Megatest

;; Copyright 2006-2017, 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.

;;======================================================================
;; launch a task - this runs on the originating host, tests themselves
;;
;;======================================================================

(use regex regex-case base64 sqlite3 srfi-18 directory-utils posix-extras z3 call-with-environment-variables csv)
(use typed-records pathname-expand matchable)

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

(declare (unit launch))
(declare (uses common))
(declare (uses configf))

	       (csvt (let-values (((fmt-cell fmt-record fmt-csv) (make-format ",")))
		       (fmt-csv (map list->csv-record csvr))))
	       (status (configf:lookup dat "final" "exit-status"))
	       (msg     (configf:lookup dat "final" "message")))
          (if csvt  ;; this if blocked stack dump caused by .dat file from logpro being 0-byte.  fixed by upgrading logpro
              (rmt:csv->test-data run-id test-id csvt)
	      (debug:print 0 *default-log-port* "ERROR: no csvdat exists for run-id: " run-id " test-id: " test-id " stepname: " stepname ", check that logpro version is 1.15 or newer"))
	  ;;  (debug:print-info 13 *default-log-port* "Error: run-id/test-id/stepname="run-id"/"test-id"/"stepname" => bad csvr="csvr)
	  ;;  )
	  (cond
	   ((equal? status "PASS") "PASS") ;; skip the message part if status is pass
	   (status (conc (configf:lookup dat "final" "exit-status") ": " (if msg msg "no message")))
	   (else #f)))
	#f)))

		    (if (steprun-good? logpro-used (launch:einf-exit-code exit-info))
			(if (not (null? tal))
			    (loop (car tal) (cdr tal) stepname))
			(debug:print 4 *default-log-port* "WARNING: step " (car ezstep) " failed. Stopping")))
		  (debug:print 4 *default-log-port* "WARNING: a prior step failed, stopping at " ezstep)))))))

(define (launch:monitor-job run-id test-id item-path fullrunscript ezsteps test-name tconfigreg exit-info m work-area runtlim misc-flags)
  (let* ((update-period (string->number (or (configf:lookup *configdat* "setup" "test-stats-update-period") "30")))
         (start-seconds (current-seconds))
	 (calc-minutes  (lambda ()
			  (inexact->exact 
			   (round 
			    (- 
			     (current-seconds) 
			     start-seconds)))))
	 (kill-tries 0))
    ;; (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area)
    ;; (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
    (tests:set-full-meta-info #f test-id run-id (calc-minutes) work-area 10)
    (let loop ((minutes   (calc-minutes))
	       (cpu-load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
	       (disk-free (get-df (current-directory)))
               (last-sync (current-seconds)))
      (let* ((over-time     (> (current-seconds) (+ last-sync update-period)))
             (new-cpu-load  (let* ((load  (alist-ref 'adj-core-load (common:get-normalized-cpu-load #f)))
                                   (delta (abs (- load cpu-load))))
                              (if (> delta 0.1) ;; don't bother updating with small changes
                                  load
                                  #f)))
             (new-disk-free (let* ((df    (if over-time ;; only get df every 30 seconds
                                              (get-df (current-directory))
                                              disk-free))
                                   (delta (abs (- df disk-free))))
                              (if (and (> df 0)
                                       (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                  df
                                  #f)))
             (do-sync       (or new-cpu-load new-disk-free over-time)))
        (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
	(set! kill-job? (or (test-get-kill-request run-id test-id) ;; run-id test-name itemdat))
			    (and runtlim (let* ((run-seconds   (- (current-seconds) start-seconds))
						(time-exceeded (> run-seconds runtlim)))
					   (if time-exceeded
					       (begin
						 (debug:print-info 0 *default-log-port* "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim)
						 #t)
					       #f)))))
        (if do-sync
            (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f))
	(if kill-job? 
	    (begin
	      (mutex-lock! m)
	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
	      ;;       between tries?
	      (let* ((pid1 (launch:einf-pid exit-info)) ;; (vector-ref exit-info 0))

	      (mutex-unlock! m)
	      ;; no point in sticking around. Exit now.
	      (exit)))
	(if (hash-table-ref/default misc-flags 'keep-going #f)
	    (begin
	      (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
	      (if (hash-table-ref/default misc-flags 'keep-going #f)  ;; keep originals for cpu-load and disk-free unless they change more than the allowed delta
		  (loop (calc-minutes)
                        (or new-cpu-load cpu-load)
                        (or new-disk-free disk-free)
                        (if do-sync (current-seconds) last-sync)))))))
    (tests:update-central-meta-info run-id test-id (get-cpu-load) (get-df (current-directory))(calc-minutes) #f #f))) ;; NOTE: Checking twice for keep-going is intentional


