Megatest

;;
;; There are two scenarios. 
;;   1. We are a test manager and we received *transport-type* and *runremote* via cmdline
;;   2. We are a run tests, list runs or other interactive process and we must figure out
;;      *transport-type* and *runremote* from the monitor.db
;;
;; client:setup
;;
;; lookup_server, need to remove *runremote* stuff
;;
(define (client:setup run-id #!key (remaining-tries 3))
  (if (<= remaining-tries 0)

      (begin
	(debug:print 0 "ERROR: failed to start or connect to server for run-id " run-id)
	(exit 1))






      (let ((server-dat (and run-id (hash-table-ref/default *runremote* run-id #f))))

	(if server-dat


	    (let ((start-res (http-transport:client-connect run-id
							    (tasks:hostinfo-get-interface server-dat)

							    (tasks:hostinfo-get-port      server-dat))))
	      (if start-res ;; sucessful login?
		  (begin
		    (hash-table-set! *runremote* run-id server-dat)
		    server-dat)

		  (begin    ;; login failed



		    (hash-table-delete! *runremote* run-id)
		    (open-run-close tasks:server-force-clean-run-record
				    tasks:open-db
				    run-id 
				    (tasks:hostinfo-get-interface server-dat)
				    (tasks:hostinfo-get-port      server-dat))
		    (client:setup run-id remaining-tries: (- remaining-tries 1)))))
	    (let* ((server-dat (open-run-close tasks:get-server tasks:open-db run-id)))
	      (if server-dat



		  (let ((start-res (http-transport:client-connect run-id
								  (tasks:hostinfo-get-interface server-dat)
								  (tasks:hostinfo-get-port      server-dat))))
		    (if start-res
			(begin
			  (hash-table-set! *runremote* run-id server-dat)
			  server-dat)
			(begin    ;; login failed
			  (hash-table-delete! *runremote* run-id)
			  (open-run-close tasks:server-force-clean-run-record
					  tasks:open-db
					  run-id 
					  (tasks:hostinfo-get-interface server-dat)
					  (tasks:hostinfo-get-port      server-dat))
			  (server:try-running run-id)
			  (thread-sleep! 3) ;; give server a little time to start up
			  (client:setup run-id remaining-tries: (- remaining-tries 1)))))
		  (begin    ;; no server registered

		    (server:try-running run-id)
		    (thread-sleep! 3) ;; give server a little time to start up
		    (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))

;; keep this as a function to ease future 
(define (client:start run-id server-info)
  (http-transport:client-connect run-id 
				 (tasks:hostinfo-get-interface server-info)
				 (tasks:hostinfo-get-port server-info)))

;; client:signal-handler
(define (client:signal-handler signum)
  (handle-exceptions
   exn
   (debug:print " ... exiting ...")
   (let ((th1 (make-thread (lambda ()

			       immediate
			       flush
			       sync
			       set-verbosity
			       killserver
			       ))

(define (db:login dbstruct calling-path calling-version run-id client-signature)
  (cond 
   ((not (equal? calling-path *toppath*))
    (list #f "Login failed due to mismatch paths: " calling-path ", " *toppath*))
   ((not (equal? *run-id* run-id))
    (list #f "Login failed due to mismatch run-id: " run-id ", " *run-id*))
   ((not (equal? megatest-version calling-version))
    (list #f "Login failed due to mismatch megatest version: " calling-version ", " megatest-version))
   (else
    (hash-table-set! *logged-in-clients* client-signature (current-seconds))
    '(#t "successful login"))))


(define (db:general-call db stmtname params)
  (let ((query (let ((q (alist-ref (if (string? stmtname)
				       (string->symbol stmtname)
				       stmtname)
				   db:queries)))
 		 (if q (car q) #f))))

all : server.pdf megatest_manual.html client.pdf

megatest_manual.html : megatest_manual.txt getting_started.txt writing_tests.txt reference.txt ../plan.txt
	asciidoc megatest_manual.txt
	dos2unix megatest_manual.html

server.pdf : server.dot
	dot -Tpdf server.dot > server.pdf
	
client.pdf : client.dot
	dot -Tpdf client.dot > client.pdf

clean:
	rm -f megatest_manual.html

digraph G {

    // put client after server so server_start node is visible
    //
    subgraph cluster_2 {
        node [style=filled,shape=box];
	
	"client:setup start"     -> runremote_lookup_server;
	runremote_lookup_server  -> login_attempt [label="have server"];
	runremote_lookup_server  -> monitordb_lookup_server [label="no server"];

	monitordb_lookup_server  -> login_attempt [label="have server"];
	monitordb_lookup_server  -> server_start_remote [label="no server"];

	server_start_remote      -> delay_2_sec;
	delay_2_sec              -> runremote_lookup_server;

	login_attempt            -> "rmt:send-receive_start" [label="login sucessful"];
	"rmt:send-receive_start" -> "rmt:send-receive_start";

	"rmt:send-receive_start" -> runremote_lookup_server [label=exception];
	login_attempt            -> clear_runremote [label="login failed"];

