Megatest

	'())))

(define (launch:execute encoded-cmd)
  (let* ((cmdinfo   (read (open-input-string (base64:base64-decode encoded-cmd)))))
    (setenv "MT_CMDINFO" encoded-cmd)
    (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))  ;; How is testpath different from work-area ??
	       (top-path  (assoc/default 'toppath   cmdinfo))
	       (work-area (assoc/default 'work-area cmdinfo))
	       (test-name (assoc/default 'test-name cmdinfo))
	       (runscript (assoc/default 'runscript cmdinfo))
	       (ezsteps   (assoc/default 'ezsteps   cmdinfo))
	       (db-host   (assoc/default 'db-host   cmdinfo))
	       (run-id    (assoc/default 'run-id    cmdinfo))
	       (test-id   (assoc/default 'test-id   cmdinfo))
	       (target    (assoc/default 'target    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))
	       (mt-bindir-path (assoc/default 'mt-bindir-path cmdinfo))
	       (fullrunscript (if (not runscript)
                                  #f
                                  (if (substring-index "/" runscript)
                                      runscript ;; use unadultered if contains slashes
                                      (let ((fulln (conc testpath "/" runscript)))
	                                  (if (and (file-exists? fulln)
                                                   (file-execute-access? fulln))
                                              fulln
                                              runscript))))) ;; assume it is on the path
	       (rollup-status 0))
	  
	  (debug:print 2 "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
	  ;; 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 ",")))

		;; (sqlite3:finalize! db)
		;; (sqlite3:finalize! tdb)
		(exit 1)))
	  ;; Can setup as client for server mode now
	  (server:client-setup)

	  (change-directory *toppath*) 
	  (set-megatest-env-vars run-id) ;; these may be needed by the launching process
	  (change-directory work-area) 

	  (open-run-close set-run-config-vars #f run-id)
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
	  (set-megatest-env-vars run-id)
	  (set-item-env-vars itemdat)
	  (save-environment-as-files "megatest")
	  (open-run-close test-set-meta-info #f test-id run-id test-name itemdat 0)
	  (tests:test-set-status! test-id "REMOTEHOSTSTART" "n/a" (args:get-arg "-m") #f)
	  (if (args:get-arg "-xterm")
	      (set! fullrunscript "xterm")
	      (if (and fullrunscript (not (file-execute-access? fullrunscript)))

						   ;;       (set! script (conc script "source " prev-env))))
						   
						   ;; call the command using mt_ezstep
						   (set! script (conc "mt_ezstep " stepname " " (if prevstep prevstep "-") " " stepcmd))

						   (debug:print 4 "script: " script)

						   (cdb:remote-run db:teststep-set-status! #f test-id stepname "start" "-" #f #f)
						   ;; now launch
						   (let ((pid (process-run script)))
						     (let processloop ((i 0))
						       (let-values (((pid-val exit-status exit-code)(process-wait pid #t)))
								   (mutex-lock! m)
								   (vector-set! exit-info 0 pid)
								   (vector-set! exit-info 1 exit-status)
								   (vector-set! exit-info 2 exit-code)
								   (mutex-unlock! m)
								   (if (eq? pid-val 0)
								       (begin
									 (thread-sleep! 2)
									 (processloop (+ i 1))))
								   ))
                                                     (let ((exinfo (vector-ref exit-info 2))
                                                           (logfna (if logpro-used (conc stepname ".html") "")))
						       ;; testing if procedures called in a remote call cause problems (ans: no or so I suspect)
						       (cdb:remote-run db:teststep-set-status! #f test-id stepname "end" exinfo #f logfna))
						     (if logpro-used
							 (cdb:test-set-log! *runremote*  test-id (conc stepname ".html")))
						     ;; set the test final status
						     (let* ((this-step-status (cond
									       ((and (eq? (vector-ref exit-info 2) 2) logpro-used) 'warn)
									       ((eq? (vector-ref exit-info 2) 0)                   'pass)
									       (else 'fail)))
							    (overall-status   (cond
									       ((eq? rollup-status 2) 'warn)

