Megatest

        (archiver       (let ((s (configf:lookup *configdat* "archive" "archiver")))
			   (if s (string->symbol s) 'bup)))
        (rsync-exe (or (configf:lookup *configdat* "archive" "rsync") "rsync"))   
        (print-prefix      "Running: ") 
        (archive-info (archive:allocate-new-archive-block blockid-cache *toppath* tsname min-space target-patt run-patt "megatest-db"))
	(archive-dir  (if archive-info (cdr archive-info) #f))
	(archive-id   (if archive-info (car archive-info) -1))
        (home-host (common:get-homehost))
        (home-host (server:get-homehost))
        (archive-time (seconds->std-time-str (current-seconds)))
        (archive-staging-db (conc *toppath* "/.db-snapshot/archive_" archive-time))
        (tmp-db-path (conc (common:get-db-tmp-area) "/megatest.db"))
        (dbfile             (conc  archive-staging-db "/megatest.db"))) 
        (create-directory archive-staging-db #t)
        (let-values (((pid-val exit-status exit-code) (run-n-wait rsync-exe params: (list "-v" (conc (car home-host) ":"tmp-db-path) archive-staging-db) print-cmd: print-prefix)))
            (if (eq? exit-code 0)   

		    (set! *runremote* (make-remote))
                    (let* ((server-info (remote-server-info *runremote*))) 
                      (if server-info
                        (begin
                          (remote-server-url-set! *runremote* (server:record->url server-info))
                          (remote-server-id-set! *runremote* (server:record->id server-info)))))))
	      (if (and host port server-id)
		  (let* ((start-res (case *transport-type*
				      ((http)(http-transport:client-connect host port server-id))))
		  (let* ((start-res (http-transport:client-connect host port server-id))
			 (ping-res  (case *transport-type* 
				      ((http)(rmt:login-no-auto-client-setup start-res)))))
			 (ping-res  (rmt:login-no-auto-client-setup start-res)))
		    (if (and start-res
			     ping-res)
			(let ((runremote (or area-dat *runremote*))) ;; it might have been generated only a few statements ago
			  (if runremote
			      (begin
				(remote-conndat-set! runremote start-res) ;; (hash-table-set! runremote run-id start-res)
				(debug:print-info 2 *default-log-port* "connected to " (http-transport:server-dat-make-url start-res))

(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 *on-exit-procs* '()) ;; add procs to this list to be executed on exit
(define *default-area-tag* "local")

;; DATABASE
;; (define *dbstruct-dbs*         #f) ;; used to cache the dbstruct in db:setup. Goal is to remove this.
;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))

    ((skip) "SKIP")
    (else "FAIL")))

(define (common:logpro-exit-code->test-status exit-code)
  (status-sym->string (common:logpro-exit-code->status-sym exit-code)))

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (hh-dat            (server:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        #f) ;; (server:check-if-running *toppath*) #f))
  (server-id         #f)
  (server-info       (if *toppath* (server:check-if-running *toppath*) #f))
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (connect-time      (current-seconds))
  (conndat           #f)
  (transport         *transport-type*)

      #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
  (mutex-lock! *homehost-mutex*)
  (cond
   (*home-host*
    (mutex-unlock! *homehost-mutex*)
    *home-host*)
   ((not *toppath*)
    (mutex-unlock! *homehost-mutex*)
    (launch:setup) ;; safely mutexed now
    (if (> trynum 0)
	(begin
	  (thread-sleep! 2)
	  (common:get-homehost trynum: (- trynum 1)))
	#f))
   (else
    (let* ((currhost (get-host-name))
	   (bestadrs (server:get-best-guess-address currhost))
	   ;; first look in config, then look in file .homehost, create it if not found
	   (homehost (or (configf:lookup *configdat* "server" "homehost" )
			 (handle-exceptions
			     exn
			     (if (> trynum 0)
				 (let ((delay-time (* (- 5 trynum) 5)))
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)"] Failed to read .homehost file, delaying "
						delay-time " seconds and trying again, message: "  ((condition-property-accessor 'exn 'message) exn)
						", exn=" exn)
				   (thread-sleep! delay-time)
				   (common:get-homehost trynum: (- trynum 1)))
				 (begin
				   (mutex-unlock! *homehost-mutex*)
				   (debug:print 0 *default-log-port* "ERROR: ["(common:human-time)
						"] 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 (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
					 (mutex-unlock! *homehost-mutex*)
					 (car (common:get-homehost))))
				     #f))))))
	   (at-home  (or (equal? homehost currhost)
			 (equal? homehost bestadrs))))
      (set! *home-host* (cons homehost at-home))
      (mutex-unlock! *homehost-mutex*)
      *home-host*))))