	"remove_running > 5s"    -> runremote_lookup_server;

	subgraph cluster_3 {
		node [style=filled];
		clear_runremote          -> "remove_running > 5s";
	}

        label = "client:setup";
        color=green;
    }

}

digraph G {



    subgraph cluster_1 {
        node [style=filled,shape=box];


	check_available_queue       -> remove_entries_over_10s_old;
	remove_entries_over_10s_old -> set_available [label="num_avail < 3"];
	remove_entries_over_10s_old -> exit [label="num_avail > 2"];

	set_available               -> delay_2s;
	delay_2s          -> check_place_in_queue;



	check_place_in_queue        -> "http:transport-launch" [label="at head"];
	check_place_in_queue        -> exit [label="not at head"];


	"client:login"              -> "server:shutdown" [label="login failed"];
	"server:shutdown"           -> exit;	

	subgraph cluster_2 {

		"http:transport-launch"       -> "http:transport-run";
		"http:transport-launch"       -> "http:transport-keep-running";

		"http:transport-keep-running" -> "tests running?";
		"tests running?"              -> "client:login" [label=yes];

		"tests running?"              -> "server:shutdown" [label=no];
		"client:login"                -> delay_5s [label="login ok"];
		delay_5s                      -> "http:transport-keep-running";
	}







	// start_server -> "server_running?";
	// "server_running?" -> set_available [label="no"];
	// "server_running?" -> delay_2s [label="yes"];
	// delay_2s -> "still_running?";
	// "still_running?" -> ping_server [label=yes];
	// "still_running?" -> set_available [label=no];
	// ping_server -> exit [label=alive];
	// ping_server -> remove_server_record [label=dead];
	// remove_server_record -> set_available;
	// set_available -> avail_delay [label="delay 3s"];
	// avail_delay -> "first_in_queue?";
	// 
	// "first_in_queue?" -> set_running [label=yes];
	// set_running -> get_next_port -> handle_requests;
	// "first_in_queue?" -> "dead_entry_in_queue?" [label=no];
	// "dead_entry_in_queue?" -> "server_running?" [label=no];
	// "dead_entry_in_queue?" -> "remove_dead_entries" [label=yes];
	// remove_dead_entries -> "server_running?";
	// 
	// handle_requests -> start_shutdown [label="no traffic\nno running tests"];
	// handle_requests -> shutdown_request;
	// start_shutdown -> shutdown_delay;
	// shutdown_request -> shutdown_delay;
	// shutdown_delay -> exit;
	
        label = "server:launch";
        color=brown;
    }

//     client_start_server -> start_server;
//     handle_requests -> read_write;
//     read_write -> handle_requests;
}

    (string-intersperse 
     (map number->string
	  (u8vector->list
	   (if res res (hostname->ip hostname)))) ".")))

(define (http-transport:run hostn run-id server-id)
  (debug:print 2 "Attempting to start the server ...")








  (let* ((db              #f) ;;        (open-db)) ;; we don't want the server to be opening and closing the db unnecesarily
	 (hostname        (get-host-name))
	 (ipaddrstr       (let ((ipstr (if (string=? "-" hostn)
					   ;; (string-intersperse (map number->string (u8vector->list (hostname->ip hostname))) ".")
					   (server:get-best-guess-address hostname)
					   #f)))
			    (if ipstr ipstr hostn))) ;; hostname))) 
	 (start-port      (open-run-close tasks:server-get-next-port tasks:open-db))






	 (link-tree-path  (config-lookup *configdat* "setup" "linktree")))
    (set! db *inmemdb*)
    (root-path     (if link-tree-path 
		       link-tree-path
		       (current-directory))) ;; WARNING: SECURITY HOLE. FIX ASAP!
    (handle-directory spiffy-directory-listing)
    ;; http-transport:handle-directory) ;; simple-directory-handler)
    ;; Setup the web server and a /ctrl interface

   exn
   (begin
     (print-error-message exn)
     (if (< portnum 9000)
	 (begin 
	   (debug:print 0 "WARNING: failed to start on portnum: " portnum ", trying next port")
	   (thread-sleep! 0.1)

	   ;; get_next_port goes here

	   (http-transport:try-start-server ipaddrstr (+ portnum 1) server-id))
	 (print "ERROR: Tried and tried but could not start the server")))
   ;; any error in following steps will result in a retry
   (set! *server-info* (list ipaddrstr portnum))
   (open-run-close tasks:server-set-interface-port 
		   tasks:open-db 
		   server-id 

  (set! *http-connections-next-cleanup* (+ (current-seconds) 10))
  (mutex-unlock! *http-mutex*))

(define (http-transport:inc-requests-and-prep-to-close-all-connections)
  (mutex-lock! *http-mutex*)
  (set! *http-requests-in-progress* (+ 1 *http-requests-in-progress*)))


;; Send "cmd" with json payload "params" to serverdat and receive result
;;



(define (http-transport:client-api-send-receive run-id serverdat cmd params #!key (numretries 30))


  ;; (let loop ((sdat  serverdat)



  ;;            (tries 10))


  ;;   (if (not sdat) ;; get #f, something went wrong. try starting the server again and reconnecting
  ;;       (begin

  ;;         ;; try to restart the server and then reconnect
  ;;         ;; (hash-table-delete! *runremote* run-id) ;; this should be taken care of by client:setup
  ;;         (thread-sleep! 1)
  ;;         (if (> tries 0)



  ;;             (let ((newsdat (client:setup run-id)))





  ;;       	(set! serverdat newsdat)











  ;;       	(loop newsdat (- tries 1)))








  ;;             (debug:print 0 "ERROR: could not connect to or start a server for run-id " run-id)))))