							  (round 
							   (- 
							    (current-seconds) 
							    start-seconds)))))
					(kill-tries 0))
				   (let loop ((minutes   (calc-minutes)))
				     (begin
				       (set! kill-job? (test-get-kill-request test-id)) ;; run-id test-name itemdat))
				       (open-run-close test-set-meta-info #f test-id run-id test-name itemdat minutes)
				       (if kill-job? 
					   (begin
					     (mutex-lock! m)
					     (let* ((pid (vector-ref exit-info 0)))
					       (if (number? pid)
						   (begin
						     (debug:print 0 "WARNING: Request received to kill job (attempt # " kill-tries ")")
						     (let ((processes (cmd-run->list (conc "pgrep -l -P " pid))))
						       (for-each 
							(lambda (p)
							  (let* ((parts  (string-split p))
								 (p-id   (if (> (length parts) 0)
									     (string->number (car parts))
									     #f)))
							    (if p-id
								(begin
								  (debug:print 0 "Killing " (cadr parts) "; kill -9  " p-id)
								  (system (conc "kill -9 " p-id))))))
							(car processes))
						       (system (conc "kill -9 -" pid))))
						   (begin
						     (debug:print 0 "WARNING: Request received to kill job but problem with process, attempting to kill manager process")
						     (tests:test-set-status! test-id "KILLED"  "FAIL"
								     (args:get-arg "-m") #f)
						     (sqlite3:finalize! tdb)
						     (exit 1))))
					     (set! kill-tries (+ 1 kill-tries))
					     (mutex-unlock! m)))
				       ;; (sqlite3:finalize! db)
				       (thread-sleep! (+ 10 (random 10))) ;; add some jitter to the call home time to spread out the db accesses
				       (loop (calc-minutes)))))))
		 (th1          (make-thread monitorjob))
		 (th2          (make-thread runit)))
	    (set! job-thread th2)
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th2)
	    (mutex-lock! m)
	    (let* ((item-path (item-list->path itemdat))
		   (testinfo  (cdb:get-test-info-by-id *runremote* test-id))) ;; )) ;; run-id test-name item-path)))
	      ;; Am I completed?
	      (if (not (equal? (db:test-get-state testinfo) "COMPLETED"))
		  (begin
		    (debug:print 2 "Test NOT logged as COMPLETED, (state=" (db:test-get-state testinfo) "), updating result, rollup-status is " rollup-status)
		    (tests:test-set-status! test-id 
				    (if kill-job? "KILLED" "COMPLETED")
				    (cond
				     ((not (vector-ref exit-info 1)) "FAIL") ;; job failed to run

;;  <linkdir> - <target> - <testname> [ - <itempath> ]
;; 
;; All log file links should be stored relative to the top of link path
;;  
;; <target> - <testname> [ - <itempath> ] 
;;
(define (create-work-area db run-id test-id test-src-path disk-path testname itemdat)
  (let* ((run-info (cdb:remote-run db:get-run-info #f run-id))
	 (item-path (item-list->path itemdat))
	 (runname  (db:get-value-by-header (db:get-row run-info)
					   (db:get-header run-info)
					   "runname"))
	 ;; convert back to db: from rdb: - this is always run at server end
	 (key-vals (cdb:remote-run db:get-key-vals #f run-id))
	 (target   (string-intersperse key-vals "/"))

	 (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))

		      (if rd rd (conc *toppath* "/runs"))))

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

    ;; Update the rundir path in the test record for all
    (cdb:test-set-rundir-by-test-id *runremote* test-id lnkpathf)

    (debug:print 2 "INFO:\n       lnkbase=" lnkbase "\n       lnkpath=" lnkpath "\n  toptest-path=" toptest-path "\n     test-path=" test-path)
    (if (not (file-exists? linktree))
	(begin
	  (debug:print 0 "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 (not (directory-exists? lnkbase))
	(create-directory lnkbase #t))
    
    ;; update the toptest record with its location rundir, cache the path
    ;; This wass highly inefficient, one db write for every subtest, potentially
    ;; thousands of unnecessary updates, cache the fact it was set and don't set it 
    ;; again. 

    ;; NB - This is not working right - some top tests are not getting the path set!!!

    (if (not (hash-table-ref/default *toptest-paths* testname #f))
	(let* ((testinfo       (cdb:get-test-info-by-id *runremote* test-id)) ;;  run-id testname item-path))
	       (curr-test-path (if testinfo (db:test-get-rundir testinfo) #f)))
	  (hash-table-set! *toptest-paths* testname curr-test-path)
	  ;; NB// Was this for the test or for the parent in an iterated test?
	  (cdb:test-set-rundir! *runremote* run-id testname "" lnkpath) ;; toptest-path)
	  (if (or (not curr-test-path)
		  (not (directory-exists? toptest-path)))
	      (begin
		(debug:print-info 2 "Creating " toptest-path " and link " lnkpath)
		(create-directory toptest-path #t)
		(hash-table-set! *toptest-paths* testname toptest-path)))))

