Megatest

util.scm at [366547935e]
Login

File util.scm artifact 8fad541d3c part of check-in 366547935e


;;======================================================================
;; Copyright 2018, Matthew Welland.
;; 
;; This file is part of Megatest.
;; 
;;     Megatest is free software: you can redistribute it and/or modify
;;     it under the terms of the GNU General Public License as published by
;;     the Free Software Foundation, either version 3 of the License, or
;;     (at your option) any later version.
;; 
;;     Megatest is distributed in the hope that it will be useful,
;;     but WITHOUT ANY WARRANTY; without even the implied warranty of
;;     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;;     GNU General Public License for more details.
;; 
;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.
;;
;;======================================================================

;;======================================================================
;; Tests
;;======================================================================

(declare (unit util))

(declare (uses common))

(module util
    *

(import chicken scheme data-structures extras srfi-13 ports )
  
  
;;======================================================================
;; L O C K I N G   M E C H A N I S M S 
;;======================================================================

;; faux-lock is deprecated. Please use simple-lock below
;;
(define (common:faux-lock keyname #!key (wait-time 8)(allow-lock-steal #t))
  (if (rmt:no-sync-get/default keyname #f) ;; do not be tempted to compare to pid. locking is a one-shot action, if already locked for this pid it doesn't actually count
      (if (> wait-time 0)
	  (begin
	    (thread-sleep! 1)
	    (if (eq? wait-time 1) ;; only one second left, steal the lock
		(begin
		  (debug:print-info 0 *default-log-port* "stealing lock for " keyname)
		  (common:faux-unlock keyname force: #t)))
	    (common:faux-lock keyname wait-time: (- wait-time 1)))
	  #f)
      (begin
        (rmt:no-sync-set keyname (conc (current-process-id)))
        (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))))

(define (common:faux-unlock keyname #!key (force #f))
  (if (or force (equal? (conc (current-process-id)) (conc (rmt:no-sync-get/default keyname #f))))
      (begin
        (if (rmt:no-sync-get/default keyname #f) (rmt:no-sync-del! keyname))
        #t)
      #f))

;; simple lock. improve and converge on this one.
;;
(define (common:simple-lock keyname)
  (rmt:no-sync-get-lock keyname))  
  
(define (common:run-a-command cmd #!key (with-vars #f))
  (let* ((pre-cmd  (dtests:get-pre-command))
         (post-cmd (dtests:get-post-command))
         (fullcmd  (if (or pre-cmd post-cmd)
                       (conc pre-cmd cmd post-cmd)
                       (conc "viewscreen " cmd))))
    (debug:print-info 02 *default-log-port* "Running command: " fullcmd)
    (if with-vars
        (common:without-vars cmd)
        (common:without-vars fullcmd "MT_.*"))))
		  
;; 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)) ;; if this host happens to have been recently used by a test reuse the load data
         (load                             (car loadinfo))
         (load-sample-time                 (cdr loadinfo))
         (load-sample-age                  (- (current-seconds) load-sample-time))
         (loadinfo-timeout-seconds         6) ;; this was 20 seconds, seems way too lax. Switch to 6 seconds
         (host-last-update-timeout-seconds 4)
         (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)))) ;; this is cheaper than you might think. get-normalized-cpu-load is cached for up to 5 seconds
     (else
      (list #f 0 -1) ;; bad host, don't use!
      ))))

(define (get-uname . params)
  (let* ((uname-res (process:cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
	 (uname #f))
    (if (null? (car uname-res))
	"unknown"
	(caar uname-res))))

(define (get-unix-df path)
  (let* ((df-results (process:cmd-run->list (conc "df " path)))
	 (space-rx   (regexp "([0-9]+)\\s+([0-9]+)%"))
	 (freespc    #f))
    ;; (write df-results)
    (for-each (lambda (l)
		(let ((match (string-search space-rx l)))
		  (if match 
		      (let ((newval (string->number (cadr match))))
			(if (number? newval)
			    (set! freespc newval))))))
	      (car df-results))
    freespc))

;; given path get free space, allows override in [setup]
;; with free-space-script /path/to/some/script.sh
;;
(define (get-df path)
  (if (configf:lookup *configdat* "setup" "free-space-script")
      (with-input-from-pipe 
       (conc (configf:lookup *configdat* "setup" "free-space-script") " " path)
       (lambda ()
	 (let ((res (read-line)))
	   (if (string? res)
	       (string->number res)))))
      (get-unix-df path)))

(define (std-exit-procedure)
  (on-exit (lambda () 0))
  ;;(debug:print-info 13 *default-log-port* "std-exit-procedure called; *time-to-exit*="*time-to-exit*)
  (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)))))
                              (http-client#close-all-connections!)
                              ;; (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)