  ;; (debug:print 0 "serverdat=" serverdat)



  (let* ((fullurl    (if (list? serverdat)
			 (cadddr serverdat) ;; this is the uri for /api
			 (begin
			   (debug:print 0 "FATAL ERROR: http-transport:client-api-send-receive called with no server info")
			   (exit 1))))
	 (res        #f))
    (handle-exceptions
     exn
     #f
     ;; (begin
       ;; TODO: Send this output to a log file so it isn't lost when running as daemon

       ;; (if (> numretries 0)
	;;    ;; on the zeroeth retry do not print the error message - this allows the call to be used as a ping (no junk on output).
	;;    (begin
	;;      (print "ERROR IN http-transport:client-api-send-receive " ((condition-property-accessor 'exn 'message) exn))
	;;      ;; try to restart the server and then reconnect
	;;      ;; (hash-table-delete! *runremote* run-id)
	;;      ;; (client:setup run-id)
	;;      ;; (http-transport:client-api-send-receive run-id serverdat cmd params numretries: (- numretries 1)))
	;;      #f) ;; simply return #f to indicate failure. The caller will need to do the retry.
	;;    #f))
     (begin
       (debug:print-info 11 "fullurl=" fullurl ", cmd=" cmd ", params=" params ", run-id=" run-id "\n")
       ;; set up the http-client here
       (max-retry-attempts 5)
       ;; consider all requests indempotent
       (retry-request? (lambda (request)
			 #t))   ;;  		 (thread-sleep! (/ (if (> numretries 100) 100 numretries) 10))
       ;; (set! numretries (- numretries 1))
       ;;  		 #t))

				  ;;     (set! res dat)
				  ;;     (http-transport:dec-requests-count-and-close-all-connections))
				  ;;   (http-transport:dec-requests-count
				  ;;    (lambda ()
				  ;;      (set! res dat)))))))
	      (time-out     (lambda ()
			      (thread-sleep! 45)
			      #f))
			      ;; (if (not res)
			      ;;     (begin
			      ;;       (debug:print 0 "WARNING: communication with the server timed out.")
			      ;;       (mutex-unlock! *http-mutex*)
			      ;;       (http-transport:client-api-send-receive run-id serverdat cmd params numretries: (- numretries 1))
			      ;;       (if (< numretries 3) ;; on last try just exit
			      ;;   	(begin
			      ;;   	  (debug:print 0 "ERROR: communication with the server timed out. Giving up.")
			      ;;   	  (exit 1)))))))
	      (th1 (make-thread send-recieve "with-input-from-request"))
	      (th2 (make-thread time-out     "time out")))
	 (thread-start! th1)
	 (thread-start! th2)
	 (thread-join! th1)
	 (thread-terminate! th2)
	 (debug:print-info 11 "got res=" res)
	 res)))))

;;
;; connect
;;
(define (http-transport:client-connect run-id iface port)

  (let* ((uri-dat     (make-request method: 'POST uri: (uri-reference (conc "http://" iface ":" port "/ctrl"))))
	 (uri-api-dat (make-request method: 'POST uri: (uri-reference (conc "http://" iface ":" port "/api"))))
	 (serverdat   (list iface port uri-dat uri-api-dat))
	 (login-res   (rmt:login-no-auto-client-setup serverdat run-id)))
    (hash-table-set! *runremote* run-id serverdat) ;; may or may not be good ...

    (if (and (list? login-res)
	     (car login-res))
	(begin
	  (debug:print-info 2 "Logged in and connected to " iface ":" port)
	  (hash-table-set! *runremote* run-id serverdat)
	  serverdat)
	(begin
	  (debug:print-info 0 "ERROR: Failed to login or connect to " iface ":" port)
	  #f))))

;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running server-id)
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown

	 (tdb         (tasks:open-db))
	 (server-timeout (let ((tmo (config-lookup  *configdat* "server" "timeout")))
			   (if (and (string? tmo)
				    (string->number tmo))
			       (* 60 60 (string->number tmo))
			       ;; default to three days
			       (* 3 24 60 60)))))
    ;;
    ;; set_running
    ;;
    (tasks:server-set-state! tdb server-id "running")
    (let loop ((count 0))
      ;; Use this opportunity to sync the inmemdb to db
      (let ((start-time (current-milliseconds))
	    (sync-time  #f)
	    (rem-time   #f))
	(if *inmemdb* (db:sync-touched *inmemdb* force-sync: #t))

      (if (or (not (equal? sdat (list iface port)))
	      (not server-id))
	  (begin 
	    (debug:print-info 0 "interface changed, refreshing iface and port info")
	    (set! iface (car sdat))
	    (set! port  (cadr sdat))))
      






      ;; Transfer *last-db-access* to last-access to use in checking that we are still alive
      (mutex-lock! *heartbeat-mutex*)
      (set! last-access *last-db-access*)
      (mutex-unlock! *heartbeat-mutex*)

