Megatest

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

(declare (unit dbmod))
(declare (uses commonmod))
(declare (uses ods))
(declare (uses configfmod))
(declare (uses margsmod))

(module dbmod
	*
	
(import commonmod)
(import ods)
(import configfmod)
(import margsmod)

(import scheme chicken data-structures extras ports)
(import
  (prefix base64 base64:)
  (prefix sqlite3 sqlite3:)
  csv
  csv-xml

  regex
  s11n
  srfi-1
  srfi-13
  stack
  z3
  )

(import (prefix dbi dbi:)
	pkts)

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

(include "db_records.scm")

(define (db:writeable dbdir dbname)
  (let* ((dbfile (conc dbdir "/" dbname)))
    (if (file-exists? dbfile)
	(file-write-access? dbfile)
	(file-write-access? *toppath*))))

;;======================================================================the end

;; state is the priority rollup of all states
;; status is the priority rollup of all completed statesfu
;;
;; if test-name is an integer work off that instead of test-name test-path
;;
(define (db:set-state-status-and-roll-up-items dbstruct run-id test-name item-path state status comment)
  ;; establish info on incoming test followed by info on top level test
  ;; BBnote - for mode itemwait, linkage between upstream test & matching item status is propagated to run queue in db:prereqs-not-met
  (let* ((testdat      (if (number? test-name)
			   (db:get-test-info-by-id dbstruct run-id test-name) ;; test-name is actually a test-id
			   (db:get-test-info       dbstruct run-id test-name item-path)))
	 (test-id      (db:test-get-id testdat))
	 (test-name    (if (number? test-name)
			   (db:test-get-testname testdat)
			   test-name))
	 (item-path    (db:test-get-item-path testdat))
         (tl-testdat   (db:get-test-info dbstruct run-id test-name ""))
         (tl-test-id   (if tl-testdat
			   (db:test-get-id tl-testdat)
			   #f)))
    (if (member state '("LAUNCHED" "REMOTEHOSTSTART")) 
	(db:general-call dbstruct 'set-test-start-time (list test-id)))
    (mutex-lock! *db-transaction-mutex*)
    (db:with-db
     dbstruct #f #f
     (lambda (db)
       (let ((tr-res
              (sqlite3:with-transaction
               db
               (lambda ()
                 ;; NB// Pass the db so it is part fo the transaction
                 (db:test-set-state-status db run-id test-id state status comment) ;; this call sets the item state/status
                 (if (not (equal? item-path "")) ;; only roll up IF incoming test is an item
                     (let* ((state-status-counts  (db:get-all-state-status-counts-for-test dbstruct run-id test-name item-path state status)) ;; item-path is used to exclude current state/status of THIS test
			      						  (state-stauses (db:roll-up-rules state-status-counts state status))
                          (newstate (car state-stauses))
                          (newstatus (cadr state-stauses)))
                       (debug:print 4 *default-log-port* "BB> tl-test-id="tl-test-id" ; "test-name":"item-path" newstate="newstate" newstatus="newstatus" len(sscs)="(length state-status-counts)  " state-status-counts: "
							(apply conc
                  (map (lambda (x)
                     (conc
                     		(with-output-to-string (lambda () (pp (dbr:counts->alist x)))) " | "))
                              state-status-counts))); end debug:print
   
                       (if tl-test-id
			   (db:test-set-state-status db run-id tl-test-id newstate newstatus #f)) ;; we are still in the transaction - must access the db and not the dbstruct
		       ))))))
         (mutex-unlock! *db-transaction-mutex*)
         (if (and test-id state status (equal? status "AUTO")) 
             (db:test-data-rollup dbstruct run-id test-id status))
         tr-res)))))

;; ;; speed up for common cases with a little logic
;; ;; NB// Ultimately this will be deprecated in deference to mt:test-set-state-status-by-id
;;
;;      NOTE: run-id is not used
;; ;;
(define (db:test-set-state-status dbstruct run-id test-id newstate newstatus newcomment)
  (db:with-db
   dbstruct
   ;; run-id
   #f
   #t
   (lambda (db)
     (cond
      ((and newstate newstatus newcomment)
       (sqlite3:execute db "UPDATE tests SET state=?,status=?,comment=? WHERE id=?;" newstate newstatus newcomment ;; (sdb:qry 'getid newcomment)
			test-id))
      ((and newstate newstatus)
       (sqlite3:execute db "UPDATE tests SET state=?,status=? WHERE id=?;" newstate newstatus test-id))
      (else
       (if newstate   (sqlite3:execute db "UPDATE tests SET state=?   WHERE id=?;" newstate   test-id))
       (if newstatus  (sqlite3:execute db "UPDATE tests SET status=?  WHERE id=?;" newstatus  test-id))
       (if newcomment (sqlite3:execute db "UPDATE tests SET comment=? WHERE id=?;" newcomment ;; (sdb:qry 'getid newcomment)
				       test-id))))))
  (mt:process-triggers dbstruct run-id test-id newstate newstatus))