;;======================================================================
;; am I on the homehost?
;;
(define (common:on-homehost?)
  (let ((hh (common:get-homehost)))
    (if hh
	(cdr hh)
	#f)))

;;======================================================================
;; do we honor the caches of the config files?
;;
(define (common:use-cache?)
  (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!

		    (host-last-used-set! rec curr-time)
		    new-best)
		  (if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))

(define (common:wait-for-homehost-load maxnormload 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)))
                     (server:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f)))
    (common:wait-for-normalized-load maxnormload msg hh)))

(define (common:get-num-cpus remote-host)
  (let* ((actual-host (or remote-host (get-host-name))))
    ;; hosts had better not be changing the number of cpus too often!
    (or (hash-table-ref/default *numcpus-cache* actual-host #f)

			(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
  (if (or (not 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))))
	    (debug:print 0 *default-log-port* "pktsdir: "pktsdir)
            (handle-exceptions
             exn
             (debug:print-info 0 "failed to write out packet to " pktsdir ", exn=" exn) ;; 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")

          (debug:print 0 *default-log-port* "Failed to find megatest.config, exiting") 
          (exit 1)
        )
    )

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


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

	       (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 (std-exit-procedure)
  ;;(common:telemetry-log-close)
  (on-exit (lambda () 0))
  (on-exit (lambda () 0)) ;; why is this here?
  ;;(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
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
    (if (and no-hurry
	     (debug:debug-mode 18))
	(rmt:print-db-stats))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-dbs* (db:close-all *dbstruct-dbs*)) ;; one second allocated
			      (if (list? *on-exit-procs*)
				  (for-each
				   (lambda (proc)
				     (proc))
				   *on-exit-procs*))
			      (if *task-db*    
				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)

			  (debug:print-info 0 *default-log-port* "Waiting for server alive signature")
                          (mutex-lock! *heartbeat-mutex*)
                          (set! sdat *server-info*)
                          (mutex-unlock! *heartbeat-mutex*)
                          (if (and sdat
				   (not changed)
				   (> (- (current-seconds) start-time) 2))
			      (let* ((servinfodir (conc *toppath*"/.servinfo"))
				     (ipaddr      (car sdat))
				     (port        (cadr sdat))
				     (servinf     (conc servinfodir"/"ipaddr":"port)))
				(if (not (file-exists? servinfodir))
				    (create-directory servinfodir #t))
				(with-output-to-file servinf
				  (lambda ()
				    (print "SERVER STARTED: "ipaddr":"port" AT "(current-seconds)" server-id: "(server:get-client-signature))
				    (print "started: "(seconds->year-week/day-time (current-seconds)))))
				(set! *on-exit-procs* (cons
						       (lambda ()
			      (begin
							 (delete-file* servinf))
						       *on-exit-procs*))
				;; put data about this server into a simple flat file host.port
				(debug:print-info 0 *default-log-port* "Received server alive signature")
                                (common:save-pkt `((action . alive)
                                #;(common:save-pkt `((action . alive)
                                                   (T      . server)
                                                   (pid    . ,(current-process-id))
                                                   (ipaddr . ,(car sdat))
                                                   (port   . ,(cadr sdat)))
                                                 *configdat* #t)
				sdat)
                              (begin
				(debug:print-info 0 *default-log-port* "Still waiting, last-sdat=" last-sdat)
                                (sleep 4)
				(if (> (- (current-seconds) start-time) 120) ;; been waiting for two minutes
				    (begin
				    (let* ((ipaddr  (car sdat))
					   (port    (cadr sdat))
					   (servinf (conc *toppath*"/.servinfo/"ipaddr":"port)))
				      (debug:print-error 0 *default-log-port* "transport appears to have died, exiting server")
				      ;; (delete-file* servinf) ;; handled by on-exit, can be removed
                                      (common:save-pkt `((action . died)
                                      #;(common:save-pkt `((action . died)
                                                         (T      . server)
                                                         (pid    . ,(current-process-id))
                                                         (ipaddr . ,(car sdat))
                                                         (port   . ,(cadr sdat))
                                                         (msg    . "Transport died?"))
						       *configdat* #t)
				      (exit))

