (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
	  (lambda (test-id run-dir uname testname item-path)
	    (if (and (equal? uname "n/a")
		     (equal? item-path "")) ;; this is a toplevel test
		;; what to do with toplevel? call rollup?
		(set! toplevels   (cons (list test-id run-dir uname testname item-path run-id) toplevels))
		(begin
		  (debug:print-info 0 *default-log-port* "Found old test in LAUNCHED state, test-id=" test-id
				    " 1 day since event_time marked")
                (set! oldlaunched (cons (list test-id run-dir uname testname item-path run-id) oldlaunched)))))


	  stmth3
	  run-id)
       
	 (debug:print-info 18 *default-log-port* "Found " (length oldlaunched) " old LAUNCHED items, "
			   (length toplevels) " old LAUNCHED toplevel tests and "
			   (length incompleted) " tests marked RUNNING but apparently dead."))

	 ;; These are defunct tests, do not do all the overhead of set-state-status. Force them to INCOMPLETE.
       ;;
	 ;; (db:delay-if-busy dbdat)







	 (let* ((min-incompleted-ids (map car incompleted)) ;; do 'em all
		(all-ids             (append min-incompleted-ids (map car oldlaunched))))
	   (if (> (length all-ids) 0)
	       (begin
		 ;; (launch:is-test-alive "localhost" 435)
		 (debug:print 0 *default-log-port* "WARNING: Marking test(s); " (string-intersperse (map conc all-ids) ", ")
			      " as DEAD")
		 (for-each
                  (lambda (test-id)
                    (let* (;; (run-dir (db:test-get-rundir-from-test-id dbstruct run-id test-id))
			   (tinfo   (db:get-test-info-by-id dbstruct run-id test-id))
			   (run-dir (db:test-get-rundir     tinfo))
			   (host    (db:test-get-host       tinfo))
			   (pid     (db:test-get-process_id tinfo))
			   (result (db:get-status-from-final-status-file run-dir)))
		      (if (and (list? result) (> (length result) 1) (equal? "PASS" (cadr result)) (equal? "COMPLETED" (car result))) 
			  (begin
			    (debug:print 0 *default-log-port* "INFO: test " test-id " actually passed, so marking PASS not DEAD")
			    (db:set-state-status-and-roll-up-items
			     dbstruct run-id test-id 'foo "COMPLETED" "PASS"
			     "Test stopped responding but it has PASSED; marking it PASS in the DB."))
			  (let ((is-alive (and (not (eq? pid 0))  ;; 0 is default in re-used field "attemptnum" where pid stored.
					       (launch:is-test-alive host pid))))
			    (if is-alive
				(debug:print 0 *default-log-port* "INFO: test " test-id " on host " host
					     " has a process on pid " pid ", NOT setting to DEAD.")
				(begin
				  (debug:print 0 *default-log-port* "INFO: test " test-id
					       " final state/status is not COMPLETED/PASS. It is " result)
				  (db:set-state-status-and-roll-up-items
				   dbstruct run-id test-id 'foo "COMPLETED" "DEAD"
				   "Test stopped responding while in RUNNING or REMOTEHOSTSTART; presumed dead.")))))))
		  ;; call end of eud of run detection for posthook - from merge, is it needed?
		  ;; (launch:end-of-run-check run-id)
		  all-ids)
		 ;;call end of eud of run detection for posthook
		 (launch:end-of-run-check run-id)
		 )))))))





















;; BUG: Probably broken - does not explicitly use run-id in the query
;;
(define (db:top-test-set-per-pf-counts dbstruct run-id test-name)
  (db:general-call dbstruct 'top-test-set-per-pf-counts (list test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name test-name)))

;; Clean out old junk and vacuum the database

		       (lambda (key)
			 (set! res (cons key res)))
		       db
		       "SELECT fieldname FROM keys ORDER BY id DESC;")))
	(set! *db-keys* res)
	res)))

;; extract index number given a header/data structure
(define (db:get-index-by-header header field)
  (list-index (lambda (x)(equal? x field)) header))

;; look up values in a header/data structure
(define (db:get-value-by-header row header field)
  (if (or (null? header) (not row))
      #f
      (let loop ((hed (car header))
                 (tal (cdr header))
                 (n   0))

(define-record simple-run target id runname state status owner event_time)
(define-record-printer (simple-run x out)
  (fprintf out "#,(simple-run ~S ~S ~S ~S)"
	   (simple-run-target x) (simple-run-id x) (simple-run-runname x) (time->string (seconds->local-time (simple-run-event_time x) ))))

