Megatest

;;======================================================================
;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.
;;======================================================================

(use srfi-1 data-structures posix regex-case (prefix base64 base64:)
     format dot-locking csv-xml z3 ;; sql-de-lite
(use srfi-1 posix regex-case base64 format dot-locking csv-xml z3 sql-de-lite hostinfo md5 message-digest typed-records directory-utils stack
     matchable)
     hostinfo md5 message-digest typed-records directory-utils stack
     matchable regex posix (srfi 18) extras ;; tcp 
(require-extension regex posix)

     (prefix nanomsg nmsg:)
(require-extension (srfi 18) extras tcp rpc)

     (prefix sqlite3 sqlite3:)
     pkts
     )
(import (prefix sqlite3 sqlite3:))
(import (prefix base64 base64:))

(declare (unit common))
(declare (uses keys))

(include "common_records.scm")


;; (require-library margs)
;; (include "margs.scm")

;; (define old-exit exit)
;; 
;; (define (exit . code)
;;   (if (null? code)
;;       (old-exit)
;;       (old-exit code)))


;; execute thunk, return value.  If exception thrown, trap exception, return #f, and emit nonfatal condition note to *default-log-port* .
;; arguments - thunk, message
(define (common:fail-safe thunk warning-message-on-exception)
  (handle-exceptions
   exn
   (begin
     (debug:print-info 0 *default-log-port* "notable but nonfatal condition - "warning-message-on-exception)
     (debug:print-info 0 *default-log-port*
                       (string-substitute "\n?Error:" "nonfatal condition:"
                                          (with-output-to-string
                                            (lambda ()
                                              (print-error-message exn) ))))
     (debug:print-info 0 *default-log-port* "    -- continuing after nonfatal condition...")
     #f)
   (thunk)))

(define getenv get-environment-variable)
(define (safe-setenv key val)
  (if (substring-index ":" key) ;; variables containing : are for internal use and cannot be environment variables.
      (debug:print-error 4 *default-log-port* "skip setting internal use only variables containing \":\"")
      (if (and (string? val)
	       (string? key))
	  (handle-exceptions
	      exn
	      (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val)
	    (setenv key val))
	  (debug:print-error 0 *default-log-port* "bad value for setenv, key=" key ", value=" val))))

(define home (getenv "HOME"))
(define user (getenv "USER"))


;; returns list of fd count, socket count
(define (get-file-descriptor-count #!key  (pid (current-process-id )))
  (list
    (length (glob (conc "/proc/" pid "/fd/*")))
    (length  (filter identity (map socket? (glob (conc "/proc/" pid "/fd/*")))))
  )
)


;; GLOBAL GLETCHES
;; GLOBALS

;; CONTEXTS
(defstruct cxt
  (taskdb #f)
  (cmutex (make-mutex)))
;; (define *contexts* (make-hash-table))
;; (define *context-mutex* (make-mutex))

;; A hash table that can be accessed by #{scheme ...} calls in
;; config files. Allows communicating between confgs
;;
(define *user-hash-data* (make-hash-table))

(define *db-keys* #f)

(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 *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-db*         #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 *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))
(define *db-transaction-mutex* (make-mutex))
(define *db-cache-path*       #f)
(define *db-with-db-mutex*    (make-mutex))
(define *db-api-call-time*    (make-hash-table)) ;; hash of command => (list of times)
;; no sync db
(define *no-sync-db*          #f)

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*    'http)             ;; override with [server] transport http|rpc|nmsg
(define *runremote*         #f)                ;; if set up for server communication this will hold <host port>
;; (define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
;; (define *server-id*         #f)
(define *server-id*         #f)
(define *server-info*       #f)  ;; good candidate for easily convert to non-global
(define *time-to-exit*      #f)
(define *server-run*        #t)
(define *run-id*            #f)
(define *server-kind-run*   (make-hash-table))
(define *home-host*         #f)
;; (define *total-non-write-delay* 0)
(define *heartbeat-mutex*   (make-mutex))
(define *api-process-request-count* 0)
(define *max-api-process-requests* 0)
(define *server-overloaded*  #f)

;; client
(define *rmt-mutex*         (make-mutex))     ;; remote access calls mutex 

;; RPC transport
(define *rpc:listener*      #f)

;; KEY info
(define *target*            (make-hash-table)) ;; cache the target here; target is keyval1/keyval2/.../keyvalN
(define *keys*              (make-hash-table)) ;; cache the keys here
(define *keyvals*           (make-hash-table))
(define *toptest-paths*     (make-hash-table)) ;; cache toptest path settings here
(define *test-paths*        (make-hash-table)) ;; cache test-id to test run paths here
(define *test-ids*          (make-hash-table)) ;; cache run-id, testname, and item-path => test-id
(define *test-info*         (make-hash-table)) ;; cache the test info records, update the state, status, run_duration etc. from testdat.db

(define *run-info-cache*     (make-hash-table)) ;; run info is stable, no need to reget
(define *launch-setup-mutex* (make-mutex))     ;; need to be able to call launch:setup often so mutex it and re-call the real deal only if *toppath* not set
(define *homehost-mutex*     (make-mutex))

;; Miscellaneous
(define *triggers-mutex*     (make-mutex))     ;; block overlapping processing of triggers

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        (if *toppath* (server:check-if-running *toppath*))) ;; (server:check-if-running *toppath*) #f))
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (conndat           #f)
  (transport         *transport-type*)
  (server-timeout    (server:get-timeout)) ;; default from server:get-timeout
  (server-timeout    (server:expiration-timeout))
  (force-server      #f)
  (ro-mode           #f)  
  (ro-mode-checked   #f)) ;; flag that indicates we have checked for ro-mode

;; launching and hosts
(defstruct host
  (reachable    #f)

(define (common:get-last-run-version-number)
  (string->number 
   (substring (common:get-last-run-version) 0 6)))

(define (common:set-last-run-version)
  (rmt:set-var "MEGATEST_VERSION" (common:version-signature)))

;; 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))))
               (common:version-signature))))

(define (common:api-changed?)
  (not (equal? (substring (->string megatest-version) 0 4)
               (substring (conc (common:get-last-run-version)) 0 4))))
  
;; 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 
(define (common:cleanup-db dbstruct #!key (full #f))
  (apply db:multi-db-sync 
   dbstruct
   'schema
   ;; 'new2old
   'killservers
   'dejunk
   ;; 'adj-testids
   'adj-target
   ;; 'old2new
   'new2old
   ;; (if full
       '(dejunk)
   'schema)
  (if (common:version-changed?)
       ;; '())
       )
  (if (common:api-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
;;          logs directory you wish to log-rotate.