    ;; 		      (if (eq? *number-non-write-queries* 0)
    ;; 			  "n/a (no queries)"
    ;; 			  (/ *total-non-write-delay* 
    ;; 			     *number-non-write-queries*))
    ;; 		      " ms")
    
    (db:print-current-query-stats)
    (common:save-pkt `((action . exit)
    #;(common:save-pkt `((action . exit)
                       (T      . server)
                       (pid    . ,(current-process-id)))
                     *configdat* #t)
    *configdat* #t)

    ;; remove .servinfo file(s) here
    
    (debug:print-info 0 *default-log-port* "Server shutdown complete. Exiting")
    (exit)))

;; all routes though here end in exit ...
;;
;; start_server? 
;;

	  (exit)))
    ;; lets not even bother to start if there are already three or more server files ready to go
    #;(let* ((num-alive   (server:get-num-alive (server:get-list *toppath*))))
      (if (> num-alive 3)
          (begin
            (cleanup-proc (conc "ERROR: Aborting server start because there are already " num-alive " possible servers either running or starting up"))
            (exit))))
  (common:save-pkt `((action . start)
		     (T      . server)
		     (pid    . ,(current-process-id)))
		   *configdat* #t)
    #;(common:save-pkt `((action . start)
		       (T      . server)
		       (pid    . ,(current-process-id)))
		     *configdat* #t)
    (let* ((th2 (make-thread (lambda ()
                               (debug:print-info 0 *default-log-port* "Server run thread started")
                               (http-transport:run 
                                (if (args:get-arg "-server")
                                    (args:get-arg "-server")
                                    "-")
                                )) "Server run"))

      (set! cmdparms (base64:base64-encode 
		      (z3:encode-buffer 
		       (with-output-to-string
			 (lambda () ;; (list 'hosts     hosts)
			   (write (list (list 'testpath  test-path)
					;; (list 'transport (conc *transport-type*))
					;; (list 'serverinf *server-info*)
					(list 'homehost  (let* ((hhdat (common:get-homehost)))
					(list 'homehost  (let* ((hhdat (server:get-homehost)))
							   (if hhdat
							       (car hhdat)
							       #f)))
					(list 'serverurl (if *runremote*
							     (remote-server-url *runremote*)
							     #f)) ;;
					(list 'areaname  (common:get-testsuite-name))

;; for some switches always print the command to stderr
;;
(if (args:any? "-run" "-runall" "-remove-runs" "-set-state-status" "-kill-runs" "-kill-rerun")
    (debug:print 0 *default-log-port* (string-intersperse (argv) " ")))

;; some switches imply homehost. Exit here if not on homehost
;;
(let ((homehost-required  (list "-cleanup-db" "-server")))
(let ((homehost-required  (list "-cleanup-db")))
  (if (apply args:any? homehost-required)
      (if (not (common:on-homehost?))
	  (for-each
	   (lambda (switch)
	     (if (args:get-arg switch)
		 (begin
		   (debug:print 0 *default-log-port* "ERROR: you must be on the homehost to run with " switch