;; from metadat lookup MEGATEST_VERSION
;;
(define (common:get-last-run-version) ;; RADT => How does this work in send-receive function??; assume it is the value saved in some DB
  (rmt:get-var "MEGATEST_VERSION"))

(define (common:set-last-run-version)
  (rmt:set-var "MEGATEST_VERSION" (common:version-signature)))

;;======================================================================
;; D I S K   S P A C E 
;;======================================================================

(define (common:get-disk-space-used fpath)
  (with-input-from-pipe (conc "/usr/bin/du -s " fpath) read))

(define (common:check-space-in-dir dirpath required)
  (let* ((dbspace  (if (directory? dirpath)
		       (get-df dirpath)
		       0)))
    (list (> dbspace required)
	  dbspace
	  required
	  dirpath)))

;; check space in dbdir and in megatest dir
;; returns: ok/not dbspace required-space
;;
(define (common:check-db-dir-space)
  (let* ((required (string->number 
		    (or (configf:lookup *configdat* "setup" "dbdir-space-required")
			"100000")))
	 (dbdir    (common:get-db-tmp-area)) ;; (db:get-dbdir))
	 (tdbspace (common:check-space-in-dir dbdir required))
	 (mdbspace (common:check-space-in-dir *toppath* required)))
    (sort (list tdbspace mdbspace) (lambda (a b)
				     (< (cadr a)(cadr b))))))
    
;; check available space in dbdir, exit if insufficient
;;
(define (common:check-db-dir-and-exit-if-insufficient)
  (let* ((spacedat (car (common:check-db-dir-space))) ;; look only at worst for now
	 (is-ok    (car spacedat))
	 (dbspace  (cadr spacedat))
	 (required (caddr spacedat))
	 (dbdir    (cadddr spacedat)))
    (if (not is-ok)
	(begin
	  (debug:print-error 0 *default-log-port* "Insufficient space in " dbdir ", require " required ", have " dbspace  ", exiting now.")
	  (exit 1)))))

;; paths is list of lists ((name path) ... )
;;
(define (common:get-disk-with-most-free-space disks minsize)
  (let ((best     #f)
	(bestsize 0))
    (for-each 
     (lambda (disk-num)
       (let* ((dirpath    (cadr (assoc disk-num disks)))
	      (freespc    (cond
			   ((not (directory? dirpath))
			    (if (common:low-noise-print 300 "disks not a dir " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a directory - ignoring it."))
			    -1)
			   ((not (file-write-access? dirpath))
			    (if (common:low-noise-print 300 "disks not writeable " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not writeable - ignoring it."))
			    -1)
			   ((not (eq? (string-ref dirpath 0) #\/))
			    (if (common:low-noise-print 300 "disks not a proper path " disk-num)
				(debug:print 0 *default-log-port* "WARNING: disk " disk-num " at path \"" dirpath "\" is not a fully qualified path - ignoring it."))
			    -1)
			   (else
			    (get-df dirpath)))))
	 (if (> freespc bestsize)
	     (begin
	       (set! best     (cons disk-num dirpath))
	       (set! bestsize freespc)))))
     (map car disks))
    (if (and best (> bestsize minsize))
	best
	#f))) ;; #f means no disk candidate found

;; see defstruct host at top of file.
;;    host: reachable last-update last-used last-cpuload
;;
(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)))