             (file-age (- (current-seconds)(file-modification-time fullname))))
        (if (or (and (string-match "^.*.log" file)
                     (> (file-size fullname) 200000))
                (and (string-match "^server-.*.log" file)
                     (> (- (current-seconds) (file-modification-time fullname))
                        (* 8 60 60))))
            (let ((gzfile (conc fullname ".gz")))
              (if (file-exists? gzfile)
              (if (common:file-exists? gzfile)
                  (begin
                    (debug:print-info 0 *default-log-port* "removing " gzfile)
                    (delete-file gzfile)))
              (debug:print-info 0 *default-log-port* "compressing " file)
              (system (conc "gzip " fullname)))
            (if (> file-age (* (string->number (or (configf:lookup *configdat* "setup" "log-expire-days") "30")) 24 3600))
                (handle-exceptions
                 exn
                 #f
                 (delete-file fullname)))))))
   '()
   "logs"))

;; 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:version-changed?)
      (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)))
                (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)
             ((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.")
                 (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 (file-exists? mtconf))
             ((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 (file-exists? dbfile))
             ((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))))
              (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
;;======================================================================

(define (make-sparse-array)
  (let ((a (make-sparse-vector)))

;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (handle-exceptions
      exn
      #f ;; don't really care what went wrong right now. NOTE: I have not seen this one actually fail.
    (if (file-exists? fname)
    (if (common:file-exists? fname)
	(if (> (- (current-seconds)(file-modification-time fname)) expire-time)
	    (begin
	      (delete-file* fname)
	      (common:simple-file-lock fname expire-time: expire-time))
	    #f)
	(let ((key-string (conc (get-host-name) "-" (current-process-id))))
	  (with-output-to-file fname
	    (lambda ()
	      (print key-string)))
	  (thread-sleep! 0.25)
	  (if (file-exists? fname)
	  (if (common:file-exists? fname)
	      (with-input-from-file fname
		(lambda ()
		  (equal? key-string (read-line))))
	      #f)))))

(define (common:simple-file-lock-and-wait fname #!key (expire-time 300))
  (let ((end-time (+ expire-time (current-seconds))))

      #f ;; I don't really care why this failed (at least for now)
    (delete-file* fname)))

;;======================================================================
;; S T A T E S   A N D   S T A T U S E S
;;======================================================================

;; BBnote: *common:std-states* - dashboard filter control and test control state buttons defined here; used in set-fields-panel and dboard:make-controls
(define *common:std-states*   
  '((0 "ARCHIVED")
(define *common:std-states*   ;; for toggle buttons in dashboard
  '(
    (0 "ARCHIVED")
    (1 "STUCK")
    (2 "KILLREQ")
    (3 "KILLED")
    (4 "NOT_STARTED")
    (5 "COMPLETED")
    (6 "LAUNCHED")
    (7 "REMOTEHOSTSTART")
    (8 "RUNNING")
    ))

(define *common:dont-roll-up-states*
  '("DELETED"
    "REMOVING"
    "CLEANING"
    "ARCHIVE_REMOVING"
    ))

;; BBnote: *common:std-statuses* dashboard filter control and test control status buttons defined here; used in set-fields-panel and dboard:make-controls
;; note these statuses are sorted from better to worse.
;; This sort order is important to dcommon:status-compare3 and db:set-state-status-and-roll-up-items
(define *common:std-statuses*
  '(;; (0 "DELETED")
  '(;; (0 "DELETED")  
    (1 "n/a")
    (2 "PASS")
    (3 "CHECK")
    (4 "SKIP")
    (5 "WARN")
    (6 "WAIVED")
    (3 "SKIP")
    (4 "WARN")
    (5 "WAIVED")
    (6 "CHECK")
    (7 "STUCK/DEAD")
    (8 "DEAD")
    (8 "FAIL")
    (9 "ABORT")))
    (9 "FAIL")
    (10 "PREQ_FAIL")
    (11 "PREQ_DISCARDED")
    (12 "ABORT")))

(define *common:ended-states*       ;; states which indicate the test is stopped and will not proceed
  '("COMPLETED" "ARCHIVED" "KILLED" "KILLREQ" "STUCK" "INCOMPLETE"))
  '("COMPLETED" "ARCHIVED" "KILLED" "KILLREQ" "STUCK" "INCOMPLETE" ))

(define *common:badly-ended-states* ;; these roll up as CHECK, i.e. results need to be checked
  '("KILLED" "KILLREQ" "STUCK" "INCOMPLETE" "DEAD"))

(define *common:well-ended-states* ;; an item's prereq in this state allows item to proceed
  '("PASS" "WARN" "CHECK" "WAIVED" "SKIP"))

;; BBnote: *common:running-states* used from db:set-state-status-and-roll-up-items
(define *common:running-states*     ;; test is either running or can be run
  '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED"))
  '("RUNNING" "REMOTEHOSTSTART" "LAUNCHED" "STARTED"))

(define *common:cant-run-states*    ;; These are stopping conditions that prevent a test from being run
  '("COMPLETED" "KILLED" "UNKNOWN" "INCOMPLETE" "ARCHIVED"))

(define *common:not-started-ok-statuses* ;; if not one of these statuses when in not_started state treat as dead
  '("n/a" "na" "PASS" "FAIL" "WARN" "CHECK" "WAIVED" "DEAD" "SKIP"))

;; group tests into buckets corresponding to rollup
;;; Running, completed-pass,  completed-non-pass + worst status, not started.
;; filter out 
;(define (common:categorize-items-for-rollup in-tests)
;  (

(define (common:special-sort items order comp)
  (let ((items-order (map reverse order))
        (acomp       (or comp >)))
    (sort items
        (lambda (a b)
          (let ((a-num (cadr (or (assoc a items-order) '(0 0))))

      (getenv "MT_TESTSUITE_NAME")
      (if (string? *toppath* )
          (pathname-file *toppath*)
          #f))) ;; (pathname-file (current-directory)))))

(define common:get-area-name common:get-testsuite-name)

(define (common:get-db-tmp-area)
(define (common:get-db-tmp-area . junk)
  (if *db-cache-path*
      *db-cache-path*
      (if *toppath*
	  (let ((dbpath (create-directory (conc "/tmp/" (current-user-name)
						"/megatest_localdb/"
						(common:get-testsuite-name) "/"
						(string-translate *toppath* "/" ".")) #t)))
	    (set! *db-cache-path* dbpath)
	    dbpath)
      (if *toppath* ;; common:get-create-writeable-dir
	  (handle-exceptions
	      exn
	      (begin
		(debug:print-error 0 *default-log-port* "Couldn't create path to " dbdir)
		(exit 1))
	    (let ((dbpath (common:get-create-writeable-dir
			   (list (conc "/tmp/" (current-user-name)
				       "/megatest_localdb/"
				       (common:get-testsuite-name) "/"
				       (string-translate *toppath* "/" ".")))))) ;;  #t))))
	      (set! *db-cache-path* dbpath)
	      dbpath))
	  #f)))

