Megatest

;; db stats
(define *db-stats*            (make-hash-table)) ;; hash of vectors < count duration-total >
(define *db-stats-mutex*      (make-mutex))
;; db access
(define *db-last-access*      (current-seconds)) ;; last db access, used in server
(define *db-write-access*     #t)
;; db sync

(define *db-last-sync*        0)                 ;; last time the sync to megatest.db happened
(define *db-sync-in-progress* #f)                ;; if there is a sync in progress do not try to start another
(define *db-multi-sync-mutex* (make-mutex))      ;; protect access to *db-sync-in-progress*, *db-last-sync*
;; task db
(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))
(define *db-cache-path*       #f)

;; SERVER

(define *keys*              (make-hash-table)) ;; cache the keys here
(define *keyvals*           (make-hash-table))
(define *toptest-paths*     (make-hash-table)) ;; cache toptest path settings here
(define *test-paths*        (make-hash-table)) ;; cache test-id to test run paths here
(define *test-ids*          (make-hash-table)) ;; cache run-id, testname, and item-path => test-id
(define *test-info*         (make-hash-table)) ;; cache the test info records, update the state, status, run_duration etc. from testdat.db

(define *run-info-cache*     (make-hash-table)) ;; run info is stable, no need to reget
(define *launch-setup-mutex* (make-mutex))     ;; need to be able to call launch:setup often so mutex it and re-call the real deal only if *toppath* not set
(define *homehost-mutex*     (make-mutex))

(defstruct remote
  (hh-dat            (common:get-homehost)) ;; homehost record ( addr . hhflag )
  (server-url        (if *toppath* (server:read-dotserver *toppath*))) ;; (server:check-if-running *toppath*) #f))
  (last-server-check 0)  ;; last time we checked to see if the server was alive
  (conndat           #f)
  (transport         *transport-type*)
  (server-timeout    (or (server:get-timeout) 100))) ;; default to 100 seconds

;; launching and hosts
(defstruct host
  (reachable    #f)
  (last-update  0)
  (last-used    0)
  (last-cpuload 1))

(define *host-loads*         (make-hash-table))

;; cache environment vars for each run here
(define *env-vars-by-run-id* (make-hash-table))

;; Testconfig and runconfig caches. 
(define *testconfigs*        (make-hash-table)) ;; test-name => testconfig
(define *runconfigs*         (make-hash-table)) ;; target    => runconfig

;; This is a cache of pre-reqs met, don't re-calc in cases where called with same params less than
;; five seconds ago
(define *pre-reqs-met-cache* (make-hash-table))

;; cache of verbosity given string
;;
(define *verbosity-cache*    (make-hash-table))

(define (common:clear-caches)
  (set! *target*             (make-hash-table))
  (set! *keys*               (make-hash-table))
  (set! *keyvals*            (make-hash-table))
  (set! *toptest-paths*      (make-hash-table))
  (set! *test-paths*         (make-hash-table))

  (message-digest-string (md5-primitive) *toppath*))

;;======================================================================
;; E X I T   H A N D L I N G
;;======================================================================

(define (common:run-sync?)
    (and (common:on-homehost?)
	 (args:get-arg "-server")))

;;   (let ((ohh (common:on-homehost?))
;; 	(srv (args:get-arg "-server")))
;;     (and ohh srv)))
    ;; (debug:print-info 0 *default-log-port* "common:run-sync? ohh=" ohh ", srv=" srv)

;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
(define (common:sync-to-megatest.db dbstruct) 
  (let ((start-time         (current-seconds))
	(res                (db:multi-db-sync dbstruct 'new2old)))
    (let ((sync-time (- (current-seconds) start-time)))
      (debug:print-info 3 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
      (if (common:low-noise-print 30 "sync new to old")
	  (debug:print-info 0 *default-log-port* "Sync of newdb to olddb completed in " sync-time " seconds pid="(current-process-id))))
    res))