(define (launch:execute encoded-cmd)
  (let* ((cmdinfo    (common:read-encoded-string encoded-cmd))
	 (tconfigreg #f))
    (setenv "MT_CMDINFO" encoded-cmd)
    ;;(bb-check-path msg: "launch:execute incoming")
    (if (list? cmdinfo) ;; ((testpath /tmp/mrwellan/jazzmind/src/example_run/tests/sqlitespeed)
	;; (test-name sqlitespeed) (runscript runscript.rb) (db-host localhost) (run-id 1))
	(let* ((testpath  (assoc/default 'testpath  cmdinfo))  ;; testpath is the test spec area
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))  ;; work-area is the test run area
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (ezsteps   (assoc/default 'ezsteps   cmdinfo))
	       ;; (runremote (assoc/default 'runremote cmdinfo))
	       ;; (transport (assoc/default 'transport cmdinfo))  ;; not used
	       ;; (serverinf (assoc/default 'serverinf cmdinfo))
	       ;; (port      (assoc/default 'port      cmdinfo))
	       (serverurl (assoc/default 'serverurl cmdinfo))
	       (homehost  (assoc/default 'homehost  cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (target    (assoc/default 'target    cmdinfo))
	       (areaname  (assoc/default 'areaname  cmdinfo))
	       (itemdat   (assoc/default 'itemdat   cmdinfo))
	       (env-ovrd  (assoc/default 'env-ovrd  cmdinfo))
	       (set-vars  (assoc/default 'set-vars  cmdinfo)) ;; pre-overrides from -setvar
	       (runname   (assoc/default 'runname   cmdinfo))
	       (megatest  (assoc/default 'megatest  cmdinfo))
	       (runtlim   (assoc/default 'runtlim   cmdinfo))
	       (contour   (assoc/default 'contour   cmdinfo))

                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
	       ) ;; (rollup-status 0)

	  (if contour (setenv "MT_CONTOUR" contour))
	  
	  ;; immediated set some key variables from CMDINFO data, yes, these will be set again below ...
	  ;;
	  (setenv "MT_TESTSUITENAME" areaname)
	  (setenv "MT_RUN_AREA_HOME" top-path)
	  (set! *toppath* top-path)
	  (setenv "MT_TEST_RUN_DIR"  work-area)

	  ;; On NFS it can be slow and unreliable to get needed startup information.
	  ;;  i. Check if we are on the homehost, if so, proceed
	  ;; ii. Check if host and port passed in via CMDINFO are valid and if
	  ;;     possible use them.
	  (let ((bestadrs (server:get-best-guess-address (get-host-name)))
		(needcare #f))
	    (if (equal? homehost bestadrs) ;; we are likely on the homehost
		(debug:print-info 0 *default-log-port* "test " test-name " appears to be running on the homehost " homehost)
		(let ((host-port (if serverurl (string-split serverurl ":") #f)))
		  (if (not *runremote*)(set! *runremote* (make-remote))) ;; init *runremote*
		  (if (string? homehost)
		      (if (and host-port
			       (> (length host-port) 1))
			  (let* ((host      (car host-port))
                                 (port      (cadr host-port))
                                 (start-res (http-transport:client-connect host port))
                                 (ping-res  (rmt:login-no-auto-client-setup start-res)))
			    (if (and start-res
				     ping-res)
				;; (begin ;; let ((url  (http-transport:server-dat-make-url start-res)))
				(begin
				  (remote-conndat-set! *runremote* start-res)
				  ;; (remote-server-url-set! *runremote* url)
				  ;; (if (server:ping url)
				  (debug:print-info 0 *default-log-port* "connected to " host ":" port " using CMDINFO data."))
				(begin
				  (debug:print-info 0 *default-log-port* "have CMDINFO data but failed to connect to " host ":" port)
				  (set! *runremote* #f))
				  ;; (remote-conndat-set! *runremote* #f))
				))
			  (begin
			    (set! *runremote* #f)
			    (debug:print-info 0 *default-log-port* (if host-port
								       (conc "received invalid host-port information " host-port)
								       "no host-port information received"))
			    ;; potential for bad situation if simultaneous starting of hundreds of jobs on servers, set needcare.
			    (set! needcare #t)))
		      (begin
			(set! *runremote* #f)
			(debug:print-info 0 *default-log-port* "received no homehost information. Please report this to support as it should not happen.")
			(set! needcare #t)))))
	    (if needcare  ;; due to very slow NFS we will do a brute force mkdir to ensure that the directory inode it truly available on this host
		(let ((logdir (conc top-path "/logs"))) ;; we'll try to create this directory
		  (handle-exceptions
		      exn
		      (debug:print 0 *default-log-port* "Failed to create directory " logdir " expect problems, message: " ((condition-property-accessor 'exn 'message) exn))
		    (create-directory logdir #t)))))
		  