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

(define (common:get-signature str)
  (message-digest-string (md5-primitive) str))

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

(define (common:run-sync?)
    (and (common:on-homehost?)
	 (args:get-arg "-server")))

;;   (let ((ohh (common:on-homehost?))
;; 	(srv (args:get-arg "-server")))
;;     (and ohh srv)))
    ;; (debug:print-info 0 *default-log-port* "common:run-sync? ohh=" ohh ", srv=" srv)

;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
(define (common:sync-to-megatest.db dbstruct) 
  (let ((start-time         (current-seconds))
	(res                (db:multi-db-sync dbstruct 'new2old)))
    (let ((sync-time (- (current-seconds) start-time)))
      (debug:print-info 3 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
      (if (common:low-noise-print 30 "sync new to old")
	  (debug:print-info 0 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))))
    res))




(define *wdnum* 0)
(define *wdnum*mutex (make-mutex))
;; 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)

        (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)
                  (let ((res (db:multi-db-sync dbstruct 'old2new)))
                    (debug:print-info 13 *default-log-port* "rosync called, " res " records transferred.")))
	      (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)))


        
(define (common:writable-watchdog dbstruct)
  (thread-sleep! 0.05) ;; delay for startup
  (let ((legacy-sync (common:run-sync?))
	(debug-mode  (debug:debug-mode 1))
	(last-time   (current-seconds))
        (this-wd-num     (begin (mutex-lock! *wdnum*mutex) (let ((x *wdnum*)) (set! *wdnum* (add1 *wdnum*)) (mutex-unlock! *wdnum*mutex) x))))
    (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)))
	  (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*)
		   (should-sync      (and (not *time-to-exit*)
                                          (> (- (current-seconds) *db-last-sync*) 5))) ;; sync every five seconds minimum
		   (start-time       (current-seconds))
		   (mt-mod-time      (file-modification-time mtpath))
		   (recently-synced  (< (- start-time mt-mod-time) 4))
		   (will-sync        (and (or need-sync should-sync)
					  (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)
	      ;; (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 ((res (common:sync-to-megatest.db dbstruct))) ;; did we sync any data? If so need to set the db touched flag to keep the server alive
		    (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"))))
	      (if will-sync
		  (begin
		    (mutex-lock! *db-multi-sync-mutex*)
		    (set! *db-sync-in-progress* #f)
		    (set! *db-last-sync* start-time)
		    (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 4)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (if (not *time-to-exit*) (loop))))
	    (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 (launch:setup)
  (if (common:on-homehost?)
      (let ((dbstruct (db:setup)))
	(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.")
	  (common:writable-watchdog dbstruct)))
	(debug:print-info 13 *default-log-port* "watchdog done."))
      (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost")))
      (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.")
	      (server:writable-watchdog dbstruct)))
	    (debug:print-info 13 *default-log-port* "watchdog done."))
	  (debug:print-info 13 *default-log-port* "no need for watchdog on non-homehost"))))


(define (std-exit-procedure)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f

				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (http-client#close-all-connections!)
                              (if (and *runremote*
                                       (remote-conndat *runremote*))
                                  (begin
                                    (http-client#close-all-connections!))) ;; for http-client
                              ;; (if (and *runremote*
                              ;;          (remote-conndat *runremote*))
                              ;;     (begin
                              ;;       (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin

  0)

(define (std-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  ;;(debug:print-info 13 *default-log-port* "got signal "signum)
  (debug:print-error 0 *default-log-port* "Received signal " signum " exiting promptly")
  (debug:print-error 0 *default-log-port* "Received signal " signum " aaa exiting promptly")
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))

(define (special-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  ;;(debug:print-info 13 *default-log-port* "got signal "signum)
  (debug:print-error 0 *default-log-port* "Received signal " signum " sending email befor exiting!!")
  ;;TODO send email to notify admin contact listed in the config that the lisner got killed
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))


(set-signal-handler! signal/int  std-signal-handler)  ;; ^C
(set-signal-handler! signal/term std-signal-handler)

;; (set-signal-handler! signal/stop std-signal-handler)  ;; ^Z NO, do NOT handle ^Z!

;;======================================================================
;; M I S C   U T I L S
;;======================================================================

;; convert stuff to a number if possible

(define (common:which cmds)
  (if (null? cmds)
      #f
      (let loop ((hed (car cmds))
		 (tal (cdr cmds)))
	(let ((res (with-input-from-pipe (conc "which " hed) read-line)))
	  (if (and (string? res)
		   (file-exists? res))
		   (common:file-exists? res))
	      res
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))
  
(define (common:get-install-area)
  (let ((exe-path (car (argv))))
    (if (file-exists? exe-path)
    (if (common:file-exists? exe-path)
	(handle-exceptions
	 exn
	 #f
	 (pathname-directory
	  (pathname-directory 
	   (pathname-directory exe-path))))
	#f)))

;; return first path that can be created or already exists and is writable
;;
(define (common:get-create-writeable-dir dirs)
  (if (null? dirs)
      #f
      (let loop ((hed (car dirs))
		 (tal (cdr dirs)))
	(let ((res (or (and (directory? hed)
			    (file-write-access? hed)
			    hed)
		       (handle-exceptions
			exn
			#f
			   exn
			   (begin
			     (debug:print-info 0 *default-log-port* "could not create " hed ", this might cause problems down the road.")
			     #f)
			(create-directory hed #t)))))
	  (if (and (string? res)
		   (directory? res))
	      res
	      (if (null? tal)
		  #f
		  (loop (car tal)(cdr tal))))))))

(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  (if (args:get-arg "--modepatt") (args:get-arg "--modepatt") "TESTPATT"))
         (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
     ((args:get-arg "--modepatt") ;; modepatt is a forced setting, when set it MUST refer to an existing PATT in the runconfig
      (if rconf
	  (runconfigs-get rconf testpatt-key)
	  #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 args-testpatt))))

(define (common:directory-exists? path-string)
  ;;;; TODO: catch permission denied exceptions and emit appropriate warnings, eg:  system error while trying to access file: "/nfs/pdx/disks/icf_env_disk001/bjbarcla/gwa/issues/mtdev/randy-slow/reproduce/q...
  (common:false-on-exception (lambda () (directory-exists? path-string))
                             message: (conc "Unable to access path: " path-string)
                             ))