    ;; Now create the link from the test path to the link tree, however
    ;; if the test is iterated it is necessary to create the parent path
    ;; to the iteration. use pathname-directory to trim the path by one
    ;; level
    (if (not not-iterated) ;; i.e. iterated
	(let ((iterated-parent  (pathname-directory (conc lnkpath "/" item-path))))
	  (debug:print-info 2 "Creating iterated parent " iterated-parent)
	  (create-directory iterated-parent #t)))

    (if (symbolic-link? lnkpath) (delete-file lnkpath))
    (if (not (or (file-exists? lnkpath)
		 (symbolic-link? lnkpath)))
	(create-symbolic-link toptest-path lnkpath))
    

    ;;   (if (and (file-exists? testlink)
    ;;            (or (regular-file? testlink)
    ;;     	   (symbolic-link? testlink)))
    ;;       (system (conc "rm -f " testlink)))
    ;;   (system  (conc "ln -sf " test-path " " testlink)))
    (if (directory? test-path)
	(begin
	  (let* ((cmd    (conc "rsync -av" (if (debug:debug-mode 1) "" "q") " " test-src-path "/ " test-path "/"))
		 (status (system cmd)))
	    (if (not (eq? status 0))
		(debug:print 2 "ERROR: problem with running \"" cmd "\"")))
	  (list lnkpathf lnkpath ))
	(list #f #f))))

;; 1. look though disks list for disk with most space

	 (work-area  #f)
	 (toptest-work-area #f) ;; for iterated tests the top test contains data relevant for all
	 (diskpath   #f)
	 (cmdparms   #f)
	 (fullcmd    #f) ;; (define a (with-output-to-string (lambda ()(write x))))
	 (mt-bindir-path #f)
	 (item-path (item-list->path itemdat))
	 (test-id    (cdb:remote-run db:get-test-id #f run-id test-name item-path))
	 (testinfo   (cdb:get-test-info-by-id *runremote* test-id))
	 (mt_target  (string-intersperse (map cadr keyvallst) "/"))
	 (debug-param (append (if (args:get-arg "-debug")  (list "-debug" (args:get-arg "-debug")) '())
			      (if (args:get-arg "-logging")(list "-logging") '()))))
    (if hosts (set! hosts (string-split hosts)))
    ;; set the megatest to be called on the remote host
    (if (not remote-megatest)(set! remote-megatest local-megatest)) ;; "megatest"))
    (set! mt-bindir-path (pathname-directory remote-megatest))
    (if launcher (set! launcher (string-split launcher)))
    ;; set up the run work area for this test
    (set! diskpath (get-best-disk *configdat*))
    (if diskpath
	(let ((dat  (open-run-close create-work-area db run-id test-id test-path diskpath test-name itemdat)))
	  (set! work-area (car dat))
	  (set! toptest-work-area (cadr dat))
	  (debug:print-info 2 "Using work area " work-area))
	(begin
	  (set! work-area (conc test-path "/tmp_run"))
	  (create-directory work-area #t)
	  (debug:print 0 "WARNING: No disk work area specified - running in the test directory under tmp_run")))
    (set! cmdparms (base64:base64-encode (with-output-to-string
					   (lambda () ;; (list 'hosts     hosts)
					     (write (list (list 'testpath  test-path)

							  (list 'ezsteps   ezsteps) 
							  (list 'target    mt_target)
							  (list 'env-ovrd  (hash-table-ref/default *configdat* "env-override" '())) 
							  (list 'set-vars  (if params (hash-table-ref/default params "-setvars" #f)))
							  (list 'runname   runname)
							  (list 'mt-bindir-path mt-bindir-path))))))) ;; (string-intersperse keyvallst " "))))
    ;; clean out step records from previous run if they exist
    ;; (debug:print-info 4 "FIXMEEEEE!!!! This can be removed some day, perhaps move all test records to the test db?")
    ;; (open-run-close db:delete-test-step-records db test-id)
    (change-directory work-area) ;; so that log files from the launch process don't clutter the test dir
    (tests:test-set-status! test-id "LAUNCHED" "n/a" #f #f) ;; (if launch-results launch-results "FAILED"))
    (cond
     ((and launcher hosts) ;; must be using ssh hostname
      (set! fullcmd (append launcher (car hosts)(list remote-megatest test-sig "-execute" cmdparms) debug-param)))
     ;; (set! fullcmd (append launcher (car hosts)(list remote-megatest test-sig "-execute" cmdparms))))
     (launcher