	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (file-exists? top-path)
		    (> count 10))
		(change-directory top-path)
		(begin
		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " top-path " not found")

	    (set-signal-handler! signal/term sighand)
	    ) ;; (set-signal-handler! signal/stop sighand)
	  
	  ;; Do not run the test if it is REMOVING, RUNNING, KILLREQ or REMOTEHOSTSTART,
	  ;; Mark the test as REMOTEHOSTSTART *IMMEDIATELY*
	  ;;
	  (let* ((test-info (rmt:get-test-info-by-id run-id test-id))
		 (test-host (if test-info
				(db:test-get-host        test-info)
				(begin
				  (debug:print 0 *default-log-port* "ERROR: failed to find a record for test-id " test-id ", exiting.")
				  (exit))))
		 (test-pid  (db:test-get-process_id  test-info)))
	    (cond
	     ((member (db:test-get-state test-info) '("INCOMPLETE" "KILLED" "UNKNOWN" "KILLREQ" "STUCK")) ;; prior run of this test didn't complete, go ahead and try to rerun
	      (debug:print 0 *default-log-port* "INFO: test is INCOMPLETE or KILLED, treat this execute call as a rerun request")
	      ;; (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
	      (rmt:test-set-state-status run-id test-id "REMOTEHOSTSTART" "n/a" #f)
	      ) ;; prime it for running

	      (debug:print-error 0 *default-log-port* "test state is " (db:test-get-state test-info) ", cannot proceed")
	      (exit))))
	  
	  (debug:print 2 *default-log-port* "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
	  (set! keys       (rmt:get-keys))
	  ;; (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals) ;; these may be needed by the launching process
	  ;; one of these is defunct/redundant ...
	  (if (not (launch:setup force-reread: #t))
	      (begin
		(debug:print 0 *default-log-port* "Failed to setup, exiting") 
		;; (sqlite3:finalize! db)
		;; (sqlite3:finalize! tdb)
		(exit 1)))
	  (change-directory *toppath*) 

	  ;; NOTE: Current order is to process runconfigs *before* setting the MT_ vars. This 
	  ;;       seems non-ideal but could well break stuff
	  ;;    BUG? BUG? BUG?
	  
	  (let ((rconfig (full-runconfigs-read)) ;; (read-config (conc  *toppath* "/runconfigs.config") #f #t sections: (list "default" target))))
		(wconfig (read-config "waivers.config" #f #t sections: `( "default" ,target )))) ;; read the waivers config if it exists
	    ;; (setup-env-defaults (conc *toppath* "/runconfigs.config") run-id (make-hash-table) keyvals target)
	    ;; (set-run-config-vars run-id keyvals target) ;; (db:get-target db run-id))
	    ;; Now have runconfigs data loaded, set environment vars
	    (for-each (lambda (section)
			(for-each (lambda (varval)
				    (let ((var (car varval))
					  (val (cadr varval)))
				      (if (and (string? var)(string? val))
					  (begin
					    (setenv var (config:eval-string-in-environment val))) ;; val)
					  (debug:print-error 0 *default-log-port* "bad variable spec, " var "=" val))))
				  (configf:get-section rconfig section)))
		      (list "default" target)))
          ;;(bb-check-path msg: "launch:execute post block 1")

