2.0.0

Added

• Add support for Prometheus as a performance data source. (#268)
• Add 'datasource' field to output summarization record. (#269)
• Add better support for the configuration check utility in supremmconf.py. (#263)
• Add support for configurable job uniqueness. (#265)

Changed

• Update from Python 2.7 to Python 3.6.
• Update template paths to explicitly include the PCP version. (#255)
• Use multithreaded archive indexing by default. (#250)

Fixed

• Update indexing in plugins to use integer division when processing jobs with greater than 64 nodes. (#264)
• Fix string encoding from byte array to UTF-8 for PCP on RHEL 8. (#261)

2.0.0-rc.3

Third release candidate for version 2.0.0

2.0.0-beta3

Adding

• Support for Python3 and Rocky 8

2.0.0-beta2

Added

• Python 3 and Rocky 8 support

2.0.0-beta1

Added

• Python 3 and Rocky 8 support

1.4.1

• Fix edge case where one type of corrupt performance counter value was not handled by the hardware performance counter plugin.
• Fix bug in the job script ingestor that caused the script data in the database to be set to the string "0" if a script data was ingested more than once.

1.4.0

Added

• Added new configuration setting that determines how the job script timestamps
  are interpreted. See https://supremm.xdmod.org/9.0/supremm-jobscript.html for details.
• Added timeseries plot generation for Lustre filesystem data.
• Added support for PBS style cgroups for CPU and memory metrics.

Changed

• Improve error reporting when the temporary job directories cannot be created.
• Improve the cpu performance counter plugin to report data on the available counters
  even if not all are present.
• Update internal developer documentation
• The supremm-setup script now requires mongo client 3 or later to run the mongo database setup
  (instructions for manual setup using the older client are at https://supremm.xdmod.org/9.0/supremm-processing-configuration.html#setup-mongodb)
• Improve the job metadata handling in the developer test harness

Fixed

• Update indexarchives.py to defer opening the connection to the XDMoD database until it
  it needed. This resolves an timeout issue when processing a large number of files
  or on a slow filesystem.

1.4.0.rc01

First release candidate for the 1.4.0 version.

Version 1.3.0

Added

• Added IPMI power timeseries plugin that generates timeseries data from
  IPMI power measurements.
• Add component tests for the performance counter plugins.

Changed

• Updated RPM dependencies to match the official PCP version in RedHat/Centos 7.
• Update the CI build to test against XDMoD version 8.1.
• Improved support for hardware performance counters in the various performance
  counter plugins.

Fixed

• Fixed bug in the performance counter preprocessor that caused it to
  incorrectly mark the performance counters as disabled by user.
• Fixed incorrect unit in schema definition for the block device timeseries metric.

Version 1.2.0

Added

• Added GPU power plugin that summarizes the power usage for nvidia GPUs.
• Added IPMI power plugin that summarizes the power usage from impi.
• Added support for scanning the YYYY/MM/HOSTNAME/YYYY-MM-DD directory format for PCP archives.
• Added --dry-run option to indexarchives.py script (used for testing purposes).

Changed

• Removed deprecated code that supported python 2.6.
• Updated templates to use the new recommended directory format.
• Updated test harness to work with current plugins.
• Database access code now sets the local_infile flag (needed for MySQL driver version 2.0.0 or later).

Fixed

• Summarization script now only includes job start and end archives that are within 30 seconds of the job start and end. This mitigates a bug caused if the resource manager reuses the same job id for two different jobs.
• The summarization software now skips corrupt PCP archives and will try to continue processing the valid archives. Previously processing would stop.
• If a PCP archive has data missing at a timestep the framework will skip the timestep and continue processing the archive. Previously processing would stop.
• The Slurm Proc preprocessor now filters non-unicode characters from command names.
• Improvements to error handing in preprocessors and in the perfevent plugin.