(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")))
         (server-start-allowance 200)
         (server-overloaded-budget 200)
         (launch-monitor-off-time (or (configf:lookup-number *configdat* "setup" "test-stats-update-period") 30))
         (launch-monitor-on-time-budget 30)
         (launch-monitor-period (+ launch-monitor-off-time launch-monitor-on-time-budget server-overloaded-budget))
         (remotehoststart-deadtime-default (+ server-start-allowance server-overloaded-budget 30))
         (remotehoststart-deadtime (or deadtime-trim remotehoststart-deadtime-default))
         (running-deadtime-default (+ server-start-allowance (* 2 launch-monitor-period)))
         (running-deadtime (or deadtime-trim running-deadtime-default)) ;; two minutes (30 seconds between updates, this leaves 3x grace period)
         )
    (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)

		 ;; MOVE TO rmt:find-and-mark-incomplete - for now always call launch:end-of-run-check after
		 ;; calling rmt:find-and-mark-incompletes
		 
		 ;;ALWAYS CALL after rmt:find-and-mark-incompletes
		 ;; call end of eud of run detection for posthook
		 ;; (launch:end-of-run-check run-id)

		 )))))))

(define (db:test-set-state-status-process-triggers  dbstruct run-id test-id newstate newstatus newcomment)
  (db:test-set-state-status dbstruct run-id test-id newstate newstatus newcomment)
  (mt:process-triggers dbstruct run-id test-id newstate newstatus))

;; get a list of servers with all relevant data
;; ( mod-time host port start-time pid )
;;
(define (server:get-list areapath #!key (limit #f))
  (let ((fname-rx    (regexp "^(|.*/)server-(\\d+)-(\\S+).log$"))
	(day-seconds (* 24 60 60)))
    ;; if the directory exists continue to get the list
    ;; otherwise attempt to create the logs dir and then
    ;; continue
    (if (if (directory-exists? (conc areapath "/logs"))
	    '()
	    (if (file-write-access? areapath)
		(begin
		  (condition-case
		   (create-directory (conc areapath "/logs") #t)
		   (exn (i/o file)(debug:print 0 *default-log-port* "ERROR: Cannot create directory at " (conc areapath "/logs")))
		   (exn ()(debug:print 0 *default-log-port* "ERROR: Unknown error attemtping to get server list. exn=" exn)))
		  (directory-exists? (conc areapath "/logs")))
		'()))
	(let* ((server-logs   (glob (conc areapath "/logs/server-[0-9]*.log")))
	       (num-serv-logs (length server-logs)))
	  (if (null? server-logs)
	      '()
	      (let loop ((hed  (car server-logs))
			 (tal  (cdr server-logs))
			 (res '()))
		(let* ((mod-time  (handle-exceptions
				   exn
				   (begin
				     (debug:print 0 *default-log-port* "failed to get modification time on " hed ", exn=" exn)
				     (current-seconds)) ;; 0
				   (file-modification-time hed))) ;; default to *very* old so log gets ignored if deleted
		       (down-time (- (current-seconds) mod-time))
		       (serv-dat  (if (or (< num-serv-logs 10)
				  	  (< down-time 900)) ;; day-seconds))
				      (server:logf-get-start-info hed)
				      '())) ;; don't waste time processing server files not touched in the 15 minutes if there are more than ten servers to look at
		       (serv-rec (cons mod-time serv-dat))
		       (fmatch   (string-match fname-rx hed))
		       (pid      (if fmatch (string->number (list-ref fmatch 2)) #f))
		       (new-res  (if (null? serv-dat)
				     res
				     (cons (append serv-rec (list pid)) res)))) ;; any changes to number of elements in new-res will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 
		  (if (null? tal)
		      (if (and limit
			       (> (length new-res) limit))
			  new-res ;; (take new-res limit)  <= need intelligent sorting before this will work
			  new-res)
		      (loop (car tal)(cdr tal) new-res)))))))))