;; does the directory exist and do we have write access?
;;
;;    returns the directory or #f
;;
(define (common:directory-writable? path-string)
  (handle-exceptions
   exn
   #f
   (if (and (directory-exists? path-string)
            (file-write-access? path-string))
       path-string
       #f)))

(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (or (and *configdat*
	       (configf:lookup *configdat* "setup" "linktree"))
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree")
	  (if *toppath*
	      (conc *toppath* "/lt")
	      (if (file-exists? "megatest.config") ;; we are in the toppath (new area, mtutils compatible)
		  (conc (current-directory) "/lt")
		  #f)))))
	      #f))))

(define (common:args-get-runname)
  (let ((res (or (args:get-arg "-runname")
		 (args:get-arg ":runname")
		 (getenv "MT_RUNNAME"))))
    ;; (if res (set-environment-variable "MT_RUNNAME" res)) ;; not sure if this is a good idea. side effect and all ...
    res))

(define (common:get-fields cfgdat)
  (let ((fields (hash-table-ref/default cfgdat "fields" '())))
    (map car fields)))

(define (common:args-get-target #!key (split #f)(exit-if-bad #f))
  (let* ((keys    (if (hash-table? *configdat*) (keys:config-get-fields *configdat*) '()))
  (let* ((keys    (if (hash-table? *configdat*) (common:get-fields *configdat*) '()))
	 (numkeys (length keys))
	 (target  (or (args:get-arg "-reqtarg")
		      (args:get-arg "-target")
		      (getenv "MT_TARGET")))
	 (tlist   (if target (string-split target "/" #t) '()))
	 (valid   (if target
		      (or (null? keys) ;; probably don't know our keys yet

	    target)
	(if target
	    (begin
	      (debug:print-error 0 *default-log-port* "Invalid target, spaces or blanks not allowed \"" target "\", target should be: " (string-intersperse keys "/") ", have " tlist " for elements")
	      (if exit-if-bad (exit 1))
	      #f)
	    #f))))

;; looking only (at least for now) at the MT_ variables craft the full testname
;;
(define (common:get-full-test-name)
  (if (getenv "MT_TEST_NAME")
      (if (and (getenv "MT_ITEMPATH")
               (not (equal? (getenv "MT_ITEMPATH") "")))
          (getenv "MT_TEST_NAME")
          (conc (getenv "MT_TEST_NAME") "/" (getenv "MT_ITEMPATH")))
      #f))

;; logic for getting homehost. Returns (host . at-home)
;; IF *toppath* is not set, wait up to five seconds trying every two seconds
;; (this is to accomodate the watchdog)
;;
(define (common:get-homehost #!key (trynum 5))
  ;; called often especially at start up. use mutex to eliminate collisions

				   (thread-sleep! delay-time)
				   (common:get-homehost trynum: (- trynum 1)))
				 (begin
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: Failed to read .homehost file after trying five times. Giving up and exiting, message: "  ((condition-property-accessor 'exn 'message) exn))
				   (exit 1)))
			   (let ((hhf (conc *toppath* "/.homehost")))
			     (if (file-exists? hhf)
			     (if (common:file-exists? hhf)
				 (with-input-from-file hhf read-line)
				 (if (file-write-access? *toppath*)
				     (begin
				       (with-output-to-file hhf
					 (lambda ()
					   (print bestadrs)))
				       (begin

    (if hh
	(cdr hh)
	#f)))

;; do we honor the caches of the config files?
;;
(define (common:use-cache?)
  (not (or (args:get-arg "-no-cache")
	   (and *configdat*
		(equal? (configf:lookup *configdat* "setup" "use-cache") "no")))))

  (let ((res #t)) ;; priority by order of evaluation
    (if *configdat* ;; sillyness here. can't use setup/use-cache to know if we can use the cached files!
	(if (equal? (configf:lookup *configdat* "setup" "use-cache") "no")
	    (set! res #f)
	    (if (equal? (configf:lookup *configdat* "setup" "use-cache") "yes")
		(set! res #t))))
    (if (args:get-arg "-no-cache")(set! res #f)) ;; overrides setting in "setup"
    (if (getenv "MT_USE_CACHE")
	(if (equal? (getenv "MT_USE_CACHE") "yes")
	    (set! res #t)
	    (if (equal? (getenv "MT_USE_CACHE") "no")
		(set! res #f))))    ;; overrides -no-cache switch
    res))
  
;; force use of server?
;;
(define (common:force-server?)
  (let* ((force-setting (configf:lookup *configdat* "server" "force"))
	 (force-type    (if force-setting (string->symbol force-setting) #f))
	 (force-result  (case force-type
			  ((#f)     #f)
			  ((always) #t)
			  ((test)   (if (args:get-arg "-execute") ;; we are in a test
					#t
					#f))
			  (else
			   (debug:print 0 *default-log-port* "ERROR: Bad server force setting " force-setting ", forcing server.")
			   #t)))) ;; default to requiring server
    (if force-result
	(begin
	  (debug:print-info 0 *default-log-port* "forcing use of server, force setting is \"" force-setting "\".")
	  #t)
	#f)))

;; do we honor the caches of the config files?
;;
(define (common:use-cache?)
  (not (or (args:get-arg "-no-cache")
	   (and *configdat*
		(equal? (configf:lookup *configdat* "setup" "use-cache") "no")))))

;;======================================================================
;; M I S C   L I S T S
;;======================================================================

;; items in lista are matched value and position in listb
;; return the remaining items in listb or #f
;;

          ((< (+ load (/ (random 250) 1000))         ;; add a random factor to keep from getting in a rut
              (+ best-load (/ (random 250) 1000))  )
           (set! best-load load)
           (set! best-host hostname)))))
     hosts)
    best-host))




(define (common:wait-for-cpuload maxload numcpus waitdelay #!key (count 1000) (msg #f)(remote-host #f))
(define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (numcpus (if (< 1 numcpus-in) ;; not possible
		      (common:get-num-cpus remote-host)
		      numcpus-in))
	 (maxload (max maxload-in 0.5)) ;; so maxload must be greater than 0.5 for now BUG - FIXME?
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload numcpus))
	 (adjload (* maxload (max 1 numcpus))) ;; possible bug where numcpus (or could be maxload) is zero, crude fallback is to at least use 1
	 (loadjmp (- first next)))
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load " first " exceeding max of " adjload (if msg msg ""))
      (debug:print-info 0 *default-log-port* "server start delayed " waitdelay " seconds due to load " first " exceeding max of " adjload " on server " (or remote-host (get-host-name)) " (normalized load-limit: " maxload ") " (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1)))
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " waitdelay " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! waitdelay)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1))))))
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host)))))