;; currently the primary job of the watchdog is to run the sync back to megatest.db from the db in /tmp
;; if we are on the homehost and we are a server (by definition we are on the homehost if we are a server)
;;
(define (common:watchdog)
  (thread-sleep! 0.05) ;; delay for startup
  (let ((legacy-sync (common:run-sync?))
	(debug-mode  (debug:debug-mode 1))
	(last-time   (current-seconds)))
    (debug:print-info 0 *default-log-port* "watchdog starting. legacy-sync is " legacy-sync" pid="(current-process-id))
    (if legacy-sync
	(let ((dbstruct (db:setup)))
	  (debug:print-info 0 *default-log-port* "Server running, periodic sync started.")
	  (let loop ()
            ;;(BB> "watchdog loop.  pid="(current-process-id))
	    ;; sync for filesystem local db writes
	    ;;
	    (mutex-lock! *db-multi-sync-mutex*)
	    (let* ((need-sync        (>= *db-last-access* *db-last-sync*)) ;; no sync since last write
		   (sync-in-progress *db-sync-in-progress*)
		   (should-sync      (> (- (current-seconds) *db-last-sync*) 5)) ;; sync every five seconds minimum
		   (will-sync        (and (or need-sync should-sync)
					  (not sync-in-progress)))
		   (start-time       (current-seconds)))
	      ;; (debug:print-info 0 *default-log-port* "need-sync: " need-sync " sync-in-progress: " sync-in-progress " should-sync: " should-sync " will-sync: " will-sync)
	      (if will-sync (set! *db-sync-in-progress* #t))

		  (if (and (not *time-to-exit*)
			   (< count 4)) ;; was 11, changing to 4. 
		      (begin
			(thread-sleep! 1)
			(delay-loop (+ count 1))))
		  (loop)))
	    (if (common:low-noise-print 30)
		(debug:print-info 0 *default-log-port* "Exiting watchdog timer, *time-to-exit* = " *time-to-exit*" pid="(current-process-id))))))))

(define (std-exit-procedure)
  (on-exit (lambda () 0))
  (let ((no-hurry  (if *time-to-exit* ;; hurry up
		       #f
		       (begin
			 (set! *time-to-exit* #t)
			 #t))))
    (debug:print-info 4 *default-log-port* "starting exit process, finalizing databases.")
    (if (and no-hurry (debug:debug-mode 18))
	(rmt:print-db-stats))
    (let ((th1 (make-thread (lambda () ;; thread for cleaning up, give it five seconds
                              (if *dbstruct-db* (db:close-all *dbstruct-db*)) ;; one second allocated
			      (if *task-db*    
				  (let ((db (cdr *task-db*)))
				    (if (sqlite3:database? db)
					(begin
					  (sqlite3:interrupt! db)
					  (sqlite3:finalize! db #t)
					  ;; (vector-set! *task-db* 0 #f)
					  (set! *task-db* #f)))))
                              (if (and *runremote*
                                       (remote-conndat *runremote*))
                                  (begin
                                    (http-client#close-all-connections!))) ;; for http-client
                              (if (not (eq? *default-log-port* (current-error-port)))
                                  (close-output-port *default-log-port*))
			      (set! *default-log-port* (current-error-port))) "Cleanup db exit thread"))
	  (th2 (make-thread (lambda ()
			      (debug:print 4 *default-log-port* "Attempting clean exit. Please be patient and wait a few seconds...")
			      (if no-hurry
                                  (begin
                                    (thread-sleep! 5)) ;; give the clean up few seconds to do it's stuff
                                  (begin
      				  (thread-sleep! 2)))
      			      (debug:print 4 *default-log-port* " ... done")
      			      )
			    "clean exit")))
      (thread-start! th1)
      (thread-start! th2)
      (thread-join! th1)
      )
    )

  0)

(define (std-signal-handler signum)
  ;; (signal-mask! signum)
  (set! *time-to-exit* #t)
  (debug:print-error 0 *default-log-port* "Received signal " signum " exiting promptly")
  ;; (std-exit-procedure) ;; shouldn't need this since we are exiting and it will be called anyway
  (exit))

(define (common:args-get-state)
  (or (args:get-arg "-state")(args:get-arg ":state")))

(define (common:args-get-status)
  (or (args:get-arg "-status")(args:get-arg ":status")))

(define (common:args-get-testpatt rconf)
  (let* ((tagexpr (args:get-arg "-tagexpr"))
         (tags-testpatt (if tagexpr (string-join (runs:get-tests-matching-tags tagexpr) ",") #f))
         (testpatt-key  (if (args:get-arg "-mode") (args:get-arg "-mode") "TESTPATT"))
         (args-testpatt (or (args:get-arg "-testpatt") (args:get-arg "-runtests") "%"))


         (rtestpatt     (if rconf (runconfigs-get rconf testpatt-key) #f)))
    (cond
     (tags-testpatt
      (debug:print-info 0 *default-log-port* "-tagexpr "tagexpr" selects testpatt "tags-testpatt)
      tags-testpatt)
     ((and (equal? args-testpatt "%") rtestpatt)
      (debug:print-info 0 *default-log-port* "testpatt defined in "testpatt-key" from runconfigs: " rtestpatt)
      rtestpatt)
     (else args-testpatt))))
     
(define (common:get-linktree)
  (or (getenv "MT_LINKTREE")
      (if *configdat*
	  (configf:lookup *configdat* "setup" "linktree"))))