;; options:
;;
;;  'killservers  - kills all servers
;;  'dejunk       - removes junk records
;;  'adj-testids  - move test-ids into correct ranges
;;  'old2new      - sync megatest.db to /tmp/.../megatest.db and /tmp/.../megatest_ref.db
;;  'new2old      - sync /tmp/.../megatest.db to megatest.db and /tmp/.../megatest_ref.db (and update data_synced)
;;  'closeall     - close all opened dbs
;;  'schema       - attempt to apply schema changes
;;  run-ids: '(1 2 3 ...) or #f (for all)
;;
(define (db:multi-db-sync dbstruct . options)
  ;; (if (not (launch:setup))
  ;;    (debug:print 0 *default-log-port* "ERROR: not able to setup up for megatest.")
  (let* ((mtdb     (dbr:dbstruct-mtdb dbstruct))
	 (tmpdb    (db:get-db dbstruct))
	 (refndb   (dbr:dbstruct-refndb dbstruct))
	 (allow-cleanup #t) ;; (if run-ids #f #t))
	 (servers (server:get-list *toppath*)) ;; (tasks:get-all-servers (db:delay-if-busy tdbdat)))
	 (data-synced 0)) ;; count of changed records (I hope)
    
    (for-each
     (lambda (option)
       
       (case option
	 ;; kill servers
	 ((killservers)
	  (for-each
	   (lambda (server)
             (handle-exceptions
             exn
             (begin 
               (debug:print-info 0 *default-log-port*  "Unable to get host and/or port from " server ", exn=" exn)     
               #f)
	     (match-let (((mod-time host port start-time server-id pid) server))
	       (if (and host pid)
		   (tasks:kill-server host pid)))))
	   servers)

          ;; /tmp/bjbarcla/megatest_localdb/fullrun/.nfs.pdx.disks.icf_env_disk001.bjbarcla.gwa.issues.mtdev.matt-bisect.megatest.ext-tests.runs.all.v1.65.1.6524.dev.bb-24-justrollup-f8.rollup.fullrun/megatest.db.lock
          (delete-file* (common:get-sync-lock-filepath))
          )
	 
	 ;; clear out junk records
	 ;;
	 ((dejunk)
	  ;; (db:delay-if-busy mtdb) ;; ok to delay on mtdb
	  (when (file-write-access? (db:dbdat-get-path mtdb)) (db:clean-up mtdb))
	  (db:clean-up tmpdb)
	  (db:clean-up refndb))

	 ;; sync runs, test_meta etc.
	 ;;
	 ((old2new)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb tmpdb refndb)
	       data-synced)))
	 
	 ;; now ensure all newdb data are synced to megatest.db
	 ;; do not use the run-ids list passed in to the function
	 ;;
	 ((new2old)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f tmpdb refndb mtdb)
	       data-synced)))

	 ((adj-target)
	  (db:adj-target (db:dbdat-get-db mtdb))
	  (db:adj-target (db:dbdat-get-db tmpdb))
	  (db:adj-target (db:dbdat-get-db refndb)))
	 
	 ((schema)
	  (db:patch-schema-maindb (db:dbdat-get-db mtdb))
	  (db:patch-schema-maindb (db:dbdat-get-db tmpdb))
	  (db:patch-schema-maindb (db:dbdat-get-db refndb))
	  (db:patch-schema-rundb  (db:dbdat-get-db mtdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db tmpdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db refndb))))
       
       (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb))
     options)
    data-synced))

(define (db:get-access-mode)
  (if (args:get-arg "-use-db-cache") 'cached 'rmt))

;; given a path to a server log return: host port startseconds
;; any changes to number of elements returned by this fuction will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 
;;
(define (server:logf-get-start-info logf)
  (let ((rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+)"))) ;; SERVER STARTED: host:port AT timesecs server id
    ;;(handle-exceptions
    ;; exn
    ;; (begin
    ;;   (debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", exn=" exn)
    ;;   (list #f #f #f #f)) ;; no idea what went wrong, call it a bad server
    (if (and (file-exists? logf)
	     (file-read-access? logf))
	(with-input-from-file
	    logf
	  (lambda ()
	    (let loop ((inl  (read-line))
		       (lnum 0))
	      (if (not (eof-object? inl))
		  (let ((mlst (string-match rx inl)))
		    (if (not mlst)
			(if (< lnum 500) ;; give up if more than 500 lines of server log read
			    (loop (read-line)(+ lnum 1))
			    (begin 
			      (debug:print-info 0 *default-log-port* "Unable to get server info from first 500 lines of " logf )
			      (list #f #f #f #f)))
			(let ((dat  (cdr mlst)))
			  (list (car dat) ;; host
				(string->number (cadr dat)) ;; port
				(string->number (caddr dat))
				(cadr (cddr dat))))))
		  (begin 
		    (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (current-seconds))
		    (list #f #f #f #f))))))
	(begin
	  (debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", file not found or not readable.")
	  (list #f #f #f #f)))))

