Megatest

(declare (uses debugprint))
(declare (uses stml2))
(declare (uses pkts))
(declare (uses processmod))
(declare (uses mtargs))
(declare (uses configfmod))

;; odd but it works?
(declare (uses itemsmod))

(module commonmod
	*
	
(import scheme
	chicken.base
	chicken.condition
	chicken.file

	mtver
	debugprint
	stml2
	pkts
	processmod
	(prefix mtargs args:)
	configfmod

	itemsmod
	)

;;======================================================================
;; CONTENTS
;;
;;  config file utils
;;  misc conversion, data manipulation functions

(define home (getenv "HOME"))
(define user (getenv "USER"))

;; Globals
;;
(define  *server-loop-heart-beat* (current-seconds))

;; copied from egg call-with-environment-variables
;;
(define (call-with-environment-variables variables thunk)
  ;; @("Sets up environment variable via dynamic-wind which are taken down after thunk."
  ;;   (variables "An alist of the form {{'((\"var\" . \"value\") ...)}}")
  ;;   (thunk "The thunk to execute with a modified environment"))
  (let ((pre-existing-variables
         (map (lambda (var-value)
                (let ((var (car var-value)))
                  (cons var (get-environment-variable var))))
              variables)))
    (dynamic-wind
        (lambda () (void))
        (lambda ()
;;           (use posix)
          (for-each (lambda (var-value)
                      (setenv (car var-value) (cdr var-value)))
            variables)
          (thunk))
        (lambda ()
          (for-each (lambda (var-value)
                      (let ((var (car var-value))
                            (value (cdr var-value)))
                        (if value
                            (setenv var value)
                            (unsetenv var))))
            pre-existing-variables)))))
     
;; dot-locking egg seems not to work, using this for now
;; if lock is older than expire-time then remove it and try again
;; to get the lock
;;
(define (common:simple-file-lock fname #!key (expire-time 300))
  (let ((fmod-time (handle-exceptions
		       ext

(define *user-hash-data* (make-hash-table))

(define *db-keys* #f)

(define *pkts-info*    (make-hash-table)) ;; store stuff like the last parent here
(define *configinfo*   #f)   ;; raw results from setup, includes toppath and table from megatest.config
(define *runconfigdat* #f)   ;; run configs data
;; (define *configdat*    #f)   ;; megatest.config data ==> moved to configfmod
(define *configstatus* #f)   ;; status of data; 'fulldata : all processing done, #f : no data yet, 'partialdata : partial read done
(define *toppath*      #f)
(define *already-seen-runconfig-info* #f)

(define *test-meta-updated* (make-hash-table))
(define *globalexitstatus*  0) ;; attempt to work around possible thread issues
(define *passnum*           0) ;; when running track calls to run-tests or similar

		    (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)))

(define (launch:is-test-alive host pid)
  (if (and host pid (not (equal? host "n/a")))
      (let* ((cmd (conc "ssh " host " pstree -A " pid))
	     (output (with-input-from-pipe cmd read-lines)))
	(debug:print 2 *default-log-port* "Running " cmd " received " output)
	(if (eq? (length output) 0)
	   #f
	   #t))
      #t))

;; common:get-host-info was here
;; common:update-host-loads-table
;; common:get-least-loaded-host
;; common:wait-for-homehost-load

(define (common:get-num-cpus remote-host)

     (else
      (if (> num-tries 0) ;; should be "num-tries-left". 
	  (if (common:low-noise-print 30 (conc (round effective-load) "-load-acceptable-" effective-host))
	      (debug:print 0 *default-log-port* "Load on " effective-host " is acceptable at effective normalized load of "
			   normalized-effective-load "  continuing."))
	  (debug:print 0 *default-log-port* "Load on " effective-host ", "
		       first" could not be retrieved. Giving up and continuing."))))))