(define (common:args-get-runname)
  (let ((res (or (args:get-arg "-runname")

      (map (lambda (res)
	     (if (eof-object? res) 9e99 res))
	   (with-input-from-pipe 
	    (conc "ssh " remote-host " cat /proc/loadavg")
	    (lambda ()(list (read)(read)(read)))))
      (with-input-from-file "/proc/loadavg" 
	(lambda ()(list (read)(read)(read))))))

;; get normalized cpu load by reading from /proc/loadavg and /proc/cpuinfo return all three values and the number of real cpus and the number of threads
;; returns alist '((adj-cpu-load . normalized-proc-load) ... etc.
;;  keys: adj-proc-load, adj-core-load, 1m-load, 5m-load, 15m-load
;;
(define (common:get-normalized-cpu-load remote-host)
  (let ((data (if remote-host
                  (with-input-from-pipe 
                   (conc "ssh " remote-host " cat /proc/loadavg;cat /proc/cpuinfo;echo end")
                   read-lines)
                  (append 
                   (with-input-from-file "/proc/loadavg" 
                     read-lines)
                   (with-input-from-file "/proc/cpuinfo"
                     read-lines)
                   (list "end"))))
        (load-rx  (regexp "^([\\d\\.]+)\\s+([\\d\\.]+)\\s+([\\d\\.]+)\\s+.*$"))
        (proc-rx  (regexp "^processor\\s+:\\s+(\\d+)\\s*$"))
        (core-rx  (regexp "^core id\\s+:\\s+(\\d+)\\s*$"))
        (phys-rx  (regexp "^physical id\\s+:\\s+(\\d+)\\s*$"))
        (max-num  (lambda (p n)(max (string->number p) n))))
    ;; (print "data=" data)
    (if (null? data) ;; something went wrong
        #f
        (let loop ((hed      (car data))
                   (tal      (cdr data))
                   (loads    #f)
                   (proc-num 0)  ;; processor includes threads
                   (phys-num 0)  ;; physical chip on motherboard
                   (core-num 0)) ;; core
          ;; (print hed ", " loads ", " proc-num ", " phys-num ", " core-num)
          (if (null? tal) ;; have all our data, calculate normalized load and return result
              (let* ((act-proc (+ proc-num 1))
                     (act-phys (+ phys-num 1))
                     (act-core (+ core-num 1))
                     (adj-proc-load (/ (car loads) act-proc))
                     (adj-core-load (/ (car loads) act-core)))
                (append (list (cons 'adj-proc-load adj-proc-load)
                              (cons 'adj-core-load adj-core-load))
                        (list (cons '1m-load (car loads))
                              (cons '5m-load (cadr loads))
                              (cons '15m-load (caddr loads)))
                        (list (cons 'proc act-proc)
                              (cons 'core act-core)
                              (cons 'phys act-phys))))
              (regex-case
               hed
               (load-rx  ( x l1 l5 l15 ) (loop (car tal)(cdr tal)(map string->number (list l1 l5 l15)) proc-num phys-num core-num))
               (proc-rx  ( x p         ) (loop (car tal)(cdr tal) loads           (max-num p proc-num) phys-num core-num))
               (phys-rx  ( x p         ) (loop (car tal)(cdr tal) loads           proc-num (max-num p phys-num) core-num))
               (core-rx  ( x c         ) (loop (car tal)(cdr tal) loads           proc-num phys-num (max-num c core-num)))
               (else 
                (begin
                  ;; (print "NO MATCH: " hed)
                  (loop (car tal)(cdr tal) loads proc-num phys-num core-num)))))))))

(define (common:unix-ping hostname)
  (let ((res (system (conc "ping -c 1 " hostname " > /dev/null"))))
    (eq? res 0)))

;; ideally put all this info into the db, no need to preserve it across moving homehost
;;
;; return list of
;;  ( reachable? cpuload update-time )
(define (common:get-host-info hostname)
  (let* ((loadinfo (rmt:get-latest-host-load hostname))
         (load (car loadinfo))
         (load-sample-time (cdr loadinfo))
         (load-sample-age (- (current-seconds) load-sample-time))
         (loadinfo-timeout-seconds 20)
         (host-last-update-timeout-seconds 10)
         (host-rec (hash-table-ref/default *host-loads* hostname #f))
         )
    (cond
     ((< load-sample-age loadinfo-timeout-seconds)
      (list #t
            load-sample-time
            load))
     ((and host-rec
           (< (current-seconds) (+ (host-last-update host-rec) host-last-update-timeout-seconds)))
      (list #t
            (host-last-update host-rec)
            (host-last-cpuload host-rec )))
     ((common:unix-ping hostname)
      (list #t
            (current-seconds)
            (alist-ref 'adj-core-load (common:get-normalized-cpu-load hostname))))
     (else
      (list #f 0 -1)))))
    