      ;; (debug:print 11 "last-access=" last-access ", server-timeout=" server-timeout)
      ;;
      ;; no_traffic
      ;;
      (if (and *server-run*
	       (> (+ last-access server-timeout)
		  (current-seconds)))
	  (begin
	    (debug:print-info 0 "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
	    (loop 0))
	  (begin
	    (debug:print-info 0 "Starting to shutdown the server.")
	    ;; need to delete only *my* server entry (future use)
	    (set! *time-to-exit* #t)
	    (if *inmemdb* (db:sync-touched *inmemdb* force-sync: #t))
	    ;;
	    ;; start_shutdown
	    ;;
	    ( tasks:server-set-state! tdb server-id "shutting-down")
	    (thread-sleep! 5)
	    (debug:print-info 0 "Max cached queries was    " *max-cache-size*)
	    (debug:print-info 0 "Number of cached writes   " *number-of-writes*)
	    (debug:print-info 0 "Average cached write time "
			      (if (eq? *number-of-writes* 0)
				  "n/a (no writes)"

				     *number-non-write-queries*))
			      " ms")
	    (debug:print-info 0 "Server shutdown complete. Exiting")
	    (tasks:server-delete-record! tdb server-id)
	    (exit))))))

;; all routes though here end in exit ...
;;
;; start_server? 
;;
(define (http-transport:launch run-id)
  (set! *run-id*   run-id)






  (if (args:get-arg "-daemonize")
      (daemon:ize))
  (let ((server-id (open-run-close tasks:server-lock-slot tasks:open-db run-id)))
    (if (not server-id)
	(begin
	  ;; since we didn't get the server lock we are going to clean up and bail out
	  (debug:print-info 2 "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
	  (open-run-close tasks:server-delete-records-for-this-pid tasks:open-db))

	(let* ((th2 (make-thread (lambda ()
				   (http-transport:run 
				    (if (args:get-arg "-server")
					(args:get-arg "-server")
					"-")
				    run-id
				    server-id)) "Server run"))
	       (th3 (make-thread (lambda ()
				   (http-transport:keep-running server-id))
				 "Keep running")))
	  ;; Database connection
	  (set! *inmemdb*  (db:setup run-id))
	  (thread-start! th2)
	  (thread-start! th3)
	  (set! *didsomething* #t)
	  (thread-join! th2)


	  (exit)))))

(define (http-transport:server-signal-handler signum)
  (handle-exceptions
   exn
   (debug:print " ... exiting ...")
   (let ((th1 (make-thread (lambda ()
			     (thread-sleep! 1))

  -cleanup-db             : remove any orphan records, vacuum the db
  -update-meta            : update the tests metadata for all tests
  -env2file fname         : write the environment to fname.csh and fname.sh
  -setvars VAR1=val1,VAR2=val2 : Add environment variables to a run NB// these are
                                 overwritten by values set in config files.
  -server -|hostname      : start the server (reduces contention on megatest.db), use
                            - to automatically figure out hostname

  -daemonize              : fork into background and disconnect from stdin/out
  -list-servers           : list the servers 
  -stop-server id         : stop server specified by id (see output of -list-servers), use
                            0 to kill all
  -repl                   : start a repl (useful for extending megatest)
  -load file.scm          : load and run file.scm
  -mark-incompletes       : find and mark incomplete tests

			":variable"
			":value"
			":expected"
			":tol"
			":units"
			;; misc
			"-server"

			"-stop-server"
			"-port"
			"-extract-ods"
			"-pathmod"
			"-env2file"
			"-setvars"
			"-set-state-status"

					      (eq? pid-val 0))
					  (begin
					    (printf "Sending signal/term to ~A\n" pid)
					    (process-signal pid signal/term))))))
		       (process:children #f))
		      (original-exit exit-code)))))







;;======================================================================
;; Misc setup stuff
;;======================================================================

(debug:setup)

(if (args:get-arg "-logging")(set! *logging* #t))

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

(if (args:get-arg "-server")

    ;; Server? Start up here.
    ;;
    (let ((tl        (setup-for-run))


	  (run-id    (and (args:get-arg "-run-id")
			  (string->number (args:get-arg "-run-id")))))

      (if run-id
	  (begin
	    (server:launch run-id)
	    (set! *didsomething* #t))
	  (debug:print 0 "ERROR: server requires run-id be specified with -run-id")))

    ;; Not a server? This section will decide how to communicate
    ;;
    ;;  Setup client for all expect listed here
    (if (null? (lset-intersection 
		     equal?
		     (hash-table-keys args:arg-hash)
		     '("-list-servers"
		       "-stop-server"
		       "-show-cmdinfo"
		       "-list-runs")))
	(if (setup-for-run)
	    (let ((run-id    (and (args:get-arg "-run-id")
				  (string->number (args:get-arg "-run-id")))))
	      ;; (set! *fdb*   (filedb:open-db (conc *toppath* "/db/paths.db")))
	      ;; if not list or kill then start a client (if appropriate)
	      (if (or (args-defined? "-h" "-version" "-gen-megatest-area" "-gen-megatest-test")
		      (eq? (length (hash-table-keys args:arg-hash)) 0))
		  (debug:print-info 1 "Server connection not needed")






















		  (begin
		    (if run-id 
			(client:launch run-id) 

			(client:launch 0)      ;; without run-id we'll start a server for "0"

			)))))))



;; MAY STILL NEED THIS
;;		       (set! *megatest-db* (make-dbr:dbstruct path: *toppath* local: #t))))))))))

(if (or (args:get-arg "-list-servers")
	(args:get-arg "-stop-server"))
    (let ((tl (setup-for-run)))
      (if tl 
	  (let* ((servers (open-run-close tasks:get-all-servers tasks:open-db))
		 (fmtstr  "~5a~12a~8a~20a~24a~10a~10a~10a~10a\n")