(define (common:wait-for-homehost-load maxload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f))
         (numcpus (common:get-num-cpus hh)))
    (common:wait-for-normalized-load maxload msg hh)))

(define (common:get-num-cpus remote-host)
  (let ((proc (lambda ()
		(let loop ((numcpu 0)
			   (inl    (read-line)))
		  (if (eof-object? inl)
		      numcpu
		      (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
				(+ numcpu 1)
				numcpu)
			    (read-line)))))))
    (if remote-host
	(with-input-from-pipe 
	 (conc "ssh " remote-host " cat /proc/cpuinfo")
	 proc)
	(with-input-from-file "/proc/cpuinfo" proc))))

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload #!key (msg #f)(remote-host #f))
(define (common:wait-for-normalized-load maxload msg remote-host)
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg remote-host: remote-host)))

(define (get-uname . params)
  (let* ((uname-res (process:cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
	 (uname #f))
    (if (null? (car uname-res))
	"unknown"
	(caar uname-res))))

       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key   (car keyval))
			     (val   (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (member key ignorevars)
			(print (if (or (member key ignorevars)
				       (string-search whitesp key))
				   "# setenv "
				   "setenv ")
			       key " " delim (mungeval val) delim)))
		    envvars)))
     (with-output-to-file (conc fname ".sh")
       (lambda ()
          (for-each (lambda (keyval)
		      (let* ((key (car keyval))
			     (val (cdr keyval))
			     (delim (if (string-search whitesp val) 
					"\""
					"")))
			(print (if (or (member key ignorevars)
				       (string-search whitesp key)
				       (string-search ":" key)) ;; internal only values to be skipped.
				   "# export "
				   "export ")
			       key "=" delim (mungeval val) delim)))
                    envvars)))))


(define (common:get-param-mapping #!key (flavor #f))
  "returns alist mapping string keys in testconfig/subrun to megatest command line switches; if flavor is switch-symbol, maps tcmt symbolic switches to megatest switches"
  (let ((default '(("tag-expr"  . "-tagexpr")
                   ("mode-patt" . "-modepatt")
                   ("run-name"  . "-runname")
                   ("contour"   . "-contour")
                   ("target"    . "-target")
                   ("test-patt" . "-testpatt")
                   ("msg"       . "-m")
                   ("log"       . "-log")
                   ("start-dir" . "-start-dir")
                   ("new"       . "-set-state-status"))))
    (if (eq? flavor 'switch-symbol)
        (map (lambda (x)
               (cons (string->symbol (conc "-" (car x))) (cdr x)))
             default)
        default)))

;; set some env vars from an alist, return an alist with original values
;; (("VAR" "value") ...)
;; a value of #f means "unset this var"
;;
(define (alist->env-vars lst)
  (if (list? lst)
      (let ((res '()))
	(for-each (lambda (p)
		    (let* ((var (car  p))
			   (val (cadr p))
			   (prv (get-environment-variable var)))
		      (set! res (cons (list var prv) res))
		      (if val 
			  (setenv var (->string val))
			  (safe-setenv var (->string val))
			  (unsetenv var))))
		  lst)
	res)
      '()))


;; clear vars matching pattern, run proc, set vars back
;; if proc is a string run that string as a command with
;; system.
;;
(define (common:without-vars proc . var-patts)
  (let ((vars (make-hash-table)))

     ((string? proc)(system proc))
     (proc          (proc)))
    (hash-table-for-each
     vars
     (lambda (var val)
       (setenv var val)))
    vars))


(define (common:run-a-command cmd #!key (with-vars #f))
  (let* ((pre-cmd  (dtests:get-pre-command))
         (post-cmd (dtests:get-post-command))
         (fullcmd  (if (or pre-cmd post-cmd)
                       (conc pre-cmd cmd post-cmd)
                       (conc "viewscreen " cmd))))
    (debug:print-info 02 *default-log-port* "Running command: " fullcmd)
    (if with-vars
        (common:without-vars cmd)
        (common:without-vars fullcmd "MT_.*"))))
		  
;;======================================================================
;; T I M E   A N D   D A T E
;;======================================================================

;; Convert strings like "5s 2h 3m" => 60x60x2 + 3x60 + 5
(define (common:hms-string->seconds tstr)
  (let ((parts     (string-split tstr))
  (let ((parts     (string-split-fields "\\w+" tstr))
	(time-secs 0)
	;; s=seconds, m=minutes, h=hours, d=days
	(trx       (regexp "(\\d+)([smhd])")))
	;; s=seconds, m=minutes, h=hours, d=days, M=months, y=years, w=weeks
	(trx       (regexp "(\\d+)([smhdMyw])")))
    (for-each (lambda (part)
		(let ((match  (string-match trx part)))
		  (if match
		      (let ((val (string->number (cadr match)))
			    (unt (caddr match)))
			(if val 
			    (set! time-secs (+ time-secs (* val
							    (case (string->symbol unt)
							      ((s) 1)
							      ((m) 60)
							      ((h) (* 60 60))
							      ((d) (* 24 60 60))
							      (else 0))))))))))
							      ((m) 60) ;; minutes
							      ((h) 3600)
							      ((d) 86400)
							      ((w) 604800)
							      ((M) 2628000) ;; aproximately one month
							      ((y) 31536000)
							      (else #f))))))))))
	      parts)
    time-secs))
		       
(define (seconds->hr-min-sec secs)
  (let* ((hrs (quotient secs 3600))
	 (min (quotient (- secs (* hrs 3600)) 60))
	 (sec (- secs (* hrs 3600)(* min 60))))

  (string-intersperse 
   (map (lambda (x)
          (number->string x 16))
        (map string->number
             (string-split instr)))
   "/"))

;;======================================================================
;; L O C K I N G   M E C H A N I S M S 
;;======================================================================

;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname)
  (if (rmt:get-var keyname)
      #f
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:set-var keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:get-var keyname))))))
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:get-var keyname))))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:get-var keyname) (rmt:del-var keyname))
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

;; simple lock. improve and converge on this one.
;;
(define (common:simple-lock keyname)
  (rmt:no-sync-get-lock keyname))
  

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

(define (common:in-running-test?)
  (and (args:get-arg "-execute") (get-environment-variable "MT_CMDINFO")))

(define (common:get-color-from-status status)
  (cond
   ((equal? status "PASS")    "green")
   ((equal? status "FAIL")    "red")
   ((equal? status "WARN")    "orange")
   ((equal? status "KILLED")  "orange")
   ((equal? status "KILLREQ") "purple")
   ((equal? status "RUNNING") "blue")
   ((equal? status "ABORT")   "brown")
   (else "black")))

;;======================================================================
;; N A N O M S G   C L I E N T
;;======================================================================

;; ;;======================================================================
;; ;; N A N O M S G   C L I E N T
;; ;;======================================================================
;; 
(define (server:get-best-guess-address hostname)
  (let ((res #f))
    (for-each 
     (lambda (adr)
       (if (not (eq? (u8vector-ref adr 0) 127))
	   (set! res adr)))
     ;; NOTE: This can fail when there is no mention of the host in /etc/hosts. FIXME
     (vector->list (hostinfo-addresses (hostname->hostinfo hostname))))
    (string-intersperse 
     (map number->string
	  (u8vector->list
	   (if res res (hostname->ip hostname)))) ".")))