;;======================================================================
;; server process management
;;======================================================================

;; no elegance here ...
;;
(define (tasks:kill-server hostname pid #!key (kill-switch ""))
  (debug:print-info 0 *default-log-port* "Attempting to kill server process " pid " on host " hostname)
  (setenv "TARGETHOST" hostname)
  (let* ((logdir (if (directory-exists? "logs")
                    "logs/"
                    ""))
         (logfile (if logdir (conc "logs/server-"pid"-"hostname".log") #f))
         (gzfile  (if logfile (conc logfile ".gz"))))
    (setenv "TARGETHOST_LOGF" (conc logdir "server-kills.log"))

    (system (conc "nbfake kill "kill-switch" "pid))

    (when logfile
      (thread-sleep! 0.5)
      (if (file-exists? gzfile) (delete-file gzfile))
      (system (conc "gzip " logfile))
      
      (unsetenv "TARGETHOST_LOGF")
      (unsetenv "TARGETHOST"))))

(define (server:get-logs-list area-path)
  (let* ((server-logs-cmd  (conc "grep -iL exiting " area-path "/logs/server-*-*.log"))
	 (server-logs   (string-split (string-chomp (with-input-from-pipe server-logs-cmd read-string)))))
    server-logs))
  
;; get a list of servers with all relevant data
;; ( mod-time host port start-time pid )
;;
(define (server:get-list areapath #!key (limit #f))
  (let ((fname-rx    (regexp "^(|.*/)server-(\\d+)-(\\S+).log$"))
	(day-seconds (* 24 60 60)))
    ;; if the directory exists continue to get the list
    ;; otherwise attempt to create the logs dir and then
    ;; continue
    (if (if (directory-exists? (conc areapath "/logs"))
	    '()
	    (if (file-writable? areapath)
		(begin
		  (condition-case
		   (create-directory (conc areapath "/logs") #t)
		   (exn (i/o file)(debug:print 0 *default-log-port* "ERROR: Cannot create directory at " (conc areapath "/logs")))
		   (exn ()(debug:print 0 *default-log-port* "ERROR: Unknown error attemtping to get server list. exn=" exn)))
		  (directory-exists? (conc areapath "/logs")))
		'()))

        ;; Get the list of server logs that do not contain "exiting". Ignore logs for servers that have exited.
	(let* ((server-logs   (server:get-logs-list areapath))
	       (num-serv-logs (length server-logs)))
	  (if (or (null? server-logs) (= num-serv-logs 0))
              (let ()
                 (debug:print 1  *default-log-port* "There are no servers running")
	         '()
              )
	      (let loop ((hed  (string-chomp (car server-logs)))
			 (tal  (cdr server-logs))
			 (res '()))
		(let* ((mod-time  (handle-exceptions
				   exn
				   (begin
				     (debug:print 0 *default-log-port* "failed to get modification time on " hed ", exn=" exn)
				     (current-seconds)) ;; 0
				   (file-modification-time hed))) ;; default to *very* old so log gets ignored if deleted
		       (down-time (- (current-seconds) mod-time))
		       (serv-dat  (if (or (< num-serv-logs 10)
				  	  (< down-time 900)) ;; day-seconds))
				      (server:logf-get-start-info hed)
				      '())) ;; don't waste time processing server files not touched in the 15 minutes if there are more than ten servers to look at
		       (serv-rec (cons mod-time serv-dat))
		       (fmatch   (string-match fname-rx hed))
		       (pid      (if fmatch (string->number (list-ref fmatch 2)) #f))
		       (new-res  (if (null? serv-dat)
				     res
				     (cons (append serv-rec (list pid)) res)))) ;; any changes to number of elements in new-res will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 
		  (if (null? tal)
		      (if (and limit
			       (> (length new-res) limit))
			  new-res ;; (take new-res limit)  <= need intelligent sorting before this will work
			  new-res)
		      (loop (string-chomp (car tal)) (cdr tal) new-res)))))))))

(define (server:get-num-alive srvlst)
  (let ((num-alive 0))
    (for-each
     (lambda (server)
        (handle-exceptions
          exn
         (begin 
          (debug:print-info 0 *default-log-port*  "Unable to get server start-time and/or mod-time from " server ", exn=" exn))
       (match-let (((mod-time host port start-time server-id pid)
		    server))
	 (let* ((uptime  (- (current-seconds) mod-time))
		(runtime (if start-time
			     (- mod-time start-time)
			     0)))
	   (if (< uptime 5)(set! num-alive (+ num-alive 1)))))))
     srvlst)
    num-alive))