	  ;; NFS might not have propagated the directory meta data to the run host - give it time if needed
	  (let loop ((count 0))
	    (if (or (file-exists? work-area)
		    (> count 10))
		(change-directory work-area)
		(begin
		  (debug:print 0 *default-log-port* "INFO: Not starting job yet - directory " work-area " not found")
		  (thread-sleep! 10)
		  (loop (+ count 1)))))
          ;;(bb-check-path msg: "launch:execute post block 1.5")
	  ;; (change-directory work-area) 
	  (set! keyvals    (keys:target->keyval keys target))
	  ;; apply pre-overrides before other variables. The pre-override vars must not
	  ;; clobbers things from the official sources such as megatest.config and runconfigs.config
	  (if (string? set-vars)
	      (let ((varpairs (string-split set-vars ",")))
		(debug:print 4 *default-log-port* "varpairs: " varpairs)
		(map (lambda (varpair)
		       (let ((varval (string-split varpair "=")))
			 (if (eq? (length varval) 2)
			     (let ((var (car varval))
				   (val (cadr varval)))
			       (debug:print 1 *default-log-port* "Adding pre-var/val " var " = " val " to the environment")
			       (setenv var val)))))
		     varpairs)))
          ;;(bb-check-path msg: "launch:execute post block 2")
	  (for-each
	   (lambda (varval)
	     (let ((var (car varval))
		   (val (cadr varval)))
	       (if val
		   (setenv var val)
		   (begin
		     (debug:print-error 0 *default-log-port* "required variable " var " does not have a valid value. Exiting")
		     (exit)))))
	     (list 
	      (list  "MT_TEST_RUN_DIR" work-area)
	      (list  "MT_TEST_NAME" test-name)
	      (list  "MT_ITEM_INFO" (conc itemdat))
	      (list  "MT_ITEMPATH"  item-path)
	      (list  "MT_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (common:get-linktree)) ;; (configf:lookup *configdat* "setup" "linktree"))
	      (list  "MT_TESTSUITENAME" (common:get-testsuite-name))))
          ;;(bb-check-path msg: "launch:execute post block 3")

	  (if mt-bindir-path (setenv "PATH" (conc (getenv "PATH") ":" mt-bindir-path)))
          ;;(bb-check-path msg: "launch:execute post block 4")
	  ;; (change-directory top-path)
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

	  
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
          ;;(bb-check-path msg: "launch:execute post block 41")
	  (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals)
          ;;(bb-check-path msg: "launch:execute post block 42")
	  (set-item-env-vars itemdat)
          ;;(bb-check-path msg: "launch:execute post block 43")
	  (save-environment-as-files "megatest")
          ;;(bb-check-path msg: "launch:execute post block 44")
	  ;; open-run-close not needed for test-set-meta-info
	  ;; (tests:set-full-meta-info #f test-id run-id 0 work-area)
	  ;; (tests:set-full-meta-info test-id run-id 0 work-area)
	  (tests:set-full-meta-info #f test-id run-id 0 work-area 10)

	  ;; (thread-sleep! 0.3) ;; NFS slowness has caused grief here

		  (tests:summarize-items run-id test-id test-name #f))
	      (tests:summarize-test run-id test-id)  ;; don't force - just update if no
	      (rmt:update-run-stats run-id (rmt:get-raw-run-stats run-id)))
	    (mutex-unlock! m)
	    (debug:print 2 *default-log-port* "Output from running " fullrunscript ", pid " (launch:einf-pid exit-info) " in work area " 
			 work-area ":\n====\n exit code " (launch:einf-exit-code exit-info) "\n" "====\n")
	    (if (not (launch:einf-exit-status exit-info))
		(exit 4))))
        )))

(define (launch:cache-config)
  ;; if we have a linktree and -runtests and -target and the directory exists dump the config
  ;; to megatest-(current-seconds).cfg and symlink it to megatest.cfg
  (if (and *configdat* 
	   (or (args:get-arg "-run")
	       (args:get-arg "-runtests")
	       (args:get-arg "-execute")))
      (let* ((linktree (common:get-linktree)) ;; (get-environment-variable "MT_LINKTREE"))
	     (target   (common:args-get-target exit-if-bad: #t))
	     (runname  (or (args:get-arg "-runname")
			   (args:get-arg ":runname")
			   (getenv "MT_RUNNAME")))
	     (fulldir  (conc linktree "/"
			     target "/"
			     runname)))
	(if (and linktree (file-exists? linktree)) ;; can't proceed without linktree
	    (begin
	      (debug:print-info 0 *default-log-port* "Have -run with target=" target ", runname=" runname ", fulldir=" fulldir ", testpatt=" (or (args:get-arg "-testpatt") "%"))
	      (if (not (file-exists? fulldir))
		  (create-directory fulldir #t)) ;; need to protect with exception handler 
	      (if (and target
		       runname
		       (file-exists? fulldir))
		  (let ((tmpfile  (conc fulldir "/.megatest.cfg." (current-seconds)))
			(targfile (conc fulldir "/.megatest.cfg-"  megatest-version "-" megatest-fossil-hash))
			(rconfig  (conc fulldir "/.runconfig." megatest-version "-" megatest-fossil-hash)))
		    (if (file-exists? rconfig) ;; only cache megatest.config AFTER runconfigs has been cached
			(begin
			  (debug:print-info 0 *default-log-port* "Caching megatest.config in " tmpfile)
                          (if (not (common:in-running-test?))
                              (configf:write-alist *configdat* tmpfile))
			  (system (conc "ln -sf " tmpfile " " targfile))))
		    )))
	    (debug:print-info 1 *default-log-port* "No linktree yet, no caching configs.")))))