(define (common:send-dboard-main-changed)
  (let* ((dashboard-ips (mddb:get-dashboards)))
    (for-each
     (lambda (ipadr)
       (let* ((soc (common:open-nm-req (conc "tcp://" ipadr)))
	      (msg (conc "main " *toppath*))
	      (res (common:nm-send-receive-timeout soc msg)))
	 (if (not res) ;; couldn't reach that dashboard - remove it from db
	     (print "ERROR: couldn't reach dashboard " ipadr))
	 res))
     dashboard-ips)))
    
    
;;======================================================================
;; D A S H B O A R D   D B 
;;======================================================================

(define (mddb:open-db)
  (let* ((db (open-database (conc (get-environment-variable "HOME") "/.dashboard.db"))))
    (set-busy-handler! db (busy-timeout 10000))
    (for-each
     (lambda (qry)
       (exec (sql db qry)))
     (list 
      "CREATE TABLE IF NOT EXISTS vars       (id INTEGER PRIMARY KEY,key TEXT, val TEXT, CONSTRAINT varsconstraint UNIQUE (key));"
      "CREATE TABLE IF NOT EXISTS dashboards (
          id         INTEGER PRIMARY KEY,
          pid        INTEGER,
          username   TEXT,
          hostname   TEXT,
          ipaddr     TEXT,
          portnum    INTEGER,
          start_time TIMESTAMP DEFAULT (strftime('%s','now')),
             CONSTRAINT hostport UNIQUE (hostname,portnum)
        );"
      ))
    db))

;; register a dashboard 
;;
(define (mddb:register-dashboard port)
  (let* ((pid      (current-process-id))
	 (hostname (get-host-name))
	 (ipaddr   (server:get-best-guess-address hostname))
	 (username (current-user-name)) ;; (car userinfo)))
	 (db      (mddb:open-db)))
    (print "Register monitor, pid: " pid ", hostname: " hostname ", port: " port ", username: " username)
    (exec (sql db "INSERT OR REPLACE INTO dashboards (pid,username,hostname,ipaddr,portnum) VALUES (?,?,?,?,?);")
	   pid username hostname ipaddr port)
    (close-database db)))

;; unregister a monitor
;;
(define (mddb:unregister-dashboard host port)
  (let* ((db      (mddb:open-db)))
    (print "Register unregister monitor, host:port=" host ":" port)
    (exec (sql db "DELETE FROM dashboards WHERE hostname=? AND portnum=?;") host port)
    (close-database db)))

;; get registered dashboards
;;
(define (mddb:get-dashboards)
  (let ((db (mddb:open-db)))
    (query fetch-column
	   (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
;; 
;; 
;; (define (common:send-dboard-main-changed)
;;   (let* ((dashboard-ips (mddb:get-dashboards)))
;;     (for-each
;;      (lambda (ipadr)
;;        (let* ((soc (common:open-nm-req (conc "tcp://" ipadr)))
;; 	      (msg (conc "main " *toppath*))
;; 	      (res (common:nm-send-receive-timeout soc msg)))
;; 	 (if (not res) ;; couldn't reach that dashboard - remove it from db
;; 	     (print "ERROR: couldn't reach dashboard " ipadr))
;; 	 res))
;;      dashboard-ips)))
;;     
;;     
;; ;;======================================================================
;; ;; D A S H B O A R D   D B 
;; ;;======================================================================
;; 
;; (define (mddb:open-db)
;;   (let* ((db (open-database (conc (get-environment-variable "HOME") "/.dashboard.db"))))
;;     (set-busy-handler! db (busy-timeout 10000))
;;     (for-each
;;      (lambda (qry)
;;        (exec (sql db qry)))
;;      (list 
;;       "CREATE TABLE IF NOT EXISTS vars       (id INTEGER PRIMARY KEY,key TEXT, val TEXT, CONSTRAINT varsconstraint UNIQUE (key));"
;;       "CREATE TABLE IF NOT EXISTS dashboards (
;;           id         INTEGER PRIMARY KEY,
;;           pid        INTEGER,
;;           username   TEXT,
;;           hostname   TEXT,
;;           ipaddr     TEXT,
;;           portnum    INTEGER,
;;           start_time TIMESTAMP DEFAULT (strftime('%s','now')),
;;              CONSTRAINT hostport UNIQUE (hostname,portnum)
;;         );"
;;       ))
;;     db))
;; 
;; ;; register a dashboard 
;; ;;
;; (define (mddb:register-dashboard port)
;;   (let* ((pid      (current-process-id))
;; 	 (hostname (get-host-name))
;; 	 (ipaddr   (server:get-best-guess-address hostname))
;; 	 (username (current-user-name)) ;; (car userinfo)))
;; 	 (db      (mddb:open-db)))
;;     (print "Register monitor, pid: " pid ", hostname: " hostname ", port: " port ", username: " username)
;;     (exec (sql db "INSERT OR REPLACE INTO dashboards (pid,username,hostname,ipaddr,portnum) VALUES (?,?,?,?,?);")
;; 	   pid username hostname ipaddr port)
;;     (close-database db)))
;; 
;; ;; unregister a monitor
;; ;;
;; (define (mddb:unregister-dashboard host port)
;;   (let* ((db      (mddb:open-db)))
;;     (print "Register unregister monitor, host:port=" host ":" port)
;;     (exec (sql db "DELETE FROM dashboards WHERE hostname=? AND portnum=?;") host port)
;;     (close-database db)))
;; 
;; ;; get registered dashboards
;; ;;
;; (define (mddb:get-dashboards)
;;   (let ((db (mddb:open-db)))
;;     (query fetch-column
;; 	   (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
    