;; given a list of servers get a list of valid servers, i.e. at least
;; 10 seconds old, has started and is less than 1 hour old and is
;; active (i.e. mod-time < 10 seconds
;;
;; mod-time host port start-time pid
;;
;; sort by start-time descending. I.e. get the oldest first. Young servers will thus drop off
;; and servers should stick around for about two hours or so.
;;
(define (server:get-best srvlst)
  (let* ((nums (server:get-num-servers))
	 (now  (current-seconds))
	 (slst (sort
		(filter (lambda (rec)
			  (if (and (list? rec)
				   (> (length rec) 2))
			      (let ((start-time (list-ref rec 3))
				    (mod-time   (list-ref rec 0)))
				;; (print "start-time: " start-time " mod-time: " mod-time)
				(and start-time mod-time
				     (> (- now start-time) 0)    ;; been running at least 0 seconds
				     (< (- now mod-time)   16)   ;; still alive - file touched in last 16 seconds
				     (or (not (configf:lookup *configdat* "server" "runtime")) ;; skip if not set
					 (< (- now start-time)       
					    (+ (- (string->number (configf:lookup *configdat* "server" "runtime"))
					       180)
					    (pseudo-random-integer 360)))) ;; under one hour running time +/- 180
				     ))
			      #f))
			srvlst)
		(lambda (a b)
		  (< (list-ref a 3)
		     (list-ref b 3))))))
    (if (> (length slst) nums)
	(take slst nums)
	slst)))

(define (server:get-first-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and srvrs
	     (not (null? srvrs)))
	(car srvrs)
	#f)))

(define (server:get-rand-best areapath)
  (let ((srvrs (server:get-best (server:get-list areapath))))
    (if (and (list? srvrs)
	     (not (null? srvrs)))
	(let* ((len (length srvrs))
	       (idx (pseudo-random-integer len)))
	  (list-ref srvrs idx))
	#f)))

(define (server:record->id servr)
  (handle-exceptions
   exn
   (begin 
     (debug:print-info 0 *default-log-port*  "Unable to get server id from " servr ", exn=" exn)     
   #f)
  (match-let (((mod-time host port start-time server-id pid)
	       servr))
    (if server-id
	server-id
	#f))))

(define (server:get-num-servers #!key (numservers 2))
  (let ((ns (string->number
	     (or (configf:lookup *configdat* "server" "numservers") "notanumber"))))
    (or ns numservers)))

;; given a path to a server log return: host port startseconds
;; any changes to number of elements returned by this fuction will dirctly affect server:record->url,server:record->id,server:kill,server:get-num-alive which uses match let 

(define (server:logf-get-start-info logf)
  (let ((server-rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+)")) ;; SERVER STARTED: host:port AT timesecs server id
        (dbprep-rx (regexp "^SERVER: dbprep"))
        (dbprep-found 0)) 
    (handle-exceptions
	exn
      (begin
	(debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", exn=" exn)
	(list #f #f #f #f)) ;; no idea what went wrong, call it a bad server
      (with-input-from-file
	  logf
	(lambda ()
	  (let loop ((inl  (read-line))
		     (lnum 0))
	    (if (not (eof-object? inl))
		(let ((mlst (string-match server-rx inl))
                      (dbprep (string-match dbprep-rx inl))
                      )
                  (if dbprep
                    (set! dbprep-found 1)
                  )
		  (if (not mlst)
		      (if (< lnum 500) ;; give up if more than 500 lines of server log read
			  (loop (read-line)(+ lnum 1))
			  (begin 
                           (debug:print-info 0 *default-log-port* "Unable to get server info from first 500 lines of " logf )
                           (list #f #f #f #f)))
		      (let ((dat  (cdr mlst)))
			(list (car dat) ;; host
			      (string->number (cadr dat)) ;; port
			      (string->number (caddr dat))
                              (cadr (cddr dat))))))
                (begin 
                   (if dbprep-found
                      (begin
                         (debug:print-info 0 *default-log-port* "Server is in dbprep at " (current-seconds))
                         (thread-sleep! 25)
                      )
                      (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (current-seconds))
                   )
		   (list #f #f #f #f)))))))))