;; Get test paths matching target, runname, and testpatt
(if (or (args:get-arg "-test-files")(args:get-arg "-test-paths"))
    ;; if we are in a test use the MT_CMDINFO data
    (if (getenv "MT_CMDINFO")
	(let* ((startingdir (current-directory))
	       (cmdinfo   (read (open-input-string (base64:base64-decode (getenv "MT_CMDINFO")))))
	       ;; (runremote (assoc/default 'runremote cmdinfo))

	       (testpath  (assoc/default 'testpath  cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (state     (args:get-arg ":state"))
	       (status    (args:get-arg ":status"))
	       (target    (args:get-arg "-target"))
	       (toppath   (assoc/default 'toppath   cmdinfo)))
	  (change-directory toppath)
	  ;; (set! *runremote* runremote)

	  (if (not target)
	      (begin
		(debug:print 0 "ERROR: -target is required.")
		(exit 1)))
	  (if (not (setup-for-run))
	      (begin
		(debug:print 0 "Failed to setup, giving up on -test-paths or -test-files, exiting")

;; Archive tests matching target, runname, and testpatt
(if (args:get-arg "-archive")
    ;; if we are in a test use the MT_CMDINFO data
    (if (getenv "MT_CMDINFO")
	(let* ((startingdir (current-directory))
	       (cmdinfo   (read (open-input-string (base64:base64-decode (getenv "MT_CMDINFO")))))
	       ;; (runremote (assoc/default 'runremote cmdinfo))

	       (testpath  (assoc/default 'testpath  cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (state     (args:get-arg ":state"))
	       (status    (args:get-arg ":status"))
	       (target    (args:get-arg "-target")))
	  (change-directory testpath)
	  ;; (set! *runremote* runremote)

	  (if (not target)
	      (begin
		(debug:print 0 "ERROR: -target is required.")
		(exit 1)))
	  (if (not (setup-for-run))
	      (begin
		(debug:print 0 "Failed to setup, giving up on -archive, exiting")

(define (megatest:step step state status logfile msg)
  (if (not (getenv "MT_CMDINFO"))
      (begin
	(debug:print 0 "ERROR: MT_CMDINFO env var not set, -step must be called *inside* a megatest invoked environment!")
	(exit 5))
      (let* ((cmdinfo   (read (open-input-string (base64:base64-decode (getenv "MT_CMDINFO")))))
	     ;; (runremote (assoc/default 'runremote cmdinfo))

	     (testpath  (assoc/default 'testpath  cmdinfo))
	     (test-name (assoc/default 'test-name cmdinfo))
	     (runscript (assoc/default 'runscript cmdinfo))
	     (db-host   (assoc/default 'db-host   cmdinfo))
	     (run-id    (assoc/default 'run-id    cmdinfo))
	     (test-id   (assoc/default 'test-id   cmdinfo))
	     (itemdat   (assoc/default 'itemdat   cmdinfo))
	     (work-area (assoc/default 'work-area cmdinfo))
	     (db        #f))
	(change-directory testpath)
	;; (set! *runremote* runremote)


	(if (not (setup-for-run))
	    (begin
	      (debug:print 0 "Failed to setup, exiting")
	      (exit 1)))
	(if (and state status)
	    (rmt:teststep-set-status! run-id test-id step state status msg logfile)
	    (begin

    (if (not (getenv "MT_CMDINFO"))
	(begin
	  (debug:print 0 "ERROR: MT_CMDINFO env var not set, commands -test-status, -runstep and -setlog must be called *inside* a megatest environment!")
	  (exit 5))
	(let* ((startingdir (current-directory))
	       (cmdinfo   (read (open-input-string (base64:base64-decode (getenv "MT_CMDINFO")))))
	       ;; (runremote (assoc/default 'runremote cmdinfo))

	       (testpath  (assoc/default 'testpath  cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))
	       (db        #f) ;; (open-db))
	       (state     (args:get-arg ":state"))
	       (status    (args:get-arg ":status")))
	  ;; (set! *runremote* runremote)

	  (if (not (setup-for-run))
	      (begin
		(debug:print 0 "Failed to setup, exiting")
		(exit 1)))

	  (if (args:get-arg "-runstep")(debug:print-info 1 "Running -runstep, first change to directory " work-area))
	  (change-directory work-area)

;;  S U P P O R T   F U N C T I O N S
;;======================================================================

;; cmd is a symbol
;; vars is a json string encoding the parameters for the call
;;
(define (rmt:send-receive cmd run-id params)







  (let* ((connection-info (hash-table-ref/default *runremote* run-id #f))
	 (jparams         (db:obj->string params))
	 (res (http-transport:client-api-send-receive run-id connection-info cmd jparams)))
    (if res
	(db:string->obj res) ;; (rmt:json-str->dat res)
	(let ((new-connection-info (client:setup run-id)))
	  (debug:print 0 "WARNING: Communication failed, trying call to http-transport:client-api-send-receive again.")
	  (rmt:send-receive-no-auto-client-setup connection-info cmd run-id params)))))

