Skip to content

Latest commit

 

History

History
288 lines (195 loc) · 10.7 KB

CHANGELOG.rst

File metadata and controls

288 lines (195 loc) · 10.7 KB
Raw

Changelog

The purpose of this document is to list all of the notable changes to this project. The format was inspired by Keep a Changelog. This project adheres to semantic versioning.

Release 0.12.2 (2019-12-11)

Bug fix for 2 most recent releases: s/OSError/EnvironmentError/g

This is a follow up to / bug fix for release 0.12 and release 0.12.1 where I accidentally used OSError when I should have used IOError or the generic parent class EnvironmentError. I've verified that EnvironmentError is compatible with both Python 2.7 and 3.

The last few releases clearly show the value of having an automated test suite to guard against regressions, unfortunately due to the nature of the Negotiator project I'm not quite sure how I would get a functional test suite up and running on Travis CI.

It's definitely possible, and on my wish list, but it is one of the higher hanging fruits on that -almost infinite- wish list, so don't hold your breath 😛.

Release 0.12.1 (2019-12-09)

Enable retry in guest CLI.

This is a follow up to release 0.12 because I neglected to add retry=True to the negotiator_guest.cli module in that release (the new behavior is opt-in in the Python API so as to improve backwards compatibility and make the API more foolproof).

Release 0.12 (2019-12-05)

Retry character device access in guest agent when EBUSY error is reported.

At my employer we operate 200+ virtual servers that have Negotiator installed and in recent months we've started building more and more monitoring on top of Negotiator, resulting in hundreds of invocations per day. This is when I started seeing intermittent errors like the following:

IOError: [Errno 16] Device or resource busy: '/dev/vport2p2'

The new retry on EBUSY behavior is intended to minimize occurrences of such race conditions.

Release 0.11 (2019-10-11)

Fix error in error handling in negotiator-host

This resolves the traceback below when negotiator-host is running but 'crashes' because virsh list fails because it can't connect to libvirtd. This situation is still an unrecoverable error, but the intention was for it to be an unrecoverable error that did not produce a traceback...

Traceback (most recent call last):
  File ".../negotiator_host/cli.py", line 117, in main
    action()
  File ".../negotiator_host/__init__.py", line 46, in __init__
    self.enter_main_loop()
  File ".../negotiator_host/__init__.py", line 53, in enter_main_loop
    self.update_workers()
  File ".../negotiator_host/__init__.py", line 62, in update_workers
    running_guests = set(find_running_guests())
  File ".../negotiator_host/__init__.py", line 270, in find_running_guests
    except ExternalCommandFailed:
NameError: global name 'ExternalCommandFailed' is not defined
Cleaned up flake8 F401 (imported but unused) warning
As a result of moving to humanfriendly.compact().
Minor changes relating to Python versions

Two minor changes relating to Python version compatibility:

  • Unbreak make docs (Sphinx insists on Python 3, and rightfully so).
  • Replace Python 2.6 reference in README with Python 2.7.

Compatibility with Python 3 hasn't been verified but is more or less expected given a pure Python code base. Maybe a Unicode error slipped in here or there, as I said "to be verified".

Release 0.10 (2019-03-03)

Clarify verbosity control
Update the command line interface usage messages to clarify that the options --verbose and --quiet can be repeated (in response to #1).
No traceback when guest discovery fails

Don't log a traceback when guest discovery using the virsh list command fails, to avoid spamming the logs about a known problem. This change was made to counteract the following interaction:

  • The negotiator documentation specifically suggests to use a process supervision solution like supervisord to automatically restart the negotiator daemon when it dies.
  • When the libvirt daemon is down virsh list will fail and the negotiator daemon dies with a rather verbose traceback (before release 0.10).
  • Because supervisord automatically restarts the negotiator daemon but doesn't know about the libvirt dependency, several restarts may be required to get the negotiator daemon up and running again.
  • This "restart until it stays up" interaction would result in quite a few useless tracebacks being logged which "polluted" the logs and might raise the impression that something is really broken (that can't be fixed by an automatic restart).

Release 0.9 (2019-03-03)

Refactored channel discovery to use virsh list and virsh dumpxml:

  • The recent addition of Ubuntu 18.04 support proved once again that the old channel discovery strategy was error prone and hard to maintain.
  • Since then it had come to my attention that on Ubuntu 18.04 guest names embedded in pathnames of UNIX sockets may be truncated in which case the domain id provides the only way to match a UNIX socket to its guest.
  • Despite the previous point, I also wanted to maintain compatibility with libvirt releases that don't embed the domain id in the pathnames. Doing so based on the old channel discovery strategy would have become messy.

So I decided to take a big step back and opted for a new strategy that will hopefully prove to be more robust and future proof. Thanks to @tarmack for initially suggesting this approach.

Release 0.8.6 (2019-02-25)

Follow-up to making channel discovery compatible with Ubuntu 18.04:

Release 0.8.5 (2019-02-23)

  • Made channel discovery compatible with Ubuntu 18.04 (related to #1).
  • Added this changelog, restructured the documentation.
  • Embedded CLI usage messages in readme and documentation.
  • Updated supervisord configuration examples to use stderr_logfile instead of redirect_stderr.
  • Other minor changes not touching the code base.

Release 0.8.4 (2016-04-08)

Follow-up to previous commit (Ubuntu 16.04 support).

Release 0.8.3 (2016-04-08)

Make channel discovery compatible with Ubuntu 16.04.

Release 0.8.2 (2015-10-29)

Make platform support more explicit in the documentation (Linux only, basically :-P).

Release 0.8.1 (2014-12-30)

Improve guest channel (re)spawning on hosts (improves robustness).

Release 0.8 (2014-11-01)

Proper sub process cleanup, more robust blocking read emulation.

Release 0.7 (2014-10-24)

Support for (custom) remote call timeouts with a default of 10s.

Release 0.6.1 (2014-09-28)

Bug fix for Python 2.6 compatibility (count() does not take keyword arguments).

Release 0.6 (2014-09-26)

  • Implemented blocking reads inside guests (don't ask me how, please ...).
  • Improved getting started instructions on adding virtual devices.
  • Rebranded s/generic/scriptable/g and improved the readme a bit.

Release 0.5.2 (2014-09-24)

Add syntax highlighting to the code and configuration samples in the readme and explicitly link to the online documentation available on Read the Docs.

Release 0.5.1 (2014-09-24)

  • Minor improvements and fixes to the documentation.
  • Properly documented the environment variables exposed to host commands.
  • Added trove classifiers to the setup.py scripts.
  • Bumped the version to release updated documentation to PyPI.

Release 0.5 (2014-09-24)

  • Support for proper bidirectional user defined command execution on both sides.
  • Improved the negotiator-guest usage message (by mentioning character device detection).

Release 0.2.1 (2014-09-22)

Fixed a typo in the readme, fixed a bug in the makefile and bumped the version so I could push a new release to PyPI because the readme was missing there (due to the makefile bug).

Release 0.2 (2014-09-22)

  • Added automatic character device selection.
  • Created online documentation on Read the Docs.

Release 0.1 (2014-09-22)

The initial commit and release.