;; cannot move to dbmod until lazy-read-testconfig is unravelled.
;;
(define (mt:process-triggers dbstruct run-id test-id newstate newstatus)
  (if test-id 
      (let* ((test-dat      (db:get-test-info-by-id dbstruct run-id test-id)))
	(if test-dat
	    (let* ((test-rundir   (db:test-get-rundir       test-dat)) ;; ) ;; )
		   (test-name     (db:test-get-testname     test-dat))
		   (item-path     (db:test-get-item-path    test-dat))
		   (duration      (db:test-get-run_duration test-dat))
		   (comment       (db:test-get-comment      test-dat))
		   (event-time    (db:test-get-event_time   test-dat))
		   (tconfig       #f)
		   (state         (if newstate  newstate  (db:test-get-state  test-dat)))
		   (status        (if newstatus newstatus (db:test-get-status test-dat))))
	      ;; (mutex-lock! *triggers-mutex*)
              (handle-exceptions
               exn
               (begin
                 (debug:print-error 0 *default-log-port* " Exception in mt:process-triggers for run-id="run-id" test-id="test-id" newstate="newstate" newstatus="newstatus
                                    "\n   error: " ((condition-property-accessor 'exn 'message) exn) ", exn=" exn
                                    "\n   test-rundir="test-rundir
                                    "\n   test-name="test-name
                                    "\n   item-path="item-path
                                    "\n   state="state
                                    "\n   status="status
                                    "\n")
                 (print-call-chain (current-error-port))
                 #f)
               (if (and test-name
                        test-rundir)   ;; #f means no dir set yet
                   ;; (common:file-exists? test-rundir)
                   ;; (directory? test-rundir))
                   (call-with-environment-variables
                    (list (cons "MT_TEST_NAME"    (or test-name "no such test"))
                          (cons "MT_TEST_RUN_DIR" (or test-rundir "no test directory yet"))
                          (cons "MT_ITEMPATH"     (or item-path "")))
                    (lambda ()
                      (if (directory-exists? test-rundir)
                          (push-directory test-rundir)
                          (push-directory *toppath*))
                      (set! tconfig (mt:lazy-read-test-config test-name))
                      (for-each (lambda (trigger)
                                  (let* ((munged-trigger (string-translate trigger "/ " "--"))
					(logname        (conc "last-trigger-" munged-trigger ".log")))
                                    ;; first any triggers from the testconfig
                                    (let ((cmd  (configf:lookup tconfig "triggers" trigger)))
                                      (if cmd (mt:run-trigger cmd test-id test-rundir trigger (conc "tconfig-" logname) test-name item-path event-time state status)))
                                    ;; next any triggers from megatest.config
                                    (let ((cmd  (configf:lookup *configdat* "triggers" trigger)))
                                      (if cmd (mt:run-trigger cmd test-id test-rundir trigger (conc "mtconfig-" logname) test-name item-path event-time state status)))))
                                (list
                                 (conc state "/" status)
                                 (conc state "/")
                                 (conc "/" status)))
		     (pop-directory))
                    )))
	      ;; (mutex-unlock! *triggers-mutex*)
	      )))))

(define (mt:lazy-read-test-config test-name)
  (let ((tconf (hash-table-ref/default *testconfigs* test-name #f)))
    (if tconf
	tconf
	(let ((test-dirs (tests:get-tests-search-path *configdat*)))
	  (let loop ((hed (car test-dirs))
		     (tal (cdr test-dirs)))
	    ;; Setting MT_LINKTREE here is almost certainly unnecessary. 
	    (let ((tconfig-file (conc hed "/" test-name "/testconfig")))
	      (if (and (common:file-exists? tconfig-file)
		       (file-read-access? tconfig-file))
		  (let ((link-tree-path (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
			(old-link-tree  (get-environment-variable "MT_LINKTREE")))
		    (if link-tree-path (setenv "MT_LINKTREE" link-tree-path))
		    (let ((newtcfg (read-config tconfig-file #f #f))) ;; NOTE: Does NOT run [system ...]
		      (hash-table-set! *testconfigs* test-name newtcfg)
		      (if old-link-tree 
			  (setenv "MT_LINKTREE" old-link-tree)
			  (unsetenv "MT_LINKTREE"))
		      newtcfg))
		  (if (null? tal)
		      (begin
			(debug:print-error 0 *default-log-port* "No readable testconfig found for " test-name)
			#f)
		      (loop (car tal)(cdr tal))))))))))


;; MUST RESOLVE mt:process-triggers before these can move to dbmod.