;;======================================================================
;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
;;
;; (define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f)(force-maxload #f)(num-tries 5))
;;   (let* ((loadavg (common:get-cpu-load remote-host))
;; 	 (numcpus (if (<= 1 numcpus-in) ;; not possible to have zero.  If we get 1, it's possible that we got the previous default, and we should check again

;;        (debug:print-info 0 *default-log-port* "common-telemetry-log closure failure")
;;        )
;;      (begin
;;        (define *common:telemetry-log-state* 'closed)
;;        (udp-close-socket *common:telemetry-log-socket*)
;;        (set! *common:telemetry-log-socket* #f)))))

;;======================================================================
;; test patt stuff
;;======================================================================

;; make a query (fieldname like 'patt1' OR fieldname 
(define (db:patt->like fieldname pattstr #!key (comparator " OR "))
  (let ((patts (if (string? pattstr)
		   (string-split pattstr ",")
		   '("%"))))
    (string-intersperse (map (lambda (patt)
			       (let ((wildtype (if (substring-index "%" patt) "LIKE" "GLOB")))
				 (conc fieldname " " wildtype " '" patt "'")))
			     (if (null? patts)
				 '("")
				 patts))
			comparator)))

(define *glob-like-match-cache* (make-hash-table))
(define (tests:cache-regexp str-in flag)
  (let* ((key (conc str-in flag)))
    (or (hash-table-ref/default *glob-like-match-cache* key #f)
	(let* ((newrx (regexp str-in flag)))
	  (hash-table-set! *glob-like-match-cache* key newrx)
	  newrx))))

;; tests:glob-like-match 
(define (tests:glob-like-match patt str) 
  (let* ((like     (substring-index "%" patt))
	 (notpatt  (equal? (substring-index "~" patt) 0))
	 (newpatt  (if notpatt (substring patt 1) patt))
	 (finpatt  (if like
		       (string-substitute (regexp "%") ".*" newpatt #f)
		       (string-substitute (regexp "\\*") ".*" newpatt #f)))
	 (rx       (tests:cache-regexp finpatt (if like #t #f)))
	 (res      (string-match rx str)))
    (if notpatt (not res) res)))

;; if itempath is #f then look only at the testname part
;;
(define (tests:match patterns testname itempath #!key (required '()))
  (if (string? patterns)
      (let ((patts (append (string-split patterns ",") required)))
	(if (null? patts) ;;; no pattern(s) means no match
	    #f
	    (let loop ((patt (car patts))
		       (tal  (cdr patts)))
	      ;; (print "loop: patt: " patt ", tal " tal)
	      (if (string=? patt "")
		  #f ;; nothing ever matches empty string - policy
		  (let* ((patt-parts (string-match (regexp "^([^\\/]*)(\\/(.*)|)$") patt))
			 (test-patt  (cadr patt-parts))
			 (item-patt  (cadddr patt-parts)))
		    ;; special case: test vs. test/
		    ;;   test  => "test" "%"
		    ;;   test/ => "test" ""
		    (if (and (not (substring-index "/" patt)) ;; no slash in the original
			     (or (not item-patt)
				 (equal? item-patt "")))      ;; should always be true that item-patt is ""
			(set! item-patt "%"))
		    ;; (print "tests:match => patt-parts: " patt-parts ", test-patt: " test-patt ", item-patt: " item-patt)
		    (if (and (tests:glob-like-match test-patt testname)
			     (or (not itempath)
				 (tests:glob-like-match (if item-patt item-patt "") itempath)))
			#t
			(if (null? tal)
			    #f
			    (loop (car tal)(cdr tal)))))))))))