(define (rmt:send-receive-no-auto-client-setup connection-info cmd run-id params)
  (let* ((jparams         (db:obj->string params)) ;; (rmt:dat->json-str params))
	 (res (http-transport:client-api-send-receive run-id connection-info cmd jparams numretries: 3)))
    (if res
	(db:string->obj res) ;; (rmt:json-str->dat res)
	(let ((connection-info (client:setup run-id)))
	  ;; something went wrong, try setting up the client again and then resend
	  (debug:print 0 "WARNING: Communication failed, trying call to http-transport:client-api-send-receive again.")
	  (rmt:send-receive-no-auto-client-setup connection-info cmd run-id params)))))

;; Wrap json library for strings (why the ports crap in the first place?)
(define (rmt:dat->json-str dat)
  (with-output-to-string 
    (lambda ()
      (json-write dat))))

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

;;======================================================================
;;  M I S C
;;======================================================================

(define (rmt:login run-id)
  (rmt:send-receive 'login run-id (list *toppath* megatest-version run-id *my-client-signature*)))

;; This login does no retries under the hood - it acts a bit like a ping.
;;
(define (rmt:login-no-auto-client-setup connection-info run-id)
  (rmt:send-receive-no-auto-client-setup connection-info 'login run-id (list *toppath* megatest-version run-id *my-client-signature*)))
  
(define (rmt:kill-server run-id)
  (rmt:send-receive 'kill-server run-id (list run-id)))

;; hand off a call to one of the db:queries statements
;; added run-id to make looking up the correct db possible 
;;
(define (rmt:general-call stmtname run-id . params)

;; S E R V E R
;;======================================================================

;; Call this to start the actual server
;;

;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id)
  ;; (if (server:check-if-running run-id)

  ;; a server is already running

  ;; (exit)


  (http-transport:launch run-id))

;; (define (server:launch-no-exit run-id)
;;   (if (server:check-if-running run-id)
;;       #t ;; if running
;;       (http-transport:launch run-id)))




;;======================================================================
;; Q U E U E   M A N A G E M E N T
;;======================================================================

;; We don't want to flush the queue if it was just flushed
(define *server:last-write-flush* (current-milliseconds))

					  (argv)))))))


;; When using zmq this would send the message back (two step process)
;; with spiffy or rpc this simply returns the return data to be returned
;; 
(define (server:reply return-addr query-sig success/fail result)





  (db:obj->string (vector success/fail query-sig result)))








(define (server:try-running run-id)






  (let ((cmdln (conc (if (getenv "MT_MEGATEST") (getenv "MT_MEGATEST") "megatest")
		     " -server - -run-id " run-id " &> " *toppath* "/db/" run-id ".log &")))
    (debug:print 0 "INFO: Starting server (" cmdln ") as none running ...")



    (push-directory *toppath*)
    (system cmdln)
    (pop-directory)))




(define (server:check-if-running run-id)


  (let loop ((server (open-run-close tasks:get-server tasks:open-db run-id))
	     (trycount 0))
    (thread-sleep! 2)
    (if server
	;; note: client:start will set *runremote*. this needs to be changed
	;;       also, client:start will login to the server, also need to change that.
	;;
	;; client:start returns #t if login was successful.
	;;
	(let ((res (client:start run-id server)))
	  ;; if the server didn't respond we must remove the record
	  (if res
	      res
	      (begin
		(open-run-close tasks:server-force-clean-running-records-for-run-id tasks:open-db run-id)
		res)))
	#f)))

(define (tasks:hostinfo-get-port        vec)    (vector-ref  vec 2))
(define (tasks:hostinfo-get-pubport     vec)    (vector-ref  vec 3))
(define (tasks:hostinfo-get-transport   vec)    (vector-ref  vec 4))
(define (tasks:hostinfo-get-pid         vec)    (vector-ref  vec 5))
(define (tasks:hostinfo-get-hostname    vec)    (vector-ref  vec 6))

(define (tasks:server-lock-slot mdb run-id)


  (tasks:server-clean-out-old-records-for-run-id mdb run-id)
  (if (< (tasks:num-in-available-state mdb run-id) 4)
      (begin 
	(tasks:server-set-available mdb run-id)
	(thread-sleep! 2) ;; Try removing this. It may not be needed.
	(tasks:server-am-i-the-server? mdb run-id))
      #f))      
	
;; register that this server may come online (first to register goes though with the process)
(define (tasks:server-set-available mdb run-id)
  (sqlite3:execute 
   mdb 
   "INSERT INTO servers (pid,hostname,port,pubport,start_time,      priority,state,mt_version,heartbeat,   interface,transport,run_id)
                   VALUES(?, ?,       ?,   ?, strftime('%s','now'), ?,       ?,    ?,-1,?,        ?,        ?);"
   (current-process-id)       ;; pid
   (get-host-name)            ;; hostname
   -1                         ;; port
   -1                         ;; pubport
   (random 1000)              ;; priority (used a tiebreaker on get-available)
   "available"                ;; state
   (common:version-signature) ;; mt_version
   -1                         ;; interface
   "http"                     ;; transport
   run-id
   ))

(define (tasks:num-in-available-state mdb run-id)
  (let ((res 0))
    (sqlite3:for-each-row
     (lambda (num-in-queue)
       (set! res num-in-queue))
     mdb
     "SELECT count(id) FROM servers WHERE run_id=?;"
     run-id)
    res))

(define (tasks:server-clean-out-old-records-for-run-id mdb run-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state in ('available','shutting-down') AND (strftime('%s','now') - start_time) > 10 AND run_id=?;" run-id))

