Ganglia Job Monarch Files

	LEGEND	f: fixed - c: changed - a: added - r: removed

1.1.2:

    jobmond)

        c: job info is now escaped: special characters (from for example job 
           names) are escaped to prevent (XML) errors
        f: no longer eats up file descriptors and crashing after running out of
           file descriptors
        f: no longer crash after Torque/PBS unavailability issue under certain
           conditions

1.1.1:

    packaging)

        f: correctly set the JOBARCHIVE_RRDS in both jobarchived.conf and
           web/conf.php.in
        f: debian init.d script names in post/pre pkg corrected to new name
        f: debian postrm was incorrectly trying redhat conditional restart

    web)

        c: column nodes renamed to: hosts
        a: sorting by hosts now implemented

    jobarchived)

        f: now properly exits on fatal xml errors
        f: prevent exception to occur when no timed out jobs are found during
           Housekeeping

    jobmond)

        f: BATCH_HOST_TRANSLATE no longer required in jobmond.conf

1.1:

    web)

        a: archive search now has "include running jobs" option
        c: rewritten short versus FQDN hostname detection: now works properly 
           with ganglia hosts not using FQDN hostnames
        f: display of xml parsetime for overview. no longer display parsetime 
           for archive (no parsing done)
        f: down/offline nodes are now properly marked in cluster image again
        f: bug where "Unavailable" row would not be shown in overview summary 
           table

    packaging)

        c: completely redone and rewritten by Olivier Lahaye - thanks!

    jobmond)

        a: now supports SLURM Workload Manager!
        a: warning if connecting to remote BATCH_SERVER is not supported by 
           selected BATCH_API
        f: bug where incorrect commandline option would trigger traceback in 
           usage()

    jobarchived)

        a: now performs regular database Housekeeping every 20 job XML 
           iterations (previously only once at startup)
        a: now checks if ARCHIVE_DATASOURCES are present in gmetad.conf
        f: prevent an Exception to occur when determining datasource polling 
           interval
        f: bug where config file handle was not closed

1.0:

    jobmond)

        a: now supports multiple udp send channels
        a: now supports job arrays
        c: updated Gmetric XDR protocol to version 3.1+ compatible

        c: gmond.conf parsing has been rewritten to handle include's and 
           multiple send channels
        c: METRIC_MAX_VAL_LEN is now determined from gmond.conf
        c: utilize new job monarch protocol

        f: can now handle new PBSQuery / pbs_python versions
        f: default gmond.conf search location is now /etc/ganglia/gmond.conf
        f: fatal error's are now printed to shell upon startup, not just syslog
        f: more error checking and miscellanious bugfixes

    jobarchived)

        r: no longer use pyPgSQL for postgres database
        c: now use psycopg2 module for postgres database

        a: job thread now utilizes db commits and rollbacks
        a: now use USER/PASS authentication to database (in stead of hostbased)

        c: database schema: changed job_id to varchar to support job arrays
        c: database schema: changed job_name max length to 255, just like 
           torque
        c: database schema: added username/password role authentication
        c: utilize new job monarch protocol

        f: job thread no longer hangs when insert/update of a job in database 
           fails 
        f: rewrite of job (finished) detection: all finished jobs again 
           properly detected
        f: job checking now done post-parsing not while parsing
        f: more error checking and miscellanious bugfixes

    web)

        r: removed Pie chart
        r: removed TemplatePower
        r: removed php ini_set's and time limit directive: should be handled in
           php.ini
        r: removed "Get Fresh Data" button: served no purpose anymore
        a: now utilize Dwoo templates for html output

        a: now use USER/PASS authentication to database (in stead of hostbased)
        a: ClusterImage now drops a shadow below nodes
        a: RRDs now show "Last: Min: Avg: Max:" values in legend

        c: utilize new job monarch protocol
        c: all templates rewritten from TemplatePower to Dwoo
        c: graph.php now used for overview and archive
        c: RRDs job start/finish line is now dashed green/red line with legend

        f: some dbase fields are now CAST to INT for php since postgres now 
           requires explicit casts
        f: sort order descending/ascending is now correct
        f: many, many speed and memory improvements
        f: more error checking and miscellanious bugfixes