;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [hosts]
;; arm cubie01 cubie02

				    fallback-launcher
				    (loop (car tal)(cdr tal)))))))
		      ;; no match, try again
		      (if (null? tal)
			  fallback-launcher
			  (loop (car tal)(cdr tal))))))))
	fallback-launcher)))
  

;;======================================================================
;; NMSG AND NEW API
;;======================================================================

;; nm based server experiment, keep around for now.
;;
(define (nm:start-server dbconn #!key (given-host-name #f))
  (let* ((srvdat    (start-raw-server given-host-name: given-host-name))
	 (host-name (srvdat-host srvdat))
	 (soc       (srvdat-soc srvdat)))
    
    ;; start the queue processor (save for second round of development)
    ;;
    (thread-start! (make-thread! (lambda ()(queue-processor dbconn) "Queue processor")))
    ;; msg is an alist
    ;;  'r host:port  <== where to return the data
    ;;  'p params     <== data to apply the command to
    ;;  'e j|s|l      <== encoding of the params. default is s (sexp), if not specified is assumed to be default
    ;;  'c command    <== look up the function to call using this key
    ;;
    (let loop ((msg-in (nn-recv soc)))
      (if (not (equal? msg-in "quit"))
	  (let* ((dat        (decode msg-in))
		 (host-port  (alist-ref 'r dat)) ;; this is for the reverse req rep where the server is a client of the original client
		 (params     (alist-ref 'p dat))
		 (command    (let ((c (alist-ref 'c dat)))(if c (string->symbol c) #f)))
		 (all-good   (and host-port params command (hash-table-exists? *commands* command))))
	    (if all-good
		(let ((cmddat (make-qitem
			       command:   command
			       host-port: host-port
			       params:    params)))
		  (queue-push cmddat) 		;; put request into the queue
		  (nn-send soc "queued"))         ;; reply with "queued"
		(print "ERROR: BAD request " dat))
	    (loop (nn-recv soc)))))
    (nn-close soc)))
  


;;======================================================================
;; D A S H B O A R D   U S E R   V I E W S
;;======================================================================

;; 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 (file-exists? mthome-cfgfile)
    (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 (file-exists? home-cfgfile)
    (if (common:file-exists? home-cfgfile)
	(read-config home-cfgfile view-cfgdat #t))
    view-cfgdat))

;;======================================================================
;; H I E R A R C H I C A L   H A S H   T A B L E S
;;======================================================================

;; Every element including top element is a vector:
;;   <vector subhash value>

(define (hh:make-hh #!key (ht #f)(value #f))
  (vector (or ht    (make-hash-table)) value))

;; used internally
(define-inline (hh:set-ht! hh ht)       (vector-set! hh 0 ht))
(define-inline (hh:get-ht hh)           (vector-ref  hh 0))
(define-inline (hh:set-value! hh value) (vector-set! hh 1 value))
(define-inline (hh:get-value  hh value) (vector-ref  hh 1))

;; given a hierarchial hash and some keys look up the value ...
;;
(define (hh:get hh . keys)
  (if (null? keys)
      (vector-ref hh 1) ;; we have reached the end of the line, return the value sought
      (let ((sub-ht (hh:get-ht hh)))
	(if sub-ht ;; yes, there is more hierarchy
	    (let ((sub-hh (hash-table-ref/default sub-ht (car keys) #f)))
	      (if sub-hh
		  (apply hh:get sub-hh (cdr keys))
		  #f))
	    #f))))

;; given a hierarchial hash, a value and some keys, add needed hierarcy and insert the value
;;
(define (hh:set! hh value . keys)
  (if (null? keys)
      (hh:set-value! hh value) ;; we have reached the end of the line, store the value
      (let ((sub-ht (hh:get-ht hh)))
	(if sub-ht ;; yes, there is more hierarchy
	    (let ((sub-hh (hash-table-ref/default sub-ht (car keys) #f)))
	      (if (not sub-hh) ;; we'll need to add the next level of hierarchy
		  (let ((new-sub-hh (hh:make-hh)))
		    (hash-table-set! sub-ht (car keys) new-sub-hh)
		    (apply hh:set! new-sub-hh value (cdr keys)))
		  (apply hh:set! sub-hh value (cdr keys))))    ;; call the sub-hierhash with remaining keys
	    (begin
	      (hh:set-ht! hh (make-hash-table))
	      (apply hh:set! hh value keys))))))
  
;; Manage pkts, used in servers, tests and likely other contexts so put
;; in common
;;======================================================================

(define common:pkts-spec
  '((default . ((parent    . P)
                (action    . a)
                (filename  . f)))
    (configf . ((parent    . P)
                (action    . a)
                (filename  . f)))
    (server  . ((action    . a)
		(pid       . d)
		(ipaddr    . i)
		(port      . p)
		(parent    . P)))
    			  
    (test    . ((cpuuse    . c)
		(diskuse   . d)
		(item-path . i)
		(runname   . r)
		(state     . s)
		(target    . t)
		(status    . u)
		(parent    . P)))))

(define (common:get-pkts-dirs mtconf use-lt)
  (let* ((pktsdirs-str (or (configf:lookup mtconf "setup"  "pktsdirs")
			   (and use-lt
				(conc (or *toppath*
					  (current-directory))
				      "/lt/.pkts"))))
	 (pktsdirs  (if pktsdirs-str
			(string-split pktsdirs-str " ")
			#f)))
    pktsdirs))

;; use-lt is use linktree "lt" link to find pkts dir
(define (common:save-pkt pktalist-in mtconf use-lt #!key (add-only #f)) ;; add-only saves the pkt only if there is a parent already
  (if (or add-only
	  (hash-table-exists? *pkts-info* 'last-parent))
      (let* ((parent   (hash-table-ref/default *pkts-info* 'last-parent #f))
	     (pktalist (if parent
			   (cons `(parent . ,parent)
				 pktalist-in)
			   pktalist-in)))
	(let-values (((uuid pkt)
		      (alist->pkt pktalist common:pkts-spec)))
	  (hash-table-set! *pkts-info* 'last-parent uuid)
	  (let ((pktsdir (or (hash-table-ref/default *pkts-info* 'pkts-dir #f)
			     (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
				    (pktsdir   (car pktsdirs))) ;; assume it is there
			       (hash-table-set! *pkts-info* 'pkts-dir pktsdir)
			       pktsdir))))
            (handle-exceptions
             exn
             (debug:print-info 0 "failed to write out packet to " pktsdir) ;; don't care if this failed for now but MUST FIX - BUG!!
             (if (not (file-exists? pktsdir))
                 (create-directory pktsdir #t))
             (with-output-to-file
                 (conc pktsdir "/" uuid ".pkt")
               (lambda ()
                 (print pkt)))))))))
	
