Megatest

;;     You should have received a copy of the GNU General Public License
;;     along with Megatest.  If not, see <http://www.gnu.org/licenses/>.

;;======================================================================

(declare (unit servermod))
(declare (uses commonmod))
(declare (uses dbmod))

(module servermod
	*
	
(import scheme chicken data-structures extras)
(import (prefix sqlite3 sqlite3:) posix typed-records srfi-18 srfi-69 format ports srfi-1 matchable)
(import commonmod)
(import dbmod)

;; (use (prefix ulex ulex:))

(include "common_records.scm")

(define (server:make-server-url hostport)
  (if (not hostport)
      #f
      (conc "http://" (car hostport) ":" (cadr hostport))))

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

;; options:
;;
;;  'killservers  - kills all servers
;;  'dejunk       - removes junk records
;;  'adj-testids  - move test-ids into correct ranges
;;  'old2new      - sync megatest.db to /tmp/.../megatest.db and /tmp/.../megatest_ref.db
;;  'new2old      - sync /tmp/.../megatest.db to megatest.db and /tmp/.../megatest_ref.db (and update data_synced)
;;  'closeall     - close all opened dbs
;;  'schema       - attempt to apply schema changes
;;  run-ids: '(1 2 3 ...) or #f (for all)
;;
(define (db:multi-db-sync dbstruct . options)
  ;; (if (not (launch:setup))
  ;;    (debug:print 0 *default-log-port* "ERROR: not able to setup up for megatest.")
  (let* ((mtdb     (dbr:dbstruct-mtdb dbstruct))
	 (tmpdb    (db:get-db dbstruct))
	 (refndb   (dbr:dbstruct-refndb dbstruct))
	 (allow-cleanup #t) ;; (if run-ids #f #t))
	 (servers (server:get-list *toppath*)) ;; (tasks:get-all-servers (db:delay-if-busy tdbdat)))
	 (data-synced 0)) ;; count of changed records (I hope)
    
    (for-each
     (lambda (option)
       
       (case option
	 ;; kill servers
	 ((killservers)
	  (for-each
	   (lambda (server)
	     (match-let (((mod-time host port start-time pid) server))
	       (if (and host pid)
		   (tasks:kill-server host pid))))
	   servers)

          ;; /tmp/bjbarcla/megatest_localdb/fullrun/.nfs.pdx.disks.icf_env_disk001.bjbarcla.gwa.issues.mtdev.matt-bisect.megatest.ext-tests.runs.all.v1.65.1.6524.dev.bb-24-justrollup-f8.rollup.fullrun/megatest.db.lock
          (delete-file* (common:get-sync-lock-filepath))
          )
	 
	 ;; clear out junk records
	 ;;
	 ((dejunk)
	  (db:delay-if-busy mtdb) ;; ok to delay on mtdb
	  (when (file-write-access? (db:dbdat-get-path mtdb)) (db:clean-up mtdb))
	  (db:clean-up tmpdb)
	  (db:clean-up refndb))

	 ;; sync runs, test_meta etc.
	 ;;
	 ((old2new)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f mtdb tmpdb refndb)
	       data-synced)))
	 
	 ;; now ensure all newdb data are synced to megatest.db
	 ;; do not use the run-ids list passed in to the function
	 ;;
	 ((new2old)
	  (set! data-synced
	    (+ (db:sync-tables (db:sync-all-tables-list dbstruct) #f tmpdb refndb mtdb)
	       data-synced)))

	 ((adj-target)
	  (db:adj-target (db:dbdat-get-db mtdb))
	  (db:adj-target (db:dbdat-get-db tmpdb))
	  (db:adj-target (db:dbdat-get-db refndb)))
	 
	 ((schema)
	  (db:patch-schema-maindb (db:dbdat-get-db mtdb))
	  (db:patch-schema-maindb (db:dbdat-get-db tmpdb))
	  (db:patch-schema-maindb (db:dbdat-get-db refndb))
	  (db:patch-schema-rundb  (db:dbdat-get-db mtdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db tmpdb))
	  (db:patch-schema-rundb  (db:dbdat-get-db refndb))))
       
       (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb))
     options)
    data-synced))

(define (db:tmp->megatest.db-sync dbstruct last-update)
  (let* ((mtdb        (dbr:dbstruct-mtdb dbstruct))
	 (tmpdb       (db:get-db dbstruct))
	 (refndb      (dbr:dbstruct-refndb dbstruct))
	 (res         (db:sync-tables (db:sync-all-tables-list dbstruct) last-update tmpdb refndb mtdb)))
    (stack-push! (dbr:dbstruct-dbstack dbstruct) tmpdb)
    res))