    
    ;; DOT SET_HOMEHOST; // leaving off - doesn't really add to the clarity
    ;; DOT MUTEXLOCK -> SET_HOMEHOST [label="no homehost?"];
    ;; DOT SET_HOMEHOST -> MUTEXLOCK;
    ;; ensure we have a homehost record
    (if (not (pair? (remote-hh-dat runremote)))  ;; not on homehost
	(thread-sleep! 0.1) ;; since we shouldn't get here, delay a little
	(remote-hh-dat-set! runremote (common:get-homehost)))
	(remote-hh-dat-set! runremote (server:get-homehost)))
    
    ;;(print "BB> readonly-mode is "readonly-mode" dbfile is "dbfile)
    (cond
     #;((> (- (current-seconds)(remote-connect-time runremote)) 180) ;; reconnect to server every 180 seconds
      (debug:print 0 *default-log-port* "Forcing reconnect to server(s) due to 180 second timeout.")
      (set! *runremote* #f)
      ;; BUG: close-connections should go here?

;;
(define  (server:run areapath) ;; areapath is *toppath* for a given testsuite area
  (let* ((curr-host   (get-host-name))
         ;; (attempt-in-progress (server:start-attempted? areapath))
         ;; (dot-server-url (server:check-if-running areapath))
	 (curr-ip     (server:get-best-guess-address curr-host))
	 (curr-pid    (current-process-id))
	 (homehost    (common:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (homehost    (server:get-homehost)) ;; configf:lookup *configdat* "server" "homehost" ))
	 (target-host (car homehost))
	 (testsuite   (common:get-testsuite-name))
	 (logfile     (conc areapath "/logs/server.log")) ;; -" curr-pid "-" target-host ".log"))
	 (profile-mode (or (configf:lookup *configdat* "misc" "profilesw")
			   ""))
	 (cmdln (conc (common:get-megatest-exe)
		      " -server " (or target-host "-") (if (equal? (configf:lookup *configdat* "server" "daemonize") "yes")

		(lambda (a b)
		  (< (list-ref a 3)
		     (list-ref b 3))))))
    (if (> (length slst) nums)
	(take slst nums)
	slst)))

;; switch from server:get-list to server:get-servers-info
;;
(define (server:get-first-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and srvrs
	     (not (null? srvrs)))
	(car srvrs)
	#f)))

        *my-client-signature*)))


;; if server-start-last exists, and wasn't old enough, wait <idle time> + 1, then call this function recursively until it is old enough.
;; if it is old enough, overwrite it and wait 0.25 seconds.
;; if it then has the wrong server key, wait <idle time> + 1 and call this function recursively.
;;
(define (server:wait-for-server-start-last-flag areapath)
#;(define (server:wait-for-server-start-last-flag areapath)
  (let* ((start-flag (conc areapath "/logs/server-start-last"))
	 ;;; THIS INTERACTS WITH [server] timeout. Suggest using 0.1 or above for timeout (6 seconds)
	 (idletime    (configf:lookup-number *configdat* "server" "idletime" default: 4))
	 (server-key (conc (get-host-name) "-" (current-process-id))))
    (if (file-exists? start-flag)
	(let* ((fmodtime (file-modification-time start-flag))
	       (delta    (- (current-seconds) fmodtime))

	       (begin
		 (debug:print-info 0 *default-log-port* "Gating server start, last start: "
				   (seconds->time-string fmodtime) ", time since last start: " delta ", required idletime: " idletime ", gating reason:" (if old-enough "another job started a server" "too soon to start another server"))
		 
		 (thread-sleep! ( + 1 idletime))
		 (server:wait-for-server-start-last-flag areapath)))))))

;; oldest server alive determines host then choose random of youngest
;; five servers on that host
;;
(define (server:get-servers-info areapath)
  (let* ((servinfodir (conc *toppath*"/.servinfo"))
	 (allfiles    (glob (conc servinfodir"/*")))
	 (res         (make-hash-table)))
    (for-each
     (lambda (f)
       (let* ((hostport  (pathname-strip-directory f))
	      (serverdat (server:logf-get-start-info f)))
	 (hash-table-set! res hostport serverdat)))
     allfiles)
    res))

