holland-coredev team mailing list archive

[Andrew Garner <1170446@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

When a holland plugin forks a command which in turn forks some subset of
commands, but that command is terminated or fails in some way that does
not kill its children, holland can sometimes leave children of its
children processes running for some time, even if the plugin otherwise
tracks the child exit status properly.

This is most common with xtrabackup, where the innobackupex perl script
fails to clean up after itself in a variety of circumstances. This can
leave the xtrabackup_* binary running in the background even after
holland has stopped.

I think we can largely resolve this by always sending a SIGTERM to the
holland process group to at least signal grandchild processes to exit
which will cover most of the common cases where this occurs - and
certainly fix a few nasty cases with innobackupex -> xtrabackup.

We can do something like:

# temporarily ignore the signal we will send our process group
with ignored_signals(SIGTERM):
    os.killpg(os.getpid(), SIGTERM)

And then exit as normal as needed.

** Affects: holland-backup
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Holland
Build, which is subscribed to holland-backup.
https://bugs.launchpad.net/bugs/1170446

Title:
  holland can leave orphan grandchildren

Status in Holland Backup Framework:
  New

Bug description:
  When a holland plugin forks a command which in turn forks some subset
  of commands, but that command is terminated or fails in some way that
  does not kill its children, holland can sometimes leave children of
  its children processes running for some time, even if the plugin
  otherwise tracks the child exit status properly.

  This is most common with xtrabackup, where the innobackupex perl
  script fails to clean up after itself in a variety of circumstances.
  This can leave the xtrabackup_* binary running in the background even
  after holland has stopped.

  I think we can largely resolve this by always sending a SIGTERM to the
  holland process group to at least signal grandchild processes to exit
  which will cover most of the common cases where this occurs - and
  certainly fix a few nasty cases with innobackupex -> xtrabackup.

  We can do something like:

  # temporarily ignore the signal we will send our process group
  with ignored_signals(SIGTERM):
      os.killpg(os.getpid(), SIGTERM)

  And then exit as normal as needed.

To manage notifications about this bug go to:
https://bugs.launchpad.net/holland-backup/+bug/1170446/+subscriptions