;; simple get-runs
;;
(define (db:simple-get-runs dbstruct runpatt count offset target last-update)
    (let* ((res       '())
	   (keys       (db:get-keys dbstruct))
	   (runpattstr (db:patt->like "runname" runpatt))
	   (remfields  (list "id" "runname" "state" "status" "owner" "event_time"))
	   (targstr    (string-intersperse keys "||'/'||"))
	   (keystr     (conc targstr " AS target,"
			     (string-intersperse remfields ",")))
	   (qrystr     (conc "SELECT " keystr " FROM runs WHERE (" runpattstr ") " ;; runname LIKE ? "
			     ;; Generate: " AND x LIKE 'keypatt' ..."
			     " AND target LIKE '" target "'"
			     " AND state != 'deleted' "
			     (if (number? last-update)
				 (conc " AND last_update >= " last-update)
				 "")
			     " ORDER BY event_time DESC "
			     (if (number? count)
				 (conc " LIMIT " count)
				 "")
			     (if (number? offset)
				 (conc " OFFSET " offset)
				 "")))
	   )
    (debug:print-info 11 *default-log-port* "db:get-runs START qrystr: " qrystr " target: " target " offset: " offset " limit: " count)
    (db:with-db dbstruct #f #f
		(lambda (db)		
		  (sqlite3:for-each-row
		   (lambda (target id runname state status owner event_time)
		     (set! res (cons (make-simple-run target id runname state status owner event_time) res)))
		   db

						     (conc " ORDER BY " sort-by " ")
						     " ")))
				(if sort-order sort-order " ")
				(if limit  (conc " LIMIT " limit)   " ")
				(if offset (conc " OFFSET " offset) " ")
				";"
				)))

    (debug:print-info 8 *default-log-port* "db:get-tests-for-run run-id=" run-id ", qry=" qry)
    (let* ((res (db:with-db dbstruct run-id #f
			    (lambda (db)
			      ;; (let* ((stmth (db:get-cache-stmth dbstruct db qry))) ;; due to use of last-update we can't efficiently cache this query
				(reverse
				 (sqlite3:fold-row
				  (lambda (res . row)

	(sqlite3:fold-row
	 (lambda (res id testname item-path state status)
	   ;;  id,run_id,testname,state,status,event_time,host,cpuload,diskfree,uname,rundir,item_path,run_duration,final_logf,comment
	   (cons (vector id run-id testname state status -1         ""     -1      -1       ""    "-"  item-path -1           "-"         "-") res))
	 '()
	 stmth
	 run-id))))))

(define (db:get-tests-for-run-state-status dbstruct run-id testpatt #!optional (last-update 0))
  (let* ((res            '())
	 (tests-match-qry (tests:match->sqlqry testpatt))
	 (qry             (conc "SELECT id,testname,item_path,state,status,event_time,run_duration FROM tests WHERE run_id=? " 
				" AND last_update > ? "
				(if tests-match-qry (conc " AND (" tests-match-qry ") ") "")
				)))
    (debug:print-info 8 *default-log-port* "db:get-tests-for-run qry=" qry)
    (db:with-db dbstruct run-id #f
		(lambda (db)
		  (sqlite3:fold-row
		   (lambda (res id testname item-path state status event-time run-duration)
		     ;;            id,run_id,testname,state,status,event_time,host,cpuload,diskfree,uname,rundir,item_path,run_duration,final_logf,comment
		     (cons (vector id run-id testname state status event-time  ""     -1      -1       ""    "-"  item-path run-duration  "-"         "-") res))
		   '()
		   db 
		   qry
		   run-id
		   (or last-update 0))))))

(define (db:get-testinfo-state-status dbstruct run-id test-id)
  (let ((res            #f))
    (db:with-db dbstruct run-id #f
		(lambda (db)
		  (sqlite3:for-each-row
		   (lambda (run-id testname item-path state status)

     #f
     (lambda (db)
       (let* ((stmth (db:get-cache-stmth dbstruct db qry)))
	 (sqlite3:first-result stmth run-id))))))

;; For a given testname how many items are running? Used to determine
;; probability for regenerating html
;;
(define (db:get-count-tests-running-for-testname dbstruct run-id testname)
  (db:with-db
   dbstruct
   run-id
   #f
   (lambda (db)
     (let* ((stmt "SELECT count(id) FROM tests WHERE state in ('RUNNING','LAUNCHED','REMOTEHOSTSTART') AND run_id=? AND NOT (uname = 'n/a' AND item_path = '') AND testname=?;")