;; gather available information, if legit read configs in this order:
;;
;;   if have cache;
;;      read it a return it
;;   else
;;     megatest.config     (do not cache)
;;     runconfigs.config   (cache if all vars avail)
;;     megatest.config     (cache if all vars avail)
;;   returns:
;;     *toppath*
;;   side effects:
;;     sets; *configdat*    (megatest.config info)
;;           *runconfigdat* (runconfigs.config info)
;;           *configstatus* (status of the read data)
;;
(define (launch:setup #!key (force-reread #f) (areapath #f))
  (mutex-lock! *launch-setup-mutex*)
  (if (and *toppath*
	   (eq? *configstatus* 'fulldata) (not force-reread)) ;; got it all
      (begin
	(debug:print 2 *default-log-port* "NOTE: skipping launch:setup-body call since we have fulldata")
	(mutex-unlock! *launch-setup-mutex*)
	*toppath*)
      (let ((res (launch:setup-body force-reread: force-reread areapath: areapath)))
	(mutex-unlock! *launch-setup-mutex*)
	res)))

;; return paths depending on what info is available.
;;
(define (launch:get-cache-file-paths areapath toppath target mtconfig)
  (let* ((use-cache (common:use-cache?))
         (runname  (common:args-get-runname))
         (linktree (common:get-linktree))
         (testname (common:get-full-test-name))
         (rundir   (if (and runname target linktree)
                       (common:directory-writable? (conc linktree "/" target "/" runname))
                       #f))
         (testdir  (if (and rundir testname)
                       (common:directory-writable? (conc rundir "/" testname))
                       #f))
         (cachedir (or testdir rundir))
         (mtcachef (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
         (rccachef (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash))))
    (debug:print-info 6 *default-log-port* 
                      "runname=" runname 
                      "\n  linktree=" linktree
                      "\n  testname=" testname
                      "\n  rundir=" rundir 
                      "\n  testdir=" testdir 
                      "\n  cachedir=" cachedir
                      "\n  mtcachef=" mtcachef
                      "\n  rccachef=" rccachef)
    (cons mtcachef rccachef)))

(define (launch:setup-body #!key (force-reread #f) (areapath #f))
  (if (and (eq? *configstatus* 'fulldata)
	   *toppath*
	   (not force-reread)) ;; no need to reprocess
      *toppath*   ;; return toppath
      (let* ((use-cache (common:use-cache?)) ;; BB- use-cache checks *configdat* for use-cache setting.  We do not have *configdat*.  Bootstrapping problem here.
	     (toppath  (or *toppath* areapath (getenv "MT_RUN_AREA_HOME"))) ;; preserve toppath

	     (target   (common:args-get-target))


	     (sections (if target (list "default" target) #f)) ;; for runconfigs
	     (mtconfig (or (args:get-arg "-config") "megatest.config")) ;; allow overriding megatest.config 
             (cachefiles (launch:get-cache-file-paths areapath toppath target mtconfig))
	     ;; checking for null cachefiles should not be necessary, I was seeing error car of '(), might be a chicken bug or a red herring ...
	     (mtcachef   (if (null? cachefiles)
			     #f
			     (car cachefiles))) ;; (and cachedir (conc cachedir "/" ".megatest.cfg-"  megatest-version "-" megatest-fossil-hash)))
	     (rccachef   (if (null? cachefiles)
			     #f
			     (cdr cachefiles)))) ;; (and cachedir (conc cachedir "/" ".runconfigs.cfg-"  megatest-version "-" megatest-fossil-hash)))


	      ;; (cancreate (and cachedir (common:file-exists? cachedir)(file-write-access? cachedir) (not (common:in-running-test?)))))