;; oldest server alive determines host then choose random of youngest
;; five servers on that host
;;
;; mode:
;;   best - get best server (random of newest five)
;;   home - get home host based on oldest server
;;   info - print info
(define (server:choose-server areapath #!optional (mode 'best))
  ;; age is current-starttime
  ;; find oldest alive
  ;;   1. sort by age ascending and ping until good
  ;; find alive rand from youngest
  ;;   1. sort by age descending
  ;;   2. take five
  ;;   3. check alive, discard if not and repeat
  (let* ((serversdat  (server:get-servers-info areapath))
	 (by-time-asc (sort (hash-table-keys serversdat) ;; list of "host:port"
			    (lambda (a b)
			      (>= (list-ref (hash-table-ref serversdat a) 2)
				  (list-ref (hash-table-ref serversdat b) 2))))))
    (if (not (null? by-time-asc))
	(let* ((oldest     (last by-time-asc))
	       (oldest-dat (hash-table-ref serversdat oldest))
	       (host       (list-ref oldest-dat 0))
	       (all-valid  (filter (lambda (x)
				     (equal? host (list-ref (hash-table-ref serversdat x) 0)))
				   by-time-asc)))
	  (case mode
	    ((info)
	     (print "oldest: "oldest-dat", selected host: "host", all-valid: "all-valid)
	     (print "youngest: "(hash-table-ref serversdat (car all-valid))))
	    ((home) host)
	    ((best)(if (> (length all-valid) 5)
		       (map (lambda (x)
			      (hash-table-ref serversdat x))
			    (take all-valid 5))
		       all-valid))
	    (else
	     (debug:print 0 *default-log-port* "ERROR: invalid command "mode)
	     #f)))
	#f)))
        

(define (server:get-homehost #!key (trynum 5))
  ;; called often especially at start up. use mutex to eliminate collisions
  (mutex-lock! *homehost-mutex*)
  (cond
   (*home-host*
    (mutex-unlock! *homehost-mutex*)
    *home-host*)
   ((not *toppath*)
    (mutex-unlock! *homehost-mutex*)
    (launch:setup) ;; safely mutexed now
    (if (> trynum 0)
	(begin
	  (thread-sleep! 2)
	  (server:get-homehost trynum: (- trynum 1)))
	#f))
   (else
    (let* ((currhost (get-host-name))
	   (bestadrs (server:get-best-guess-address currhost))
	   (homehost (server:choose-server *toppath* 'home))
	   (at-home  (or (equal? homehost currhost)
			 (equal? homehost bestadrs))))

      ;; if no homehost start server, wait a bit and check again
      (if homehost
	  (begin
	    (set! *home-host* (cons homehost at-home))
	    (mutex-unlock! *homehost-mutex*)
	    *home-host*)
	  (begin
	    (server:kind-run *toppath*)
	    (thread-sleep! 5)
	    (server:get-homehost trynum: (- trynum 1))))))))

;;======================================================================
;; am I on the homehost?
;;
(define (common:on-homehost?)
  (let ((hh (server:get-homehost)))
    (if hh
	(cdr hh)
	#f)))

	  
;; kind start up of server, wait before allowing another server for a given
;; area to be launched
;;
(define (server:kind-run areapath)
  ;; look for $MT_RUN_AREA_HOME/logs/server-start-last
  ;; and wait for it to be at least <server idletime> seconds old
  (server:wait-for-server-start-last-flag areapath)
  (if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
  ;; (server:wait-for-server-start-last-flag areapath)
  (server:run areapath)
  #;(if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
      (let* ((lock-file    (conc areapath "/logs/server-start.lock")))
	(let* ((start-flag (conc areapath "/logs/server-start-last")))
	  (common:simple-file-lock-and-wait lock-file expire-time: 25)
	  (debug:print-info  2 *default-log-port* "server:kind-run: touching " start-flag)
	  (system (conc "touch " start-flag)) ;; lazy but safe
	  (server:run areapath)
	  (thread-sleep! 20) ;; don't release the lock for at least a few seconds. And allow time for the server startup to get to "SERVER STARTED".