(define (common:with-queue-db mtconf proc #!key (use-lt #f)(toppath-in #f))
  (let* ((pktsdirs (common:get-pkts-dirs mtconf use-lt))
	 (pktsdir  (if pktsdirs (car pktsdirs) #f))
	 (toppath  (or (configf:lookup mtconf "scratchdat" "toppath")
		       toppath-in))
	 (pdbpath  (or (configf:lookup mtconf "setup"  "pdbpath") pktsdir)))
    (cond
     ((not (and  pktsdir toppath pdbpath))
      (debug:print 0 *default-log-port* "ERROR: settings are missing in your megatest.config for area management.")
      (debug:print  0 *default-log-port* "  you need to have pktsdir in the [setup] section."))
     ((not (common:file-exists? pktsdir))
      (debug:print 0 *default-log-port* "ERROR: pkts directory not found " pktsdir))
     ((not (equal? (file-owner pktsdir)(current-effective-user-id)))
      (debug:print 0 *default-log-port* "ERROR: directory " pktsdir " is not owned by " (current-effective-user-name)))
     (else
	(let* ((pdb  (open-queue-db pdbpath "pkts.db"
				    schema: '("CREATE TABLE groups (id INTEGER PRIMARY KEY,groupname TEXT, CONSTRAINT group_constraint UNIQUE (groupname));"))))
	  (proc pktsdirs pktsdir pdb)
	  (dbi:close pdb))))))

(define (common:load-pkts-to-db mtconf #!key (use-lt #f))
  (common:with-queue-db
   mtconf
   (lambda (pktsdirs pktsdir pdb)
     (for-each
      (lambda (pktsdir) ;; look at all
	(cond
	 ((not (common:file-exists? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory " pktsdir " does not exist."))
	 ((not (directory? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not a directory."))
	 ((not (file-read-access? pktsdir))
	  (debug:print 0 *default-log-port* "ERROR: packets directory path " pktsdir " is not readable."))
	 (else
	  (debug:print-info 0 *default-log-port* "Loading packets found in " pktsdir)
	  (let ((pkts (glob (conc pktsdir "/*.pkt"))))
	    (for-each
	     (lambda (pkt)
	       (let* ((uuid    (cadr (string-match ".*/([0-9a-f]+).pkt" pkt)))
		      (exists  (lookup-by-uuid pdb uuid #f)))
		 (if (not exists)
		     (let* ((pktdat (string-intersperse
				     (with-input-from-file pkt read-lines)
				     "\n"))
			    (apkt   (pkt->alist pktdat))
			    (ptype  (alist-ref 'T apkt)))
		       (add-to-queue pdb pktdat uuid (or ptype 'cmd) #f 0)
		       (debug:print 4 *default-log-port* "Added " uuid " of type " ptype " to queue"))
		     (debug:print 4 *default-log-port* "pkt: " uuid " exists, skipping...")
		     )))
	     pkts)))))
      pktsdirs))
   use-lt: use-lt))

(define (common:get-pkt-alists pkts)
  (map (lambda (x)
	 (alist-ref 'apkt x)) ;; 'pkta pulls out the alist from the read pkt
       pkts))

;; given list of pkts (alist mode) return list of D cards as Unix epoch, sorted descending
;; also delete duplicates by target i.e. (car pkt)
;;
(define (common:get-pkt-times pkts)
  (delete-duplicates
   (sort 
    (map (lambda (x)
	   `(,(alist-ref 't x) . ,(string->number (alist-ref 'D x))))
	 pkts)
    (lambda (a b)(> (cdr a)(cdr b))))      ;; sort descending
   (lambda (a b)(equal? (car a)(car b))))) ;; remove duplicates by target



;; accept an alist or hash table containing envvar/env value pairs (value of #f causes unset) 
;;   execute thunk in context of environment modified as per this list
;;   restore env to prior state then return value of eval'd thunk.
;;   ** this is not thread safe **
(define (common:with-env-vars delta-env-alist-or-hash-table thunk)
  (let* ((delta-env-alist (if (hash-table? delta-env-alist-or-hash-table)
                              (hash-table->alist delta-env-alist-or-hash-table)
                              delta-env-alist-or-hash-table))
         (restore-thunks
          (filter
           identity
           (map (lambda (env-pair)
                  (let* ((env-var     (car env-pair))
                         (new-val     (cadr env-pair))
                         (current-val (get-environment-variable env-var))
                         (restore-thunk
                          (cond
                           ((not current-val) (lambda () (unsetenv env-var)))
                           ((not (string? new-val)) #f)
                           ((eq? current-val new-val) #f)
                           (else 
                            (lambda () (setenv env-var current-val))))))
                    ;;(when (not (string? new-val))
                    ;;    (debug:print 0 *default-log-port* " PROBLEM: not a string: "new-val"\n from env-alist:\n"delta-env-alist)
                    ;;    (pp delta-env-alist)
                    ;;    (exit 1))
                        
                    
                    (cond
                     ((not new-val)  ;; modify env here
                      (unsetenv env-var))
                     ((string? new-val)
                      (setenv env-var new-val)))
                    restore-thunk))
                delta-env-alist))))
    (let ((rv (thunk)))
      (for-each (lambda (x) (x)) restore-thunks) ;; restore env to original state
      rv)))

(define *common:thread-punchlist* (make-hash-table))
(define (common:send-thunk-to-background-thread thunk #!key (name #f))
  ;;(BB> "launched thread " name)

  ;; we need a unique name for the thread.
  (let* ((realname (if name
                       (if (not (hash-table-ref/default *common:thread-punchlist* name #f))
                           name
                           (conc name"-" (symbol->string (gensym))))
                       (conc "anonymous-"(symbol->string (gensym)))))
         (realthunk (lambda ()
                      (let ((res (thunk)))
                        (hash-table-delete! *common:thread-punchlist* realname)
                        res)))
         (thread   (make-thread realthunk realname)))
    (hash-table-set! *common:thread-punchlist* realname thread)
    (thread-start! thread)
    ))

(define (common:join-backgrounded-threads)
  ;; may need to trap and ignore exceptions -- dunno how atomic threads are...
  (for-each
   (lambda (thread-name)
     (let* ((thread (hash-table-ref/default *common:thread-punchlist* thread-name #f)))
       (if thread
           (handle-exceptions
           exn
           #t ;; just ignore it, it might have died in the meantime so joining it will throw an exception
           (thread-join! thread))
           )))
   (hash-table-keys *common:thread-punchlist*)))