	(set! *toppath* toppath) ;; This is needed when we are running as a test using CMDINFO as a datasource
        ;;(BB> "launch:setup-body -- cachefiles="cachefiles)
	(cond



	 ;; if mtcachef exists just read it, however we need to assume toppath is available in $MT_RUN_AREA_HOME
	 ((and (not force-reread) mtcachef (common:file-exists? mtcachef) (get-environment-variable "MT_RUN_AREA_HOME") use-cache)
          ;;(BB> "launch:setup-body -- cond branch 1 - use-cache")
          (set! *configdat*    (configf:read-alist mtcachef))
          ;;(BB> "launch:setup-body -- 1 set! *configdat*="*configdat*)
	  (set! *runconfigdat* (configf:read-alist rccachef))
	  (set! *configinfo*   (list *configdat*  (get-environment-variable "MT_RUN_AREA_HOME")))
	  (set! *configstatus* 'fulldata)
	  (set! *toppath*      (get-environment-variable "MT_RUN_AREA_HOME"))
	  *toppath*)
	 ;; we have all the info needed to fully process runconfigs and megatest.config
	 ((and (not force-reread) mtcachef) ;; BB- why are we doing this without asking if caching is desired?
          ;;(BB> "launch:setup-body -- cond branch 2")
	  (let* ((first-pass    (find-and-read-config        ;; NB// sets MT_RUN_AREA_HOME as side effect
				 mtconfig
				 environ-patt: "env-override"
				 given-toppath: toppath
				 pathenvvar: "MT_RUN_AREA_HOME"))
		 (first-rundat  (let ((toppath (if toppath 
						   toppath
						   (car first-pass))))
				  (read-config ;; (conc toppath "/runconfigs.config") ;; this should be converted to runconfig:read but it is non-trivial, leaving it for now.
				   (conc (if (string? toppath)
					     toppath
					     (get-environment-variable "MT_RUN_AREA_HOME"))
					 "/runconfigs.config")
				   *runconfigdat* #t 
				   sections: sections))))
	    (set! *runconfigdat* first-rundat)
	    (if first-pass  ;; 
		(begin
                  ;;(BB> "launch:setup-body -- \"first-pass\"=first-pass")
		  (set! *configdat*  (car first-pass))
                  ;;(BB> "launch:setup-body -- 2 set! *configdat*="*configdat*)
		  (set! *configinfo* first-pass)
		  (set! *toppath*    (or toppath (cadr first-pass))) ;; use the gathered data unless already have it
		  (set! toppath      *toppath*)
		  (if (not *toppath*)
		      (begin
			(debug:print-error 0 *default-log-port* "you are not in a megatest area!")
			(exit 1)))
		  (setenv "MT_RUN_AREA_HOME" *toppath*)
		  ;; the seed read is done, now read runconfigs, cache it then read megatest.config one more time and cache it
		  (let* ((keys         (rmt:get-keys))
			 (key-vals     (keys:target->keyval keys target))
			 (linktree     (common:get-linktree)) ;; (or (getenv "MT_LINKTREE")(if *configdat* (configf:lookup *configdat* "setup" "linktree") #f)))
					;     (if *configdat*
					; 	   (configf:lookup *configdat* "setup" "linktree")
					; 	   (conc *toppath* "/lt"))))
			 (second-pass  (find-and-read-config
					mtconfig
					environ-patt: "env-override"
					given-toppath: toppath
					pathenvvar: "MT_RUN_AREA_HOME"))
			 (runconfigdat (begin     ;; this read of the runconfigs will see any adjustments made by re-reading megatest.config
					 (for-each (lambda (kt)
						     (setenv (car kt) (cadr kt)))
						   key-vals)
					 (read-config (conc toppath "/runconfigs.config") *runconfigdat* #t ;; consider using runconfig:read some day ...
						      sections: sections)))
                         (cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
                         (mtcachef     (car cachefiles))
                         (rccachef     (cdr cachefiles)))
		    (if rccachef (configf:write-alist runconfigdat rccachef))
		    (set! *runconfigdat* runconfigdat)
		    (if mtcachef (configf:write-alist *configdat* mtcachef))
		    (if (and rccachef mtcachef) (set! *configstatus* 'fulldata))))
		;; no configs found? should not happen but let's try to recover gracefully, return an empty hash-table
		(begin (set! *configdat* (make-hash-table))
                       ;;(BB> "launch:setup-body -- 3 set! *configdat*="*configdat*)
                       )
		)))
	 ;; else read what you can and set the flag accordingly
	 (else
          ;;(BB> "launch:setup-body -- cond branch 3 - else")
	  (let* ((cfgdat   (find-and-read-config 
			    (or (args:get-arg "-config") "megatest.config")
			    environ-patt: "env-override"
			    given-toppath: (get-environment-variable "MT_RUN_AREA_HOME")
			    pathenvvar: "MT_RUN_AREA_HOME")))

