Collectl Release Information

This section tries to highlight the changes from release to release, starting with Version 2.3.0. Since most releases contain minor bug fixes they will not be highlighted here. For more details of what specifically changed you should always consult the actual release notes which can be found in /opt/hp/collectl/docs/RELEASE-collectl.

Version 4.3.1 Oct 31 , 2018

• do NOT divide disk wait times by inteval when using -sD
• a few other minor changes described in RELEASE-collectl

Version 4.3.0 Oct 3 , 2016

• very minor, disabled -sL as I forgot to when I disabled -sl

Version 4.2.0 Jun 12 , 2016

• Fixed disk summary stats bug which caused them to mostly be reported as 0s
• InfinBand OPA V4 now reads 64 but counters from /sys

Version 4.1.3 May 12 , 2016

• very minor bug fix to uninitialized variable for working with SUSE

Version 4.1.2 Feb 2, 2017

• minor bug fixes

Version 4.0.5 Jul 13, 2016

• minor bug fixes
• added hed as new ethernet device

Version 4.0.4 Jan 29, 2016

• new switch: --dskremap
• added AnonHuge memory to verbose mem stats
• added en as new ethernet device

Version 4.0.2 May 27, 2015

• minor bug fixes and some cleanup

Version 4.0.0 March 9, 2015

• New plugins for monitoring OpenStack VMs and Swift
• Colmux has been moved out of collectl-utils package into this one

Version 3.7.4 September, 2014

• Native lustre support is no longer enabled by default. You can still use it but you must explicitly specify 'l' as a requested subsystem.
• A couple of new switches but all pretty minor. Run collectl --whatsnew for details

Version 3.7.3 April, 2014

• Support for infiniband extended counters also allows multiple copies to run
• Removed myrinet and quadrics support. Also dropped nvidia and sexpr as promised
• New switch --cpufilt, allows display a subset of CPUs for machines with high cpu counts
• Support for RHEL7

Version 3.6.9 Oct, 2013

• Big oops, introduced accidential infinite loop for -sN -P so don't use 3.6.8 anymore!

Version 3.6.8 Oct, 2013

• new switch: --procopt u[width] allows different username width in proc display
• new switch: --procopt k, removes known shells (like /bin/sh, /usr/bin/perl, etc) from process name string
• new switch: --dskopt/--netopt o will exclude unused devices from plot order
• new switch: --intfilt works like other filters for interrupe names/numbers

Version 3.6.7 Oct, 2013

• new switch: --cpuopts z, disables detail lines with idle CPUs
• a couple of new switches for graphite, e and r
• added disk detail counters to lexpr

Version 3.6.5 Nov, 2012

• sexpr is deprecated and will be removed in 2014
• Lots of testing within the HP Public Cloud at 1-sec monitoring interval
• New align switch in lexpr makes it possible to report data aligned at 1 minute boundary, making it possible to accurately loadi data into RRD at different interval/li>
• Removed experimental SNMP export module, merging its functionality into -st
• Now handles dynamic disks/networks correctly
• No longer requires ethtool to get network speeds
• Added a variety of switches and options, see release notes for details

Version 3.6.3 Mar, 2012

• New switches, --rawdskfilt and --rawnetfilt filter data at time of collection.
• Fixed some problems with process filtering when long process names/owners
• Removed readS from kit and added hello and graphite which somehow never got in
• Added exit codes for success/failure
• Added some new systems to envrules.txt: SL230, SL250 and SE1170

Version 3.6.1 Feb 20, 2012

• Support for Graphite
• Lots of new switches, see: collectl --whatsnew
• Experimental module for snmp support, think netstat -s
• Can now run as non-root daemon

Version 3.6.0 Oct 17, 2011

• New subsystem: -sM. memory detail now shows numa info
• Bunch of new/updated switches: --dskopts, --netopts, --xopts, --extract
• Enhanced disk/network output filtering, allowing for exclusion of instance names
• Added new sections to documentation
• Finally dropped support for 2.4 kernels

Version 3.5.1 May 27, 2011

• new switch: --memopt R, show incremental changes to memory like sar -R
• miscellaneous fixes obscure problems that only I trip over, maybe...

