Megatest

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

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

(include "common_records.scm")
(include "key_records.scm")
(include "db_records.scm")

;;======================================================================
;; ezsteps

	       (rollup-status 0))
	  (change-directory top-path)
	  (debug:print 2 "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
	  ;; Setup the *runremote* global var
	  (if *runremote* (debug:print 2 "ERROR: I'm not expecting *runremote* to be set at this time"))
	  ;; (set! *runremote* runremote)
	  ;; (set! *transport-type* (string->symbol transport))
	  (set! keys       (rmt:get-keys))
	  (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 "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 "Adding pre-var/val " var " = " val " to the environment")
			       (setenv var val)))))
		     varpairs)))
	  (for-each
	   (lambda (varval)
	     (let ((var (car varval))
		   (val (cadr varval)))
	       (if val
		   (setenv var val)
		   (begin
		     (debug:print 0 "ERROR: 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_RUNNAME"   runname)
	      (list  "MT_MEGATEST"  megatest)
	      (list  "MT_TARGET"    target)
	      (list  "MT_LINKTREE"  (configf:lookup *configdat* "setup" "linktree"))))
	  (if mt-bindir-path (setenv "PATH" (conc (getenv "PATH") ":" mt-bindir-path)))
	  ;; (change-directory top-path)
	  (if (not (setup-for-run))
	      (begin
		(debug:print 0 "Failed to setup, exiting") 
		;; (sqlite3:finalize! db)
		;; (sqlite3:finalize! tdb)
		(exit 1)))
	  ;; Can setup as client for server mode now
	  ;; (client:setup)

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

	  (set-run-config-vars run-id keyvals target) ;; (db:get-target db run-id))
	  ;; environment overrides are done *before* the remaining critical envars.
	  (alist->env-vars env-ovrd)
	  (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! test-id "REMOTEHOSTSTART" "n/a")
	  (thread-sleep! 0.3) ;; NFS slowness has caused grief here

	  ;; (cdb:set-test-start-time!  *runremote* test-id)
	  (if (args:get-arg "-xterm")

						   ;;   (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! 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! 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)))
							    (overall-status   (cond
									       ((eq? rollup-status 2) 'warn)
									       ((eq? rollup-status 0) 'pass)
									       (else 'fail)))
							    (next-status      (cond 
									       ((eq? overall-status 'pass) this-step-status)
									       ((eq? overall-status 'warn)
										(if (eq? this-step-status 'fail) 'fail 'warn))
									       (else 'fail)))
							    (next-state       ;; "RUNNING") ;; WHY WAS THIS CHANGED TO NOT USE (null? tal) ??
							                       (cond
									       ((null? tal) ;; more to run?
									        "COMPLETED")
									       (else "RUNNING")))
							    )
						       (debug:print 4 "Exit value received: " (vector-ref exit-info 2) " logpro-used: " logpro-used 
								    " this-step-status: " this-step-status " overall-status: " overall-status 
								    " next-status: " next-status " rollup-status: " rollup-status)
						       (case next-status
							 ((warn)
							  (set! rollup-status 2)
							  ;; NB// test-set-status! does rdb calls under the hood
							  (tests:test-set-status! test-id next-state "WARN" 
									  (if (eq? this-step-status 'warn) "Logpro warning found" #f)
									  #f))
							 ((pass)
							  (tests:test-set-status! test-id next-state "PASS" #f #f))
							 (else ;; 'fail
							  (set! rollup-status 1) ;; force fail, this used to be next-state but that doesn't make sense. should always be "COMPLETED" 
							  (tests:test-set-status! test-id "COMPLETED" "FAIL" (conc "Failed at step " stepname) #f)
							  ))))
						   (if (and (steprun-good? logpro-used (vector-ref exit-info 2))
							    (not (null? tal)))
						       (loop (car tal) (cdr tal) stepname)))
						 (debug:print 4 "WARNING: a prior step failed, stopping at " ezstep))))))))
		 (monitorjob   (lambda ()
				 (let* ((start-seconds (current-seconds))
					(calc-minutes  (lambda ()
							 (inexact->exact 
							  (round 
							   (- 
							    (current-seconds) 
							    start-seconds)))))
					(kill-tries 0))
				   (tests:set-full-meta-info test-id run-id (calc-minutes) work-area)
				   (let loop ((minutes   (calc-minutes)))
				     (begin
				       (set! kill-job? (or (test-get-kill-request 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 "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim)
										#t)
									      #f)))))
				       ;; open-run-close not needed for test-set-meta-info
				       (tests:set-partial-meta-info test-id run-id minutes work-area)
				       (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* ((pid (vector-ref exit-info 0)))