;; go through the hosts from least recently used to most recently used, pick the first that meets the load criteral from the
;; [host-rules] section.
;;
(define (common:get-least-loaded-host hosts-raw host-type configdat)
  (let* ((rdat       (configf:lookup configdat "host-rules" host-type))
	 (rules      (common:val->alist (or rdat "") convert: #t))   ;; maxnload, maxnjobs, maxjobrate
	 (maxnload   (common:alist-ref/default 'maxnload rules 1.5)) ;; max normalized load
	 (maxnjobs   (common:alist-ref/default 'maxnjobs rules 1.5)) ;; max normalized number of jobs
	 (maxjobrate (common:alist-ref/default 'maxjobrate rules (/ 1 6))) ;; max rate of submitting jobs to a given host in jobs/second
	 (hosts      (filter (lambda (x)
			       (string-match (regexp "^\\S+$") x))
			     hosts-raw))
         ;; (best-host #f)
	 (get-rec    (lambda (hostname)
		       ;; (print "get-rec hostname=" hostname)
		       (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)))))
         (best-load 99999)
         (curr-time (current-seconds))
	 (get-hosts-sorted (lambda (hosts)
			     (sort hosts (lambda (a b)
					   (let ((a-rec (get-rec a))
						 (b-rec (get-rec b)))
					     ;; (print "a=" a " a-rec=" a-rec " host-last-used=" (host-last-used a-rec))
					     ;; (print "b=" b " b-rec=" b-rec " host-last-used=" (host-last-used b-rec))
					     (< (host-last-used a-rec)
						(host-last-used b-rec))))))))
    (debug:print 0 *default-log-port* "INFO: hosts-sorted=" (get-hosts-sorted hosts))
    (if (null? hosts)
	#f ;; no hosts to select from. All done and giving up now.
	(let ((hosts-sorted (get-hosts-sorted hosts)))
	  (common:update-host-loads-table hosts)
	  (let loop ((hostname  (car hosts-sorted))
		     (tal       (cdr hosts-sorted))
		     (best-host #f))
	    (let* ((rec       (get-rec hostname))
		   (reachable (host-reachable     rec))
		   (load      (host-last-cpuload  rec))
		   (last-used (host-last-used     rec))
		   (delta     (- curr-time last-used))
		   (job-rate  (if (> delta 0)
				  (/ 1 delta)
				  999)) ;; jobs per second
		   (new-best  
		    (cond
		     ((not reachable)
		      (debug:print 0 *default-log-port* "Skipping host " hostname " as it cannot be reached.")
		      best-host)
		     ((and (< load maxnload)        ;; load is acceptable
			   (< job-rate maxjobrate)) ;; job rate is acceptable
		      (set! best-load load)
		      hostname)
		     (else best-host))))
	      (debug:print 0 *default-log-port* "INFO: Trying host " hostname " with load " load ", last used " delta " seconds ago, with job-rate " job-rate " for running a test." )
	      (if new-best
		  (begin ;; found a host, return it
		    (debug:print 0 *default-log-port* "INFO: Found host: " new-best " load: " load " last-used: " delta " seconds ago, with job-rate: " job-rate)
		    (host-last-used-set! rec curr-time)
		    new-best)
		  (if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))

(define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f))
  (let* ((loadavg (common:get-cpu-load remote-host))
	 (numcpus (if (< 1 numcpus-in) ;; not possible
		      (common:get-num-cpus remote-host)
		      numcpus-in))
	 (maxload (max maxload-in 0.5)) ;; so maxload must be greater than 0.5 for now BUG - FIXME?
	 (first   (car loadavg))
	 (next    (cadr loadavg))
	 (adjload (* maxload (max 1 numcpus))) ;; possible bug where numcpus (or could be maxload) is zero, crude fallback is to at least use 1
	 (loadjmp (- first next))
         (adjwait (+ (random 10)(/ (- 1000 count) 10) waitdelay))) ;; add some randomness to the time to break any alignment where netbatch dumps many jobs to machines simultaneously
    (cond
     ((and (> first adjload)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "server start delayed " adjwait " seconds due to load " first " exceeding max of " adjload " on server " (or remote-host (get-host-name)) " (normalized load-limit: " maxload ") " (if msg msg ""))
      (thread-sleep! adjwait)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
     ((and (> loadjmp numcpus)
	   (> count 0))
      (debug:print-info 0 *default-log-port* "waiting " adjwait " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
      (thread-sleep! adjwait)
      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host)))))

(define (common:wait-for-homehost-load maxload msg)
  (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                     #f
                     (common:get-homehost)))
         (hh     (if hh-dat (car hh-dat) #f))
         (numcpus (common:get-num-cpus hh)))
    (common:wait-for-normalized-load maxload msg hh)))

(define (common:get-num-cpus remote-host)
  (let* ((actual-host (or remote-host (get-host-name))))
    (or (common:get-cached-info actual-host "num-cpus" age: 86400) ;; hosts had better not be changing the number of cpus too often!
	(let* ((proc   (lambda ()
			 (let loop ((numcpu 0)
				    (inl    (read-line)))
			   (if (eof-object? inl)
			       (begin
				 (common:write-cached-info remote-host "num-cpus" numcpu)
				 numcpu)
			       (loop (if (string-match "^processor\\s+:\\s+\\d+$" inl)
					 (+ numcpu 1)
					 numcpu)
				     (read-line))))))
	       (result (if remote-host
			   (with-input-from-pipe 
			       (conc "ssh " remote-host " cat /proc/cpuinfo")
			     proc)
			   (with-input-from-file "/proc/cpuinfo" proc))))
	  (common:write-cached-info actual-host "num-cpus" result)
	  result))))

;; wait for normalized cpu load to drop below maxload
;;
(define (common:wait-for-normalized-load maxload msg remote-host)
  (let ((num-cpus (common:get-num-cpus remote-host)))
    (common:wait-for-cpuload maxload num-cpus 15 msg: msg remote-host: remote-host)))


)