(define (tasks:server-force-clean-running-records-for-run-id mdb run-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state = 'running' AND run_id=?;" run-id))

(define (tasks:server-force-clean-run-record mdb run-id iface port)
  (sqlite3:execute mdb "DELETE FROM servers WHERE state = 'running' AND run_id=? AND interface=? AND port=?;"
		   run-id iface port))

(define (tasks:server-set-state! mdb server-id state)
  (sqlite3:execute mdb "UPDATE servers SET state=? WHERE id=?;" state server-id))

(define (tasks:server-delete-record! mdb server-id)
  (sqlite3:execute mdb "DELETE FROM servers WHERE id=?;" server-id))

(define (tasks:server-delete-records-for-this-pid mdb)
  (sqlite3:execute mdb "DELETE FROM servers WHERE hostname=? AND pid=?;" (get-host-name) (current-process-id)))

(define (tasks:server-set-interface-port mdb server-id interface port)
  (sqlite3:execute mdb "UPDATE servers SET interface=?,port=? WHERE id=?;" interface port server-id))

(define (tasks:server-get-next-port mdb)
  (let ((res         #f)
	(port-param  (if (and (args:get-arg "-port")
			      (string->number (args:get-arg "-port")))
			 (string->number (args:get-arg "-port"))
			 #f))
	(config-port (if (and (config-lookup  *configdat* "server" "port")
			      (string->number (config-lookup  *configdat* "server" "port")))
			 (string->number (config-lookup  *configdat* "server" "port"))
			 #f)))
    (sqlite3:for-each-row
     (lambda (port)
       (set! res (+ port 1))) ;; set to next
     mdb
     "SELECT max(port) FROM servers;")
    (cond
     ((and port-param res)   (if (> res port-param) res port-param))
     (port-param             port-param)
     ((and config-port res)  (if (> res config-port) res config-port))
     (config-port            config-port)
     ((and res (> res 8080)) res)
     (else                   (+ 5000 (random 1001))))))

(define (tasks:server-am-i-the-server? mdb run-id)
  (let* ((all    (tasks:server-get-servers-vying-for-run-id mdb run-id))
	 (first  (if (null? all)
		     (begin (debug:print 0 "ERROR: no servers listed, should be at least one by now.") 
			    (sqlite3:finalize! mdb)
			    (exit 1))
		     (car (db:get-rows all))))

     (lambda (a . b)
       (set! res (cons (apply vector a b) res)))
     mdb
     (conc "SELECT " selstr " FROM servers WHERE run_id=? ORDER BY start_time DESC;")
     run-id)
    (vector header res)))























(define (tasks:get-server mdb run-id)
  (let ((res  #f)
	(best #f))
    (sqlite3:for-each-row
     (lambda (id interface port pubport transport pid hostname)
       (set! res (vector id interface port pubport transport pid hostname)))
     mdb
     ;; removed:
     ;; strftime('%s','now')-heartbeat < 10 AND mt_version = ?
     "SELECT id,interface,port,pubport,transport,pid,hostname FROM servers

          WHERE run_id=? AND state='running'
          ORDER BY start_time DESC LIMIT 1;" run-id) ;; (common:version-signature) run-id)
    res))













(define (tasks:get-all-servers mdb)
  (let ((res '()))
    (sqlite3:for-each-row
     (lambda (id pid hostname interface port pubport start-time priority state mt-version last-update transport)
       (set! res (cons (vector id pid hostname interface port pubport start-time priority state mt-version last-update transport) res)))
     mdb
     "SELECT id,pid,hostname,interface,port,pubport,start_time,priority,state,mt_version,strftime('%s','now')-heartbeat AS last_update,transport FROM servers ORDER BY start_time DESC;")
    res))

(define (tasks:kill-server status hostname port pid)
  (debug:print-info 1 "Removing defunct server record for " hostname ":" port)
  (if port
      (open-run-close tasks:server-deregister tasks:open-db hostname port: port)
      (open-run-close tasks:server-deregister tasks:open-db hostname pid:  pid))
  (if status ;; #t means alive
      (begin
	(if (equal? hostname (get-host-name))
	    (handle-exceptions
	     exn
	     (debug:print-info 0 "server may or may not be dead, check for megatest -server running as pid " pid "\n"
			       "  EXCEPTION: " ((condition-property-accessor 'exn 'message) exn))
	     (debug:print 1 "Sending signal/term to " pid " on " hostname)
	     (process-signal pid signal/term)
	     (thread-sleep! 5) ;; give it five seconds to die peacefully then do a brutal kill
	     ;;(process-signal pid signal/kill)
	     ) ;; local machine, send sig term
	    (begin
	      ;;(debug:print-info 1 "Stopping remote servers not yet supported."))))
	      (debug:print-info 1 "Telling alive server on " hostname ":" port " to commit servercide")
	      (let ((serverdat (list hostname port)))

		(http-transport:client-connect hostname port)

	      	(cdb:kill-server serverdat pid)))))    ;; remote machine, try telling server to commit suicide
      (begin
	(if status 
	    (if (equal? hostname (get-host-name))
		(begin
		  (debug:print-info 1 "Sending signal/term to " pid " on " hostname)
		  (process-signal pid signal/term)  ;; local machine, send sig term