;; set tests with state currstate and status currstatus to newstate and newstatus
;; use currstate = #f and or currstatus = #f to apply to any state or status respectively
;; WARNING: SQL injection risk. NB// See new but not yet used "faster" version below
;;
;;  AND NOT (item_path='' AND testname in (SELECT DISTINCT testname FROM tests WHERE testname=? AND item_path != ''));")))
;;  (debug:print 0 *default-log-port* "QRY: " qry)
;;  (db:delay-if-busy)
;;
;; NB// This call only operates on toplevel tests. Consider replacing it with more general call
;;
(define (db:set-tests-state-status dbstruct run-id testnames currstate currstatus newstate newstatus)
  (let ((test-ids '()))
    (for-each
     (lambda (testname)
       (let ((qry (conc "UPDATE tests SET state=?,status=? WHERE "
			(if currstate  (conc "state='" currstate "' AND ") "")
			(if currstatus (conc "status='" currstatus "' AND ") "")
			" run_id=? AND testname LIKE ?;"))
	     (test-id (db:get-test-id dbstruct run-id testname "")))
	 (db:with-db
	  dbstruct
	  run-id
	  #t
	  (lambda (db)
	    (sqlite3:execute db qry
			     (or newstate  currstate "NOT_STARTED")
			     (or newstatus currstate "UNKNOWN")
			     run-id testname)))
	 (if test-id
	     (begin
	       (set! test-ids (cons test-id test-ids))
	       (mt:process-triggers dbstruct run-id test-id newstate newstatus)))))
     testnames)
    test-ids))


;;======================================================================
;; 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 (launch:setup)
      (if (common:on-homehost?)
	  (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.")
              (let* ((syncer (or (configf:lookup *configdat* "server" "sync-method") "brute-force-sync")))
                (cond
                 ((equal? syncer "brute-force-sync")
                  (server:writable-watchdog-bruteforce dbstruct))
                 ((equal? syncer "delta-sync")
                  (server:writable-watchdog-deltasync dbstruct))
                 (else
                  (debug:print-error 0 *default-log-port* "Unknown server/sync-method specified ("syncer") - valid values are brute-force-sync and delta-sync.")
                  (exit 1)))
                ;;(debug:print 1 *default-log-port* "INFO: ["(common:human-time)"] Syncer started (method="syncer")")
                )))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))

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

;;======================================================================
;; 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)))
	    (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))))

;;======================================================================
;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)
;;
(define (common:readonly-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (debug:print-info 13 *default-log-port* "common:readonly-watchdog entered.")
  ;; sync megatest.db to /tmp/.../megatst.db
  (let* ((sync-cool-off-duration   3)
        (golden-mtdb     (dbr:dbstruct-mtdb dbstruct))
        (golden-mtpath   (db:dbdat-get-path golden-mtdb))
        (tmp-mtdb        (dbr:dbstruct-tmpdb dbstruct))
        (tmp-mtpath      (db:dbdat-get-path tmp-mtdb)))
    (debug:print-info 0 *default-log-port* "Read-only periodic sync thread started.")
    (let loop ((last-sync-time 0))
      (debug:print-info 13 *default-log-port* "loop top tmp-mtpath="tmp-mtpath" golden-mtpath="golden-mtpath)
      (let* ((duration-since-last-sync (- (current-seconds) last-sync-time)))
        (debug:print-info 13 *default-log-port* "duration-since-last-sync="duration-since-last-sync)
        (if (and (not *time-to-exit*)
                 (< duration-since-last-sync sync-cool-off-duration))
            (thread-sleep! (- sync-cool-off-duration duration-since-last-sync)))
        (if (not *time-to-exit*)
            (let ((golden-mtdb-mtime (file-modification-time golden-mtpath))
                  (tmp-mtdb-mtime    (file-modification-time tmp-mtpath)))
	      (if (> golden-mtdb-mtime tmp-mtdb-mtime)
		  (if (< golden-mtdb-mtime (- (current-seconds) 3)) ;; file has NOT been touched in past three seconds, this way multiple servers won't fight to sync back
		      (let ((res (db:multi-db-sync dbstruct 'old2new)))
			(debug:print-info 13 *default-log-port* "rosync called, " res " records transferred."))))
              (loop (current-seconds)))
            #t)))
    (debug:print-info 0 *default-log-port* "Exiting readonly-watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id)" mtpath="golden-mtpath)))

