Megatest

;;  PURPOSE.

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

(use regex regex-case base64 sqlite3 srfi-18 directory-utils posix-extras)
(import (prefix base64 base64:))
(import (prefix sqlite3 sqlite3:))

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

	  (set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals)
	  (set-item-env-vars itemdat)
	  (save-environment-as-files "megatest")
	  ;; open-run-close not needed for test-set-meta-info
	  (tests:set-full-meta-info test-id run-id 0 work-area)

	  ;; (tests:test-set-status! test-id "REMOTEHOSTSTART" "n/a" (args:get-arg "-m") #f)
	  (tests:test-force-state-status! run-id test-id "REMOTEHOSTSTART" "n/a")
	  (thread-sleep! 0.3) ;; NFS slowness has caused grief here

	  (if (args:get-arg "-xterm")
	      (set! fullrunscript "xterm")
	      (if (and fullrunscript (not (file-execute-access? fullrunscript)))
		  (system (conc "chmod ug+x " fullrunscript))))
	  ;; We are about to actually kick off the test

				 ;; (tests:test-set-status! test-id "RUNNING" "n/a" #f #f)
				 ;; Since we should have a clean slate at this time there is no need to do 
				 ;; any of the other stuff that tests:test-set-status! does. Let's just 
				 ;; force RUNNING/n/a
				 

				 (thread-sleep! 0.3)
				 (tests:test-force-state-status! run-id test-id "RUNNING" "n/a")
				 (thread-sleep! 0.3) ;; NFS slowness has caused grief here

				 ;; if there is a runscript do it first
				 (if fullrunscript
				     (let ((pid (process-run fullrunscript)))
				       (let loop ((i 0))
					 (let-values

						   ;;   (if (and prevstep (file-exists? prev-env))
						   ;;       (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)
						   (rmt:teststep-set-status! run-id 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") "")))
						       (rmt:teststep-set-status! run-id test-id stepname "end" exinfo #f logfna))
						     (if logpro-used
							 (rmt:test-set-log! 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)))

					     (set! kill-tries (+ 1 kill-tries))
					     (mutex-unlock! m)))
				       (if keep-going
					   (begin
					     (thread-sleep! 3) ;; (+ 3 (random 6))) ;; add some jitter to the call home time to spread out the db accesses
					     (if keep-going
						 (loop (calc-minutes)))))))
				   (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
		 (th1          (make-thread monitorjob "monitor job"))
		 (th2          (make-thread runit "run job")))
	    (set! job-thread th2)
	    (thread-start! th1)
	    (thread-start! th2)
	    (thread-join! th2)
	    (set! keep-going #f)
	    (thread-join! th1)
	    (thread-sleep! 1)       ;; give thread th1 a chance to be done TODO: Verify this is needed. At 0.1 I was getting fail to stop, increased to total of 1.1 sec.
	    (mutex-lock! m)
	    (let* ((item-path (item-list->path itemdat))
		   ;; only state and status needed - use lazy routine
		   (testinfo  (rmt:get-testinfo-state-status run-id test-id))) ;;;(cdb:get-test-info-by-id *runremote* test-id))) ;; )) ;; run-id test-name item-path)))
	      ;; Am I completed?
	      (if (member (db:test-get-state testinfo) '("REMOTEHOSTSTART" "RUNNING")) ;; NOTE: It should *not* be REMOTEHOSTSTART but for reasons I don't yet understand it sometimes gets stuck in that state ;; (not (equal? (db:test-get-state testinfo) "COMPLETED"))
		  (let ((new-state  (if kill-job? "KILLED" "COMPLETED") ;; (if (eq? (vector-ref exit-info 2) 0) ;; exited with "good" status
				                                        ;; "COMPLETED"
							                ;; (db:test-get-state testinfo)))   ;; else preseve the state as set within the test
				    )
			(new-status (cond

	 (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 (filedb:register-path *fdb* lnkpathf))
    (rmt:general-call 'test-set-rundir-shortdir run-id lnkpathf test-path testname item-path)

    (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 (and (not (directory-exists? lnkbase))
	     (not (file-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. 





















    ;; 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)

	(handle-exceptions
	 exn
	 (begin
	   (debug:print 0 "ERROR:  Failed to create symlink " lnkpath ((condition-property-accessor 'exn 'message) exn) ", exiting")
	   (exit 1))
	 (create-symbolic-link toptest-path lnkpath)))
    
    ;; NB - This was not working right - some top tests are not getting the path set!!!
    ;;
    ;; Do the setting of this record after the paths are created so that the shortdir can 
    ;; be set to the real directory location. This is safer for future clean up if the link
    ;; tree is damaged or lost.
    ;; 
    (if (not (hash-table-ref/default *toptest-paths* testname #f))
	(let* ((testinfo       (rmt:get-test-info-by-id run-id test-id)) ;;  run-id testname item-path))
	       (curr-test-path (if testinfo ;; (filedb:get-path *fdb*
							     ;; (db:get-path dbstruct
				   ;; (rmt:sdb-qry 'getstr 
				   (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 "" (filedb:register-path *fdb* lnkpath)) ;; toptest-path)
	  (rmt:general-call 'test-set-rundir-shortdir run-id lnkpath 
			    (if (file-exists? lnkpath)
				(resolve-pathname lnkpath)
				lnkpath)
			    testname "")
	  ;; (rmt:general-call 'test-set-rundir run-id lnkpath testname "") ;; 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)))))

    ;; The toptest path has been created, the link to the test in the linktree has
    ;; been created. Now, if this is an iterated test the real test dir must be created
    (if (not not-iterated) ;; this is an iterated test
	(let ((lnktarget (conc lnkpath "/" item-path)))
	  (debug:print 2 "Setting up sub test run area")
	  (debug:print 2 " - creating run area in " test-path)
	  (handle-exceptions