            (if (and cfgdat (list? cfgdat) (> (length cfgdat) 0) (hash-table? (car cfgdat)))
		(let* ((toppath  (or (get-environment-variable "MT_RUN_AREA_HOME")(cadr cfgdat)))
		       (rdat     (read-config (conc toppath  ;; convert this to use runconfig:read!
						    "/runconfigs.config") *runconfigdat* #t sections: sections)))
		  (set! *configinfo*   cfgdat)
		  (set! *configdat*    (car cfgdat))
		  (set! *runconfigdat* rdat)
		  (set! *toppath*      toppath)
		  (set! *configstatus* 'partial))
		(begin
		  (debug:print-error 0 *default-log-port* "No " mtconfig " file found. Giving up.")
		  (exit 2))))))
	;; additional house keeping
	(let* ((linktree (common:get-linktree)))

	  (if linktree
	      (begin
		(if (not (common:file-exists? linktree))
		    (begin
		      (handle-exceptions
			  exn
			  (begin
			    (debug:print-error 0 *default-log-port* "Something went wrong when trying to create linktree dir at " linktree)
			    (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn))
			    (exit 1))
			(create-directory linktree #t))))
		(handle-exceptions
		    exn
		    (begin
		      (debug:print-error 0 *default-log-port* "Something went wrong when trying to create link to linktree at " *toppath*)
		      (debug:print 0 *default-log-port* " message: " ((condition-property-accessor 'exn 'message) exn)))
		  (let ((tlink (conc *toppath* "/lt")))
		    (if (not (file-exists? tlink))
			(create-symbolic-link linktree tlink)))))
	      (begin
		(debug:print-error 0 *default-log-port* "linktree not defined in [setup] section of megatest.config")
		)))
	(if (and *toppath*
		 (directory-exists? *toppath*))
	    (begin
	      (setenv "MT_RUN_AREA_HOME" *toppath*)
	      (setenv "MT_TESTSUITENAME" (common:get-testsuite-name)))
	    (begin
	      (debug:print-error 0 *default-log-port* "failed to find the top path to your Megatest area.")
	      ;;(exit 1)
	      (set! *toppath* #f) ;; force it to be false so we return #f
	      #f
	      ))
        ;; one more attempt to cache the configs for future reading
        (let* ((cachefiles   (launch:get-cache-file-paths areapath toppath target mtconfig))
               (mtcachef     (car cachefiles))
               (rccachef     (cdr cachefiles)))
          (if (and rccachef *runconfigdat*) (configf:write-alist *runconfigdat* rccachef))
          (if (and mtcachef *configdat*)    (configf:write-alist *configdat* mtcachef))
          (if (and rccachef mtcachef *runconfigdat* *configdat*)
              (set! *configstatus* 'fulldata)))

	;; if have -append-config then read and append here
	(let ((cfname (args:get-arg "-append-config")))
	  (if (and cfname
		   (file-read-access? cfname))
	      (read-config cfname *configdat* #t))) ;; values are added to the hash, no need to do anything special.
	*toppath*)))

(define (get-best-disk confdat testconfig)
  (let* ((disks   (or (and testconfig (hash-table-ref/default testconfig "disks" #f))
		      (hash-table-ref/default confdat "disks" #f)))
	 (minspace (let ((m (configf:lookup confdat "setup" "minspace")))
		     (string->number (or m "10000")))))
    (if disks 

(define (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat #!key (remtries 2))
  (let* ((item-path (if (string? itemdat) itemdat (item-list->path itemdat))) ;; if pass in string - just use it
	 (runname   (if (string? run-info) ;; if we pass in a string as run-info use it as run-name.
			run-info
			(db:get-value-by-header (db:get-rows run-info)
						(db:get-header run-info)
						"runname")))
	 (contour   #f) ;; NOT READY FOR THIS (args:get-arg "-contour"))
	 ;; convert back to db: from rdb: - this is always run at server end
	 (target   (string-intersperse (map cadr keyvals) "/"))

	 (not-iterated  (equal? "" item-path))