;;======================================================================
;; first read ~/views.config if it exists, then read $MTRAH/views.config if it exists
;;
(define (common:load-views-config)
  (let* ((view-cfgdat    (make-hash-table))
	 (home-cfgfile   (conc (get-environment-variable "HOME") "/.mtviews.config"))
	 (mthome-cfgfile (conc *toppath* "/.mtviews.config")))
    (if (common:file-exists? mthome-cfgfile)
	(read-config mthome-cfgfile view-cfgdat #t))
    ;; we load the home dir file AFTER the MTRAH file so the user can clobber settings when running the dashboard in read-only areas
    (if (common:file-exists? home-cfgfile)
	(read-config home-cfgfile view-cfgdat #t))
    view-cfgdat))

;;======================================================================
;; T A R G E T S  ,   S T A T E ,   S T A T U S ,   
;;                    R U N N A M E    A N D   T E S T P A T T
;;======================================================================

;;======================================================================
;; (map print (map car (hash-table->alist (read-config "runconfigs.config" #f #t))))
;;
(define (common:get-runconfig-targets #!key (configf #f))
  (let ((targs       (sort (map car (hash-table->alist
				     (or configf ;; NOTE: There is no value in using runconfig:read here.
					 (read-config (conc *toppath* "/runconfigs.config")
						      #f #t)
					 (make-hash-table))))
			   string<?))
	(target-patt (args:get-arg "-target")))
    (if target-patt
	(filter (lambda (x)
		  (patt-list-match x target-patt))
		targs)
	targs)))

;;======================================================================
;; Lookup a value in runconfigs based on -reqtarg or -target
;; 
(define (runconfigs-get config var)
  (let ((targ (common:args-get-target))) ;; (or (args:get-arg "-reqtarg")(args:get-arg "-target")(getenv "MT_TARGET"))))
    (if targ
	(or (configf:lookup config targ var)
	    (configf:lookup config "default" var))
	(configf:lookup config "default" var))))

(define (common:args-get-state)
  (or (args:get-arg "-state")(args:get-arg ":state")))

(define (common:args-get-status)
  (or (args:get-arg "-status")(args:get-arg ":status")))

(define (common:args-get-testpatt rconf)
  (let* (;; (tagexpr       (args:get-arg "-tagexpr"))
         ;; (tags-testpatt (if tagexpr (string-join (runs:get-tests-matching-tags tagexpr) ",") #f))
         (testpatt-key  (or (args:get-arg "-modepatt") (args:get-arg "--modepatt") "TESTPATT"))
         (args-testpatt (or (args:get-arg "-testpatt") (args:get-arg "-runtests") "%"))
         (rtestpatt     (if rconf (runconfigs-get rconf testpatt-key) #f)))
    (cond
     ((or (args:get-arg "--modepatt") (args:get-arg "-modepatt")) ;; modepatt is a forced setting, when set it MUST refer to an existing PATT in the runconfig
      (if rconf
	  (let* ((patts-from-mode-patt	  (runconfigs-get rconf testpatt-key)))
	    (debug:print-info 0 *default-log-port* "modepatt defined is: "testpatt-key" runconfigs values for  " testpatt-key " "  patts-from-mode-patt)
	    patts-from-mode-patt)
	  (begin
	    (debug:print-info 0 *default-log-port* " modepatt defined is: "testpatt-key" runconfigs values for  " testpatt-key) ;;  " " patts-from-mode-patt)
	    #f)))     ;; We do NOT fall back to "%"
     ;; (tags-testpatt
     ;;  (debug:print-info 0 *default-log-port* "-tagexpr "tagexpr" selects testpatt "tags-testpatt)
     ;;  tags-testpatt)
     ((and (equal? args-testpatt "%") rtestpatt)
      (debug:print-info 0 *default-log-port* "testpatt defined in "testpatt-key" from runconfigs: " rtestpatt)
      rtestpatt)
     (else 
      (debug:print-info 0 *default-log-port* "using testpatt " args-testpatt " rtestpatt:" rtestpatt)
      args-testpatt))))