(define (common:update-host-loads-table hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw)))
    (for-each
     (lambda (hostname)
       (let* ((rec       (let ((h (hash-table-ref/default *host-loads* hostname #f)))
                          (if h
                              h
                              (let ((h (make-host)))
                                (hash-table-set! *host-loads* hostname h)
                                h))))
              (host-info         (common:get-host-info hostname))
              (is-reachable      (car host-info))
              (last-reached-time (cadr host-info))
              (load              (caddr host-info)))
         (host-reachable-set!    rec is-reachable)
         (host-last-update-set!  rec last-reached-time)
         (host-last-cpuload-set! rec load)))
     hosts)))

(define (common:get-least-loaded-host hosts-raw)
  (let* ((hosts (filter (lambda (x)
                          (string-match (regexp "^\\S+$") x))
                        hosts-raw))
         (best-host #f)
         (best-load 99999)
         (curr-time (current-seconds)))
    (common:update-host-loads-table hosts)
    (for-each
     (lambda (hostname)
       (let* ((rec
               (let ((h (hash-table-ref/default *host-loads* hostname #f)))
                 (if h
                     h
                     (let ((h (make-host)))
                       (hash-table-set! *host-loads* hostname h)
                       h))))
              (reachable (host-reachable rec))
              (load      (host-last-cpuload   rec)))
         (cond
          ((not reachable) #f)
          ((< (+ load (/ (random 250) 1000))         ;; add a random factor to keep from getting in a rut
              (+ best-load (/ (random 250) 1000))  )
           (set! best-load load)
           (set! best-host hostname)))))
     hosts)
    best-host))




(define (common:wait-for-cpuload maxload numcpus waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload numcpus))
	 (loadjmp (- first next)))

    (query fetch-column
	   (sql db "SELECT ipaddr || ':' || portnum FROM dashboards;"))))
    
;;======================================================================
;;  T E S T   L A U N C H I N G   P E R   I T E M   W I T H   H O S T   T Y P E S
;;======================================================================
;; 
;; [hosts]
;; arm cubie01 cubie02
;; x86_64 zeus xena myth01
;; allhosts #{g hosts arm} #{g hosts x86_64}
;; 
;; [host-types]
;; general #MTLOWESTLOAD #{g hosts allhosts}
;; arm     #MTLOWESTLOAD #{g hosts arm}
;; nbgeneral nbjob run JOBCOMMAND -log $MT_LINKTREE/$MT_TARGET/$MT_RUNNAME.$MT_TESTNAME-$MT_ITEM_PATH.lgo
;; 
;; [launchers]
;; envsetup general
;; xor/%/n 4C16G
;; % nbgeneral
;; 
;; [jobtools]

;; # if defined and not "no" flexi-launcher will bypass "launcher" unless no match.

;; flexi-launcher yes  
;; launcher nbfake
;;
(define (common:get-launcher configdat testname itempath)
  (let ((fallback-launcher (configf:lookup configdat "jobtools" "launcher")))
    (if (and (configf:lookup configdat "jobtools" "flexi-launcher") ;; overrides launcher
	     (not (equal? (configf:lookup configdat "jobtools" "flexi-launcher") "no")))
	(let* ((launchers         (hash-table-ref/default configdat "launchers" '())))
	  (if (null? launchers)
	      fallback-launcher
	      (let loop ((hed (car launchers))
			 (tal (cdr launchers)))
		(let ((patt      (car hed))
		      (host-type (cadr hed)))
		  (if (tests:match patt testname itempath)
		      (begin
			(debug:print-info 2 *default-log-port* "Have flexi-launcher match for " testname "/" itempath " = " host-type)
			(let ((launcher (configf:lookup configdat "host-types" host-type)))
			  (if launcher
			      (let* ((launcher-parts (string-split launcher))
				     (launcher-exe   (car launcher-parts)))
				(if (equal? launcher-exe "#MTLOWESTLOAD") ;; this is our special case, we will find the lowest load and craft a nbfake commandline
				    (let ((targ-host (common:get-least-loaded-host (cdr launcher-parts))))
				      (conc "remrun " targ-host))
				    launcher))
			      (begin
				(debug:print-info 0 *default-log-port* "WARNING: no launcher found for host-type " host-type)
				(if (null? tal)
				    fallback-launcher
				    (loop (car tal)(cdr tal)))))))
		      ;; no match, try again
		      (if (null? tal)