[Bug 1315021] Re: [SRU] Service restart fails because process has not exited

[Chris J Arges]

Sponsored for Trusty/Utopic.

-- 
You received this bug notification because you are a member of Ubuntu
Sponsors Team, which is subscribed to the bug report.
https://bugs.launchpad.net/bugs/1315021

Title:
  [SRU] Service restart fails because process has not exited

Status in puppetmaster package in Ubuntu:
  Fix Released
Status in puppetmaster source package in Trusty:
  New
Status in puppetmaster source package in Utopic:
  New

Bug description:
  [Impact]

  Users restarting puppet and puppetmaster may have problems restarting the
  daemons, in systems under heavy load may be more frequent, because if the stop
  operation takes more than 1 second the start operation fails due that the
  daemon is still running.

  If the daemon isn't started in a machine managed by puppet (or the
  puppetmaster) will require manual intervention to bring up the system.

  [Test]

  - Create a small VM (512MB, 1 vcpu)
  - Install puppet and puppetmaster packages
  - Restart each daemon twice with a 10 seconds pause
    # service puppetmaster restart; sleep 10; service puppetmaster restart

  Expected results:

  - The service is restarted

  Actual results:

  - The service is stopped, but it isn't started again the console
  prints:

  root at vivid-puppetmaster:~# service puppetmaster restart ; sleep 10 ; service puppetmaster restart
   * Restarting puppet master
     ...done.
   * Restarting puppet master
     ...fail!

  [Regression Potential]

  - If after 10 seconds the daemon isn't exited the start-stop-daemon command
    will send the KILL signal and wait for 5 seconds, this is different from the
    current behavior that just informs the restart failure to the user and
    doesn't send the KILL signal.

  [Other Info]

  When restarting the puppetmaster process (via the service command, or calling
  the initscript directly), the restart can fail if the original process does
  not exit immediately. This is because the initscript is not written to wait
  for the original process to exit.

  In my testing under Trusty, adding "-R n" to the start-stop-daemon command in
  the initscript fixed the problem. It's difficult to determine an appropriate
  value for n, but it could be tunable via /etc/default/puppetmaster. It's
  definitely a problem not to wait at all for the process to exit.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/puppetmaster/+bug/1315021/+subscriptions