;; if itempath is #f then look only at the testname part
;;
(define (tests:match->sqlqry patterns)
  (if (string? patterns)
      (let ((patts (string-split patterns ",")))
	(if (null? patts) ;;; no pattern(s) means no match, we will do no query
	    #f
	    (let loop ((patt (car patts))
		       (tal  (cdr patts))
		       (res  '()))
	      ;; (print "loop: patt: " patt ", tal " tal)
	      (let* ((patt-parts (string-match (regexp "^([^\\/]*)(\\/(.*)|)$") patt))
		     (test-patt  (cadr patt-parts))
		     (item-patt  (cadddr patt-parts))
		     (test-qry   (db:patt->like "testname" test-patt))
		     (item-qry   (db:patt->like "item_path" item-patt))
		     (qry        (conc "(" test-qry " AND " item-qry ")")))
		;; (print "tests:match => patt-parts: " patt-parts ", test-patt: " test-patt ", item-patt: " item-patt)
		(if (null? tal)
		    (string-intersperse (append (reverse res)(list qry)) " OR ")
		    (loop (car tal)(cdr tal)(cons qry res)))))))
      #f))

;; itemmap is a list of testname patterns to maps
;;     test1 .*/bar/(\d+) foo/\1
;;     %     foo/([^/]+)  \1/bar
;;
;; # NOTE: the line with the single % could be the result of
;; #       itemmap entry in requirements (legacy). The itemmap
;; #       requirements entry is deprecated
;;
(define (tests:get-itemmaps tconfig)
  (let ((base-itemmap  (configf:lookup tconfig "requirements" "itemmap"))
	(itemmap-table (configf:get-section tconfig "itemmap")))
    (append (if base-itemmap
		(list (list "%" base-itemmap))
		'())
	    (if itemmap-table
		itemmap-table
		'()))))

;; given a list of itemmaps (testname . map), return the first match
;;
(define (tests:lookup-itemmap itemmaps testname)
  (let ((best-matches (filter (lambda (itemmap)
				(tests:match (car itemmap) testname #f))
			      itemmaps)))
    (if (null? best-matches)
	#f
	(let ((res (car best-matches)))
	  ;; (debug:print 0 *default-log-port* "res=" res)
	  (cond
	   ((string? res) res) ;;; FIX THE ROOT CAUSE HERE ....
	   ((null? res)   #f)
	   ((string? (cdr res)) (cdr res))  ;; it is a pair
	   ((string? (cadr res))(cadr res)) ;; it is a list
	   (else cadr res))))))

;; return items given config
;;
(define (tests:get-items tconfig)
  (let ((items      (hash-table-ref/default tconfig "items" #f)) ;; items 4
	(itemstable (hash-table-ref/default tconfig "itemstable" #f))) 
    ;; if either items or items table is a proc return it so test running
    ;; process can know to call items:get-items-from-config
    ;; if either is a list and none is a proc go ahead and call get-items
    ;; otherwise return #f - this is not an iterated test
    (cond
     ((procedure? items)      
      (debug:print-info 4 *default-log-port* "items is a procedure, will calc later")
      items)            ;; calc later
     ((procedure? itemstable)
      (debug:print-info 4 *default-log-port* "itemstable is a procedure, will calc later")
      itemstable)       ;; calc later
     ((filter (lambda (x)
		(let ((val (car x)))
		  (if (procedure? val) val #f)))
	      (append (if (list? items) items '())
		      (if (list? itemstable) itemstable '())))
      'have-procedure)
     ((or (list? items)(list? itemstable)) ;; calc now
      (debug:print-info 4 *default-log-port* "items and itemstable are lists, calc now\n"
			"    items: " items " itemstable: " itemstable)
      (items:get-items-from-config tconfig))
     (else #f))))                           ;; not iterated

(define (tests:get-tests-search-path cfgdat)
  (let ((paths (let ((section (if cfgdat
				  (configf:get-section cfgdat "tests-paths")
				  #f)))
		 (if section
		     (map cadr section)
		     '()))))
    (filter (lambda (d)
	      (if (directory-exists? d)
		  d
		  (begin
		    ;; (if (common:low-noise-print 60 "tests:get-tests-search-path" d)
		    ;;	(debug:print 0 *default-log-port* "WARNING: problem with directory " d ", dropping it from tests path"))
		    #f)))
	    (append paths (list (conc *toppath* "/tests"))))))

)