;;======================================================================
;; '(print (string-intersperse (map cadr (hash-table-ref/default (read-config "megatest.config" \#f \#t) "disks" '"'"'("none" ""))) "\n"))'
(define (common:get-disks #!key (configf #f))
  (hash-table-ref/default 
   (or configf (read-config "megatest.config" #f #t))
   "disks" '("none" "")))

;;======================================================================
;; postive number if megatest version > db version
;; negative number if megatest version < db version
(define (common:version-db-delta)
         (- megatest-version (common:get-last-run-version-number)))

(define (common:version-changed?)
  (not (equal? (common:get-last-run-version)
               (common:version-signature))))

(define (common:api-changed?)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (common:get-last-run-version)) 0 4))))

(define (server:writable-watchdog-bruteforce dbstruct)
  (thread-sleep! 1) ;; delay for startup
  (let* ((do-a-sync  (server:get-bruteforce-syncer dbstruct))
         (final-sync (server:get-bruteforce-syncer dbstruct fork-to-background: #t persist-until-sync: #t)))
    (when (and (not (args:get-arg "-sync-to-megatest.db")) ;; conditions under which we do not run the sync
	       (args:get-arg "-server"))
      
      (let loop ()
	(do-a-sync)
        (if (not *time-to-exit*) (loop))) ;; keep going unless time to exit

      ;; time to exit, close the no-sync db here
      (final-sync)

      (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)
			    )))))

(define (server:writable-watchdog-deltasync dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (let ((legacy-sync  (common:run-sync?))
        (sync-stale-seconds (configf:lookup-number *configdat* "server" "sync-stale-seconds" default: 300))
	(debug-mode   (debug:debug-mode 1))
	(last-time    (current-seconds))
	(no-sync-db   (db:open-no-sync-db))
	(stmt-cache   (dbr:dbstruct-stmt-cache dbstruct))
        (sync-duration 0) ;; run time of the sync in milliseconds
        )
    (set! *no-sync-db* no-sync-db) ;; make the no sync db available to api calls
    (debug:print-info 2 *default-log-port* "Periodic sync thread started.")
    (debug:print-info 3 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id)  );;  " this-wd-num="this-wd-num)
    (if (and legacy-sync (not *time-to-exit*))
	(let* (;;(dbstruct (db:setup))
	       (mtdb       (dbr:dbstruct-mtdb dbstruct))
	       (mtpath     (db:dbdat-get-path mtdb))
	       (tmp-area   (common:get-db-tmp-area))
	       (start-file (conc tmp-area "/.start-sync"))
	       (end-file   (conc tmp-area "/.end-sync")))
	  (debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
	  (let loop ()
	    ;; sync for filesystem local db writes
	    ;;
	    (mutex-lock! *db-multi-sync-mutex*)
	    (let* ((need-sync        (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
		   (sync-in-progress *db-sync-in-progress*)
                   (min-intersync-delay (configf:lookup-number *configdat* "server" "minimum-intersync-delay" default: 5))
		   (should-sync      (and (not *time-to-exit*)
                                          (> (- (current-seconds) *db-last-sync*) min-intersync-delay))) ;; sync every five seconds minimum, deprecated logic, can probably be removed
		   (start-time       (current-seconds))
                   (cpu-load-adj     (alist-ref 'adj-proc-load (common:get-normalized-cpu-load #f)))
		   (mt-mod-time      (file-modification-time mtpath))
		   (last-sync-start  (if (common:file-exists? start-file)
					 (file-modification-time start-file)
					 0))
		   (last-sync-end    (if (common:file-exists? end-file)
					 (file-modification-time end-file)
					 10))
                   (sync-period      (+ 3 (* cpu-load-adj 30))) ;; as adjusted load increases increase the sync period
		   (recently-synced  (and (< (- start-time mt-mod-time) sync-period) ;; not useful if sync didn't modify megatest.db!
					  (< mt-mod-time last-sync-start)))
		   (sync-done        (<= last-sync-start last-sync-end))
                   (sync-stale       (> start-time (+ last-sync-start sync-stale-seconds)))
		   (will-sync        (and (not *time-to-exit*)       ;; do not start a sync if we are in the process of exiting
                                          (or need-sync should-sync)
					  (or sync-done sync-stale)
					  (not sync-in-progress)
					  (not recently-synced))))
              (debug:print-info 13 *default-log-port* "WD writable-watchdog top of loop.  need-sync="need-sync" sync-in-progress=" sync-in-progress
				" should-sync="should-sync" start-time="start-time" mt-mod-time="mt-mod-time" recently-synced="recently-synced" will-sync="will-sync
				" sync-done=" sync-done " sync-period=" sync-period)
              (if (and (> sync-period 5)
                       (common:low-noise-print 30 "sync-period"))
                  (debug:print-info 0 *default-log-port* "Increased sync period due to long sync times, sync took: " sync-period " seconds."))
	      ;; (if recently-synced (debug:print-info 0 *default-log-port* "Skipping sync due to recently-synced flag=" recently-synced))
	      ;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
	      (if will-sync (set! *db-sync-in-progress* #t))
	      (mutex-unlock! *db-multi-sync-mutex*)
	      (if will-sync
                  (let (;; (max-sync-duration  (configf:lookup-number *configdat* "server" "max-sync-duration")) ;; KEEPING THIS AVAILABLE BUT SHOULD NOT USE, I'M PRETTY SURE IT DOES NOT WORK!
                        (sync-start         (current-milliseconds)))
		    (with-output-to-file start-file (lambda ()(print (current-process-id))))
		    