;;;; run-ids
;;    if #f use *db-local-sync* : or 'local-sync-flags
;;    if #t use timestamps      : or 'timestamps
;;
;;  NB// no-sync-db is the db handle, not a flag!
;;
(define (db:sync-to-megatest.db dbstruct #!key (no-sync-db #f)) 
  (let* ((start-time         (current-seconds))
	 (last-full-update   (if no-sync-db
				 (db:no-sync-get/default no-sync-db "LAST_FULL_UPDATE" 0)
				 0))
	 (full-sync-needed   (> (- start-time last-full-update) 3600)) ;; every hour do a full sync
	 (last-update        (if full-sync-needed
				 0
				 (if no-sync-db
				     (db:no-sync-get/default no-sync-db "LAST_UPDATE" 0)
				     0))) ;; (or (db:get-var dbstruct "LAST_UPDATE") 0))
	 (sync-needed        (> (- start-time last-update) 6))
	 (res                (if (or sync-needed ;; don't sync if a sync already occurred in the past 6 seconds
				     full-sync-needed)
				 (begin
				   (if no-sync-db
				       (begin
					 (if full-sync-needed (db:no-sync-set no-sync-db "LAST_FULL_UPDATE" start-time))
					 (db:no-sync-set no-sync-db "LAST_UPDATE" start-time)))
				   (db:tmp->megatest.db-sync dbstruct last-update))
				 0))
	 (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")
          (if sync-needed
              (debug:print-info 0 *default-log-port* "Sync of " res " records from newdb to olddb completed in " sync-time " seconds pid="(current-process-id))
              (debug:print-info 0 *default-log-port* "No sync needed, last updated " (- start-time last-update) " seconds ago")))
      res))

;;======================================================================
;; server stuff that operates on the server log files
;;======================================================================

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

;; 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-write-access? 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.")))
		  (directory-exists? (conc areapath "/logs")))
		'()))
	(let* ((server-logs   (glob (conc areapath "/logs/server-*.log")))
	       (num-serv-logs (length server-logs)))
	  (if (null? server-logs)
	      '()
	      (let loop ((hed  (car server-logs))
			 (tal  (cdr server-logs))
			 (res '()))
		(let* ((mod-time  (handle-exceptions
				      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))))
		(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 (car tal)(cdr tal) new-res)))))))))
;; given a path to a server log return: host port startseconds
;;
(define (server:logf-get-start-info logf)
  (let ((rx (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+)"))) ;; SERVER STARTED: host:port AT timesecs
    (handle-exceptions
	exn
	(list #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 rx inl)))
		  (if (not mlst)
		      (if (< lnum 500) ;; give up if more than 500 lines of server log read
			  (loop (read-line)(+ lnum 1))
			  (list #f #f #f))
		      (let ((dat  (cdr mlst)))
			(list (car dat) ;; host
			      (string->number (cadr dat)) ;; port
			      (string->number (caddr dat))))))
		(list #f #f #f))))))))

(define (server:get-num-alive srvlst)
  (let ((num-alive 0))
    (for-each
     (lambda (server)
       (match-let (((mod-time host port start-time 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
				     (< (- now start-time)       
					(+ (- (string->number (or (configf:lookup *configdat* "server" "runtime") "3600"))
					      180)
					   (random 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 (random len)))
	  (list-ref srvrs idx))
	#f)))


(define (server:record->url servr)
  (match-let (((mod-time host port start-time pid)
	       servr))
    (if (and host port)
	(conc host ":" port)
	#f)))


;; timeout is hms string: 1h 5m 3s, default is 1 minute
;;
(define (server:expiration-timeout)
  (let ((tmo (configf:lookup *configdat* "server" "timeout")))
    (if (and (string? tmo)
	     (common:hms-string->seconds tmo)) ;; BUG: hms-string->seconds is broken, if given "10" returns 0. Also, it doesn't belong in this logic unless the string->number is changed below
        (* 3600 (string->number tmo))
	60)))

;; ping the given server
;;
(define (server:check-server server-record)
  (let* ((server-url (server:record->url server-record))
         (res        (case *transport-type*
                       ((http)(server:ping server-url))
                       ;; ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
                       )))
    (if res
        server-url
	#f)))

;; no longer care if multiple servers are started by accident. older servers will drop off in time.
;;
(define (server:check-if-running areapath) ;;  #!key (numservers "2"))
  (let* ((ns            (server:get-num-servers))
	 (servers       (server:get-best (server:get-list areapath))))
    ;; (print "servers: " servers " ns: " ns)
    (if (or (and servers
		 (null? servers))
	    (not servers)
	    (and (list? servers)
		 (< (length servers) (random ns)))) ;; somewhere between 0 and numservers
        #f
        (let loop ((hed (car servers))
                   (tal (cdr servers)))
          (let ((res (server:check-server hed)))
            (if res
                res
                (if (null? tal)
                    #f
                    (loop (car tal)(cdr tal)))))))))

;;======================================================================
;; P K T S   S T U F F 
;;======================================================================

;; ???

;;======================================================================

		(server:kind-run areapath))
	    (thread-sleep! 5)
	    (loop (server:check-if-running areapath)
		  (+ try-num 1)))))))

(define server:try-running server:run) ;; there is no more per-run servers ;; REMOVE ME. BUG.


































(define (server:kill servr)
  (match-let (((mod-time hostname port start-time pid)
	       servr))
    (tasks:kill-server hostname pid)))

;; called in megatest.scm, host-port is string hostname:port
;;

(define (server:login toppath)
  (lambda (toppath)
    (set! *db-last-access* (current-seconds)) ;; might not be needed.
    (if (equal? *toppath* toppath)
	#t
	#f)))























;; (define server:sync-lock-token "SERVER_SYNC_LOCK")
;; (define (server:release-sync-lock)
;;   (db:no-sync-del! *no-sync-db* server:sync-lock-token))
;; (define (server:have-sync-lock?)
;;   (let* ((have-lock-pair (db:no-sync-get-lock *no-sync-db* server:sync-lock-token))
;;          (have-lock?     (car have-lock-pair))
;;          (lock-time      (cdr have-lock-pair))