0.4:

	jobmond)
		a:	SGE support
			thanks to: Dave Love - d(d.o.t)love(a.t)liverpool(d.o.t)ac(d.o.t)uk
			for writing it!
		a:	LSF support
			thanks to: Mahmoud Hanafi - mhanafi(a.t)csc(d.o.t)com
			for writing it!
		a:	GMETRIC_TARGET is now parsed from gmond.conf
		a:	GMETRIC_BINARY is now looked for in PATH
		f:	queue selection support is now working
			thanks to: Craig West - cwest(a.t)astro(d.o.t)umass(d.o.t)edu
			for the patch
	web)
		a:	large graphs link for job report
			thanks to: Craig West - cwest(a.t)astro(d.o.t)umass(d.o.t)edu
		a:	SHOW_EMPTY_COLUMN, SHOW_EMPTY_ROW options for ClusterImage hostname parsing

0.3.1:

	other)
		f:	updated INSTALL since "addons" directory is not included by default anymore in Ganglia
			thanks to: Steven DuChene linux(d.a.s.h)clusters(a.t)mindspring(d.o.t)com
			for reporting it

	rpm)
		f:	add "addons" directory since it's not included by default anymore in Ganglia
		f:	properly rewrite WEBDIR path in %files when rebuilding rpms with Makefile

	web)
		f:	typo in empty_cpu variable: causing incorrect 'free cpu' count and similar errors
			thanks to: Craig West - cwest(a.t)astro(d.o.t)umass(d.o.t)edu
			for reporting it
		f:	changed erroneous domain detection a little
			thanks to: Craig West - cwest(a.t)astro(d.o.t)umass(d.o.t)edu
			for reporting it
		a:	now properly detects whether or not to use FQDN or short hostnames w/o domain
			thanks to: Craig West - cwest(a.t)astro(d.o.t)umass(d.o.t)edu
			thanks to: Jeffrey Sarlo - JSarlo(a.t)Central(d.o.t)UH(d.o.t)EDU
			for the many testing and reporting it

			SPECIAL THANKS to the University of Houston for sending me a shirt!

	jobarchived)
		f:	properly catch postgres exception
		f:	don't use debug_message while loading config file

0.3:

	web)
		a:	allow per-cluster settings/override options: see CLUSTER_CONFS option
		a:	clusterimage can now draw nodes at x,y position parsed from hostname
			see SORTBY_HOSTNAME for this in clusterconf/example.php
		a:	clusterimage nodes are now clickable: has link to all jobs from that host
		a:	clusterimage nodes now have a tooltip: displays hostname and jobids for now
		a:	jobmonarch logo image
			thank to: Robin Day
			for the design
		a:	rrd graph of running/queued jobs to overview
		a:	per-cluster settings for archive database
			thanks to: Alexis Michon - alexis(d.o.t)michon(a.t)ibcp(d.o.t)fr
			for the patch

		c:	host archive view is now more complete and detailed in the same manner as
			Ganglia's own host view
		c:	host archive view available metric list is now compiled from disk,
			so that the detailed archive host view works even when the node is currently down.
		c:	removed size restrictions from detailed host archive view

		f:	compatibility: removed php5 call
			thanks to: Alexis Michon - alexis(d.o.t)michon(a.t)ibcp(d.o.t)fr
			for the patch
		f:	prevent negative cpu/node calculation
			thanks to: aloga(a.t)ifca(d.o.t)unican(d.o.t)es
			for the patch
		f:	archive search not properly resetting nodes list
			thanks to: Alexis Michon - alexis(d.o.t)michon(a.t)ibcp(d.o.t)fr
			for the patch 
		f:	detailed host view from jobarchive was broken since hostbased support of 0.2
			now host view is properly set and parsed again
			thanks to: Alexis Michon - alexis(d.o.t)michon(a.t)ibcp(d.o.t)fr
			for reporting the bug and suggesting a patch
		f:	bug where jobstart redline indicator in host detail graphs was set incorrectly
			or not at all due to a miscalculation in job times
		f:	bug where hostimage headertext xoffset was miscalculated, causing the column names
			to overlap their position when the columnname was longer than the columnvalues

	jobmond)

		a:	syslog support
		a:	report number of running/queued jobs as seperate metrics
		a:	native gmetric support, much faster and cleaner!
			thanks to: Nick Galbreath - nickg(a.t)modp(d.o.t)com
			for writing it and allowing inclusion in jobmond