	 ;; all tests are found at <rundir>/test-base or <linkdir>/test-base
	 (testtop-base (conc target "/" runname "/" testname))
	 (test-base    (conc testtop-base (if not-iterated "" "/") item-path))

	 ;; nb// if itempath is not "" then it is prefixed with "/"
	 (toptest-path (conc disk-path (if contour (conc "/" contour) "") "/" testtop-base))
	 (test-path    (conc disk-path (if contour (conc "/" contour) "") "/" test-base))

	 ;; ensure this exists first as links to subtests must be created there
	 (linktree  (common:get-linktree))
	 ;; WAS: (let ((rd (config-lookup *configdat* "setup" "linktree")))
	 ;;         (if rd rd (conc *toppath* "/runs"))))
	 ;; which seems wrong ...

	 (lnkbase   (conc linktree (if contour (conc "/" contour) "") "/" target "/" runname))
	 (lnkpath   (conc lnkbase "/" testname))
	 (lnkpathf  (conc lnkpath (if not-iterated "" "/") item-path))
	 (lnktarget (conc lnkpath "/" item-path)))

    ;; Update the rundir path in the test record for all, rundir=physical, shortdir=logical
    ;;                                                 rundir   shortdir
    (rmt:general-call 'test-set-rundir-shortdir run-id lnkpathf test-path testname item-path run-id)

    (debug:print 2 *default-log-port* "INFO:\n       lnkbase=" lnkbase "\n       lnkpath=" lnkpath "\n  toptest-path=" toptest-path "\n     test-path=" test-path)
    (if (not (common:file-exists? linktree))
	(begin
	  (debug:print 0 *default-log-port* "WARNING: linktree did not exist! Creating it now at " linktree)
	  (create-directory linktree #t))) ;; (system (conc "mkdir -p " linktree))))
    ;; create the directory for the tests dir links, this is needed no matter what...
    (if (and (not (common:directory-exists? lnkbase))
	     (not (common:file-exists? lnkbase)))
	(handle-exceptions
	 exn
	 (begin
	   (debug:print-error 0 *default-log-port* "Problem creating linktree base at " lnkbase)
	   (print-error-message exn (current-error-port)))
	 (create-directory lnkbase #t)))
    

;; 4. remotely run the test on allocated host
;;    - could be ssh to host from hosts table (update regularly with load)
;;    - could be netbatch
;;      (launch-test db (cadr status) test-conf))
(define (launch-test test-id run-id run-info keyvals runname test-conf test-name test-path itemdat params)
  (mutex-lock! *launch-setup-mutex*) ;; setting variables and processing the testconfig is NOT thread-safe, reuse the launch-setup mutex
  (let* ((item-path       (item-list->path itemdat))
	 (contour         #f)) ;; NOT READY FOR THIS (args:get-arg "-contour")))
    (let loop ((delta        (- (current-seconds) *last-launch*))
	       (launch-delay (configf:lookup-number *configdat* "setup" "launch-delay" default: 1)))
      (if (> launch-delay delta)
	  (begin
	    (if (common:low-noise-print 1200 "test launch delay") ;; every two hours or so remind the user about launch delay.
		(debug:print-info 0 *default-log-port* "NOTE: test launches are delayed by " launch-delay " seconds. See megatest.config launch-delay setting to adjust.")) ;; launch of " test-name " for " (- launch-delay delta) " seconds"))
	    (thread-sleep! (- launch-delay delta))
	    (loop (- (current-seconds) *last-launch*) launch-delay))))
    (change-directory *toppath*)
    (alist->env-vars ;; consolidate this code with the code in megatest.scm for "-execute", *maybe* - the longer they are set the longer each launch takes (must be non-overlapping with the vars)
     (append
      (list
       (list "MT_RUN_AREA_HOME" *toppath*)

	    (create-directory work-area #t)
	    (debug:print 0 *default-log-port* "WARNING: No disk work area specified - running in the test directory under tmp_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)))
							   (if hhdat
							       (car hhdat)
							       #f)))
					(list 'serverurl (if *runremote*
							     (remote-server-url *runremote*)
							     #f)) ;;
					(list 'areaname  (common:get-testsuite-name))
					(list 'toppath   *toppath*)
					(list 'work-area work-area)
					(list 'test-name test-name) 
					(list 'runscript runscript) 
					(list 'run-id    run-id   )
					(list 'test-id   test-id  )
					;; (list 'item-path item-path )