[validvalues]
state start end 0 1 - 2
status pass fail n/a 0 1 running - 2

# These are set before all tests, override them 
# in the testconfig [pre-launch-env-overrides] section
[env-override]
# This variable is honored by the loadrunner script. The value is in percent
# a value of 200 will stop new jobs from starting.
MAX_ALLOWED_LOAD 200

# MT_XTERM_CMD overrides the terminal command
# MT_XTERM_CMD xterm -bg lightgreen -fg black

SPECIAL_ENV_VARS overide them here - should be seen at launch and in the runs
TESTVAR [system echo $PWD]
DEADVAR [system ls]
VARWITHDOLLAR $HOME/.zshrc

#!/mfs/pkgs/chicken/4.8.0.5/bin/csi -nbq

(use regex srfi-69 srfi-13)

(define targs #f) 
(define files (cddddr (argv)))

(let ((targdat (cadddr (argv))))
  (if (equal? targdat "-")
      (set! targs files)
      (set! targs (string-split targdat ","))))

(define filedat-defns (make-hash-table))
(define filedat-usages (make-hash-table))

(define defn-rx (regexp "^\\s*\\(define\\s+\\(([^\\s\\)]+).*"))
(define all-regexs (make-hash-table))

(define all-fns '())

(define (print-err . data)
  (with-output-to-port (current-error-port)
    (lambda ()
      (apply print data))))

(print-err "Making graph for files: " (string-intersperse targs ", "))
(print-err "Looking at files: " (string-intersperse files ", "))

;; Gather the functions
;;
(for-each 
 (lambda (fname)
   (print-err "Processing file " fname)
   (with-input-from-file fname
     (lambda ()
       (let loop ((inl (read-line)))
	 (if (not (eof-object? inl))
	     (let ((match (string-match defn-rx inl)))
	       (if match 
		   (let ((fnname (cadr match)))
		     ;; (print "   " fnname)
		     (set! all-fns (cons fnname all-fns))
		     (hash-table-set! 
		      filedat-defns 
		      fname
		      (cons fnname (hash-table-ref/default filedat-defns fname '())))
		     ))
	       (loop (read-line))))))))
 files)

;; fill up the regex hash
(print-err "Make the huge regex hash")
(for-each
 (lambda (fnname)
   (hash-table-set! all-regexs fnname (regexp (conc "^(|.*[^a-zA-Z]+)" fnname "([^a-zA-Z]+|)$"))))
 (cons "toplevel" all-fns))

(define breadcrumbs (make-hash-table))

(define (have-function inl)
  (let loop ((hed (car all-fns))
	     (tal (cdr all-fns)))
    (if (string-contains inl hed)
	#t
	(if (null? tal)
	    #f
	    (loop (car tal)(cdr tal))))))

(define (look-for-all-calls inl fnname)
  (if (have-function inl) ;; (string-search have-function-rx inl)
      (let loop ((hed (car all-fns))
		 (tal (cdr all-fns))
		 (res '()))
	(let ((match (string-match (hash-table-ref all-regexs hed) inl)))
	  (if match
	      (let ((newres (cons hed res)))
		(if (null? tal)
		    newres
		    (loop (car tal)
			  (cdr tal)
			  newres)))
	      (if (null? tal)
		  res
		  (loop (car tal)(cdr tal) res)))))
      '()))
  
;; Gather the usages
(print "digraph G {")
(define curr-cluster-num 0)
(define function-calls '())

(for-each
 (lambda (fname)
   (let ((last-func #f))
     (print-err "Processing file " fname)
     (print "subgraph cluster_" curr-cluster-num " {")
     (set! curr-cluster-num (+ curr-cluster-num 1))
     (with-input-from-file fname
       (lambda ()
	 (with-output-to-port (current-error-port)
	   (lambda ()
	     (print "Analyzing file " fname)))
	 (print "label=\"" fname "\";")
	 (let loop ((inl    (read-line))
		    (fnname "toplevel")
		    (allcalls '()))
	   (if (eof-object? inl)
	       (begin
		 (set! function-calls (cons (list fnname allcalls) function-calls))
		 (for-each 
		  (lambda (call-name)
		    (hash-table-set! breadcrumbs call-name #t))
		  allcalls)
		 (print-err "function: " fnname " allcalls: " allcalls))
	       (let ((match (string-match defn-rx inl)))
		 (if match
		     (let ((func-name (cadr match)))
		       (if last-func
			   (print "\"" func-name "\" -> \"" last-func "\";")
			   (print "\"" func-name "\";"))
		       (set! last-func func-name)
		       (hash-table-set! breadcrumbs func-name #t)
		       (loop (read-line)
			     func-name
			     allcalls))
		     (let ((calls (look-for-all-calls inl fnname)))
		       (loop (read-line) fnname (append allcalls calls)))))))))
     (print "}")))
 targs)

(print-err "breadcrumbs: " (hash-table-keys breadcrumbs))
(print-err "function-calls: " function-calls)

(for-each 
 (lambda (function-call)
   (print-err "function-call: " function-call)
   (let ((fnname (car function-call))
	 (calls  (cadr function-call)))
     (for-each
      (lambda (callname)
	(print (if (hash-table-ref/default breadcrumbs callname #f) "" "// ")
	       "\"" fnname "\" -> \"" callname "\";"))
      calls)))
 function-calls)

(print "}")

(exit)