		f:	crashing jobmond when multiple nodes amounts are requested in
			a queued job: numeric_node variable not initialized properly
			thanks to: aloga(a.t)ifca(d.o.t)unican(d.o.t)es
			for supplying the patch
			and many others for reporting and helping debug this
		f:	hanging/blocked, increased cpu usage and halted reporting
			thanks to: Bas van der Vlies - basv(a.t)sara(d.o.t)nl
			for discovering the origin of the bug
			thanks to: Mickael Gastineau - gastineau(a.t)imcce(d.o.t)fr
			for reporting it and testing the fix
			thanks to: Craig West - cwest(a.t)astro(d.o.t)umass(d.o.t)edu
			for reporting it and testing the fix
		f:	uninitialized variable in checkGmetricVersion()
			thanks to: Peter Kruse - pk(a.t)q-leap(d.o.t)com
			for the patch
		f:	undefined PBSError
			thanks to: Peter Kruse - pk(a.t)q-leap(d.o.t)com
			for reporting it

		r:	SGE support broken

	jobarchived)

		a:	can now use py-rrdtool api instead of pipes, much faster!
			install py-rrdtool to use this
			backwards compatible fails back to pipes if module not installed

		c:	all XML input was uniencoded, which could cause errors,
			now all properly converted to normal strings

		f:	when XML data source (gmetad) is unavailable parsethread didn't return correctly
			which caused a large number of threads to spawn while consuming large amounts of memory
		f:	autocreate clusterdirs in archivedir
		f:	unhandled gather exception
		f:	incorrect stop_timestamping when jobs finished
			thanks to: Alexis Michon - alexis(d.o.t)michon(a.t)ibcp(d.o.t)fr
			for finding and debugging/testing it

0.2:

	web)
		f:	misc. optimization and bugfixes
		f:	now fully compatible with latest PHP5 and PHP4

		c:	cluster image now incorporates small text descr.
		c:	monarch (cluster/host) images no longer displayed
			for clusters that are not jobmond enabled
		c:	pie chart percentages are now cpu-based instead of node-based

		a:	host template for Ganglia
			adds a extra monarch host image to Ganglia's host overview
			which displays/links to the jobs on that host
			NOTE!: be sure to copy/install new template from addons/templates
		a:	(optional) nodes hostnames column
			thanks to: Daniel Barthel - daniel(d.o.t)barthel(a.t)nottingham(d.o.t)ac(d.o.t)uk
			for the suggestion

	jobmond)

		f:	when a job metric is longer than maximum metric length,
			the info is split up amongst multiple metrics
		f: 	no longer exit when batch server is unavailable
			thanks to: Peter Kruse - pk(a.t)q-leap(d.o.t)com
			for the patch
		f:	fd closure bug causing stderr/stdout to remain open after daemonizing

		c:	rearranged code to allow support for other batch systems

		a:	(experimental) SGE (Sun Grid Engine) support as batch server
			thanks to: Babu Sundaram - babu(a.t)cs(d.o.t)uh(d.o.t)edu
			who developed it for a OSCAR's Google-SoC project
		a:	pidfile support 
			thanks to: Michael Jeanson - michael(a.t)ccs(d.o.t)usherbrooke(d.o.t)ca
			for the patch
		a:	usage display
			thanks to: Michael Jeanson - michael(a.t)ccs(d.o.t)usherbrooke(d.o.t)ca
			for the patch
		a:	queue selection support: ability to specify which QUEUE's to get jobinfo from
			thanks to: Michael Jeanson - michael(a.t)ccs(d.o.t)usherbrooke(d.o.t)ca
			for the patch

	jobarchived)

		f:	XML retrieval for Ganglia version >= 3.0.3 working properly again
		f:	database storing for Ganglia version >= 3.0.3 working properly again
		f:	fd closure bug causing stderr/stdout to remain open after daemonizing

		c:	misc. bugfixes to optimize XML connections
		c:	misc. bugfixes for misc. minor issues

		a:	cleaning of stale jobs in dbase: see JOB_TIMEOUT option

0.1.1: 

	web)

		f:	misc. layout bugs for overview & search
		f:	bug that occured when calculating the number of nodes when there
			was more than one job running on a machine

		c:	column requested memory is now optional through conf.php
		c:	search and overview tables are now full screen (100%)
		c:	overview jobnames are now cutoff at max 9 characters
			to prevent (layout) scews in the tables
		c:	overview graphs are no longer downsized

		a:	(optional) column 'queued' (since) in overview
		a:	search results (can) now have a SEARCH_RESULT_LIMIT
			this increases performance of the query's significantly!
		a:	date/time format as displayed is now configurable through conf.php

	jobmond)

		a:	now reports 'queued since' (or creation time) of jobs

	documentation)

		f:	wrong e-mail adress in INSTALL (doh!)

0.1:

	- First public release