						   ;;      	  (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)
						     (exit 1) ;; IS THIS NECESSARY OR WISE???
						     )))
					     (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 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 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

		    ;; (if (not (equal? item-path ""))
		    ;;     (begin
		    ;;       (thread-sleep! 0.1) ;; give other processes an opportunity to access the db as rollup is lower priority
		    ;;       (cdb:roll-up-pass-fail-counts *runremote* run-id test-name item-path new-status)))
		    ))
	      ;; for automated creation of the rollup html file this is a good place...
	      (if (not (equal? item-path ""))
		  (tests:summarize-items run-id test-id test-name #f))) ;; don't force - just update if no
	    (mutex-unlock! m)
	    (debug:print 2 "Output from running " fullrunscript ", pid " (vector-ref exit-info 0) " in work area " 
			 work-area ":\n====\n exit code " (vector-ref exit-info 2) "\n" "====\n")
	    (if (not (vector-ref exit-info 1))
		(exit 4)))))))

;; set up the very basics needed for doing anything here.

			    (if (args:get-arg "-config")(args:get-arg "-config") "megatest.config")
			    environ-patt: "env-override"
			    given-toppath: (get-environment-variable "MT_RUN_AREA_HOME")
			    pathenvvar: "MT_RUN_AREA_HOME"))
	(set! *configdat*  (if (car *configinfo*)(car *configinfo*) #f))
	(set! *toppath*    (if (car *configinfo*)(cadr *configinfo*) #f))
	(if *toppath*
	    (let ((dbdir (conc *toppath* "/db")))
	      (handle-exceptions
	       exn
	       (debug:print 0 "ERROR: failed to create the " dbdir " area for your database files")
	       (if (not (directory-exists? dbdir))(create-directory dbdir)))
	      (setenv "MT_RUN_AREA_HOME" *toppath*))
	    (debug:print 0 "ERROR: failed to find the top path to your Megatest area."))))
  *toppath*)

(define (get-best-disk confdat)
  (let* ((disks    (hash-table-ref/default confdat "disks" #f))
	 (best     #f)
	 (bestsize 0))

		      (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
    (rmt:general-call 'test-set-rundir-by-test-id lnkpathf test-id)

    (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. 

    ;; 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       (rmt:get-test-info-by-id 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?
	  (rmt:general-call 'test-set-rundir lnkpath run-id 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)))))

		       " - creating link from: " test-path "\n"
		       "                   to: " lnktarget)

	  ;; If there is already a symlink delete it and recreate it.
	  (handle-exceptions
	   exn
	   (begin
	     (debug:print 0 "ERROR:  Failed to re-create link " linktarget ((condition-property-accessor 'exn 'message) exn) ", exiting")
	     (exit))
	   (if (symbolic-link? lnktarget)     (delete-file lnktarget))
	   (if (not (file-exists? lnktarget)) (create-symbolic-link test-path lnktarget)))))










    (if (directory? test-path)
	(begin
	  (let* ((ovrcmd (let ((cmd (config-lookup *configdat* "setup" "testcopycmd")))
			   (if cmd
			       ;; substitute the TEST_SRC_PATH and TEST_TARG_PATH
			       (string-substitute "TEST_TARG_PATH" test-path
						  (string-substitute "TEST_SRC_PATH" test-src-path cmd #t) #t)

	 (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   (rmt:get-test-info-by-id test-id))
	 (mt_target  (string-intersperse (map cadr keyvals) "/"))
	 (debug-param (append (if (args:get-arg "-debug")  (list "-debug" (args:get-arg "-debug")) '())
			      (if (args:get-arg "-logging")(list "-logging") '()))))
    (setenv "MT_ITEMPATH" item-path)
    (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"))