[Bug 2019293] Re: mgr: relax "pending_service_map.epoch > service_map.epoch" assert

[Dave Jones]

A question about the patch: it doesn't apply cleanly against the current
focal version of ceph. Specifically, the parts that add the patch into
the packaging (the changes directly under debian/) are fine, but there's
other changes under src/test which appear to re-create pre-existing
files (and which wouldn't be acceptable anyway as they'd be touching the
orig tar-ball content).

In addition to the changes requested by Lucas, could you upload another
version without the changes under src/test?

-- 
You received this bug notification because you are a member of Ubuntu
OpenStack, which is subscribed to ceph in Ubuntu.
https://bugs.launchpad.net/bugs/2019293

Title:
  mgr: relax "pending_service_map.epoch > service_map.epoch" assert

Status in ceph package in Ubuntu:
  New
Status in ceph source package in Focal:
  New

Bug description:
  [Impact]

  This issue has been observed from ubuntu Octopus release. 
  An assert will be triggered during the mgr fail-over process if the new active