		    ;; put lock here
		    
                    ;; (if (or (not max-sync-duration)
                    ;;        (< sync-duration max-sync-duration)) ;; NOTE: db:sync-to-megatest.db keeps track of time of last sync and syncs incrementally
                        (let ((res        (db:sync-to-megatest.db dbstruct no-sync-db: no-sync-db))) ;; did we sync any data? If so need to set the db touched flag to keep the server alive
                          (set! sync-duration (- (current-milliseconds) sync-start))
                          (if (> res 0) ;; some records were transferred, keep the db alive
                              (begin
                                (mutex-lock! *heartbeat-mutex*)
                                (set! *db-last-access* (current-seconds))
                                (mutex-unlock! *heartbeat-mutex*)
                                (debug:print-info 0 *default-log-port* "sync called, " res " records transferred."))
                              (debug:print-info 2 *default-log-port* "sync called but zero records transferred")))))
;;                         ;; TODO: factor this next routine out into a function
;;                         (with-input-from-pipe ;; this should not block other threads but need to verify this
;;                          (conc "megatest -sync-to-megatest.db -m testsuite:" (common:get-area-name) ":" *toppath*)
;;                          (lambda ()
;;                            (let loop ((inl (read-line))
;;                                       (res #f))
;;                              (if (eof-object? inl)
;;                                  (begin
;;                                    (set! sync-duration (- (current-milliseconds) sync-start))
;;                                    (cond
;;                                     ((not res)
;;                                      (debug:print 0 *default-log-port* "ERROR: sync from /tmp db to megatest.db appears to have failed. Recommended that you stop your runs and run \"megatest -cleanup-db\""))
;;                                     ((> res 0)
;;                                      (mutex-lock! *heartbeat-mutex*)
;;                                      (set! *db-last-access* (current-seconds))
;;                                      (mutex-unlock! *heartbeat-mutex*))))
;;                                  (let ((num-synced (let ((matches (string-match "^Synced (\\d+).*$" inl)))
;;                                                      (if matches
;;                                                          (string->number (cadr matches))
;;                                                          #f))))
;;                                    (loop (read-line)
;;                                          (or num-synced res))))))))))
	      (if will-sync
		  (begin
		    (mutex-lock! *db-multi-sync-mutex*)
		    (set! *db-sync-in-progress* #f)
		    (set! *db-last-sync* start-time)
		    (with-output-to-file end-file (lambda ()(print (current-process-id))))

		    ;; release lock here

		    (mutex-unlock! *db-multi-sync-mutex*)))
	      (if (and debug-mode
		       (> (- start-time last-time) 60))
		  (begin
		    (set! last-time start-time)
		    (debug:print-info 4 *default-log-port* "timestamp -> " (seconds->time-string (current-seconds)) ", time since start -> " (seconds->hr-min-sec (- (current-seconds) *time-zero*))))))
	    
	    ;; keep going unless time to exit
	    ;;
	    (if (not *time-to-exit*)
		(let delay-loop ((count 0))
                  ;;(debug:print-info 13 *default-log-port* "delay-loop top; count="count" pid="(current-process-id)" this-wd-num="this-wd-num" *time-to-exit*="*time-to-exit*)
                                                            
		  (if (and (not *time-to-exit*)
			   (< count 6)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (if (not *time-to-exit*) (loop))))
	    ;; time to exit, close the no-sync db here
	    (db:no-sync-close-db no-sync-db stmt-cache)
	    (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)))))))


)