Version 3.5.0 Feb 22, 2011

• new import module for monitoring nvidia GPUs
• new switch: --netopts allows more details about network errors
• new switch: --utime allows logging of micro-times for perforamance analysis
• new switch: --showcolheaders, primarily to support new colmux in collectl-utils, which now provided cluster-top functionality!!!
• new addition to --procopt R, causes real-time priorities to be displayed rather than RT
• enacements to gexpr to make more ganglia friendly

Version 3.4.3 Sept 8, 2010

• new switch: --netfilt
• fixed a few problems between OFED 1.4 and 1.5

Version 3.4.2 Jul 21, 2010

• new switch: --dskfilt for filtering disk output
• enhancement to -- procfilt to all UID range selection
• support for CPUs going offline/online dynamically

Version 3.4.1 Apr 01, 2010

• several new for impi monitoring
• new switch --umask for setting mode to files collectl creates
• new switch --whatsnew that give a brief summary of enhancements by version
• support for Fusion-IO disks

Version 3.4.0 Jan 07, 2010

• New installation tree structure
• Support for OFED 1.5 (perfquery output format slightly different

Version 3.3.6 Sept 16, 2009

• A number of minor bug fixes

Version 3.3.5 Jul 21, 2009

• Significant work on -G/--group code to clean of a number of playback formatting problems
• Improved performance of playback which is particularly noticable for very large files with a lot of process data
• Fixed --vmstat which got broken in earlier version
• Added col2tlviz utility to kit which converts collectl tab files to tlviz format

Version 3.3.4 Jun 14, 2009

• Detect changes in disk configuration
• If disk read/write rate is too high to be real, ignore and report all 0s
• Added LSB headers to all start scripts

Version 3.3.2 Apr 28, 2009

• A bug slipped in that prevented collectl from running as a daemon!

Version 3.3.1 Apr 27, 2009

• New data collection/reporting API, including hello.ph and misc.ph
• New export: gexpr, for exporting data to ganglia
• support for multiple daemons

Version 3.2.1 Mar 12, 2009

• Added NFS V4 support and collection
• Removed restriction on only monitoring one type of nfs date [client|server][2|3] to all
• Added new subsystem, buddyinfo, which shows memory fragmentation
• Expanded contents of verbose memory output

Version 3.1.3 Jan 23, 2009

• Fixed fatal bugs in previous release as well as some minor ones

Version 3.1.2 Jan 20, 2009 - warning - this release has fatal bugs!

• Added Power to environmental monitoring
• When contiguous files are played back together, will use last interval of one file as the base counter for the new file such that there won't be any holes in the data
• Fixed bug in --from/--thru in the were no honoring the date field
• Allow multiple clients to connect when is server mode
• Cleaned up some subtle formatting problems with less common switch combinations

Version 3.1.1 Nov 07, 2008

• Added fields to --top for monitoring slabs
• New switch --slabtop reports summary of allocated memory usage for each slab
• New 'c' process option to report accumulated dead children cpu time with parent
• Miscellaneous small bug fixes, most with effect switch validation and formatting

Version 3.0.1 Sep 03, 2008

• Expanded --top functionality. See --showtopoptions
• Experimental IPMI monitoring
• Enhanced MDS stats to support lustre 1.6.5
• Added --grep switch to allow searching/reporting data in raw files
• Fixed lots of little things in formatting that nobody even noticed!
• Fixed some subtle bugs that only show up with weird switch combinations

Version 3.0.0 Jul 01, 2008

• Big enhancement to --top functionality to allow subsystems! Try the command collectl --top -scdn to see for yourself
• A couple of switch names changed to be more consistent
• New --procanalyze can produce spreadsheet loadable summary of process activity
• More consistent formats of I/O related data for disks, network, infiniband and lustre
• Improved help
• Miscellaneous bug fixes, none effecting data collection

Version 2.6.4 Jun 04, 2008

• Extended --top to sort by I/O or Page Faults
• Allow filters with --top
• Allow --top to be used in playback mode
• Replaced -Y, -Z and some of -O options by --procopts, --slabopts, --procfilt and --slabfilt
• Added new process options to look for all threads, include syscalls in output and include command arguments in output
• New switch --iosize to include average I/O size to brief output
• Always include average I/O sizes to verbose and detail output
• Changed data that gets displayed for -si

Version 2.6.2 Apr 29, 2008

• Changed --custom with --export which now adds ability to export data over a socket
• Added server as option for -A which tells collectl to run as a server as opposed to client and in this mode doesn't exit when the other end of the connection goes away, but rather listens for a new connection
• If collectl can determine the height of the terminal window via resize it sill set --headerrepeat to that. Also extended how values for that switch are interpretted and so eliminated the need for -oh and -oH which will be dropped in a future release.
• Added new swicth --all, which will show all subsystem data

Version 2.5.1 Mar 21, 2008

• Support for OFED 1.3
• Support for Lustre 1.6.4.3
• More documentation updates

Version 2.5.0 Feb 29, 2008

• Added Interrupts by CPU reporting
• Moved a LOT of man page content to new on-line user documentation at http://collectl.sourceforge.net/Documentation.html and /opt/hp/collectl/docs

Version 2.4.3 Feb 05, 2008

• Fixed bug in CPU utilization calculation
• Mapped memory pages now includes anonymous pages
• Was incorrectly including network bond traffic with summary
• Allow disabling of checking for duplicate collectl instances monitoring IB
• Detection of logical disks created after collectl starts
• Improved/simplified algorithm to check for bogus network data
• Default display now fits in 80 column wide window!

Version 2.4.2 Jan 16, 2008

• Now purges both raw and plottable files
• Fixed old slab reporting (collection always worked) which new slab reporting broke

Version 2.4.1 Jan 05, 2008

• Added support for new SLUB allocator which is used on kernels 2.6.22 and above
• Fixed a bug that was causing causing cpu percentages to be miscalculated

Version 2.4.0 Dec 23, 2007

• Added display of process i/o stats on kernels that support it

Version 2.3.4 Dec 13, 2007

• Added IB support for Mellanox IB Infinihost III card
• Forgot to include irq, soft and steal cpu counters in s-expressions
• Removed -H command since nobody uses it or at least shouldn't
• Removed requirement that --sexpr always writes to file and now permits going over socket. Also did a general cleanup of logging code and updated collectl-logging man page

Version 2.3.3 Oct 16, 2007

• Added 3 new cpu field: irq, soft and steal which results in some output format changes
• Added mlx4 as a supported Infiniband HCA. Thanks Fred.
• >Added support for displaying device-mapper devices in disk details. Thanks again, Rene.
• Lspci changed some of its output formats as of version 2.2 and so now collectls checks its version before parsing the output. Thanks Peter.

Version 2.3.2 Sep 04, 2007

• Added new memory data element, Commited_AS. Thanks Rene.
• Added 2 new switches, --utc and --sep, which lets you specify date/time in plot format appear in UTC format while sep allows you to change the plot format field separator. The combination of the two will therefore allow you to generate plot files in rrd format.

Version 2.3.1 Aug 09, 2007

• Cleaned up Lustre OST to filesystem name mapping for RPC buffer stats.

Version 2.3.0 Jul 25, 2007

• The format of the perfquery command to read/clear the Infiniband counters was also clearing the error counters. It has been changed to only clear the performance related ones.
• The location collectl looks for the perfquery command has been changed to a list since its location is different in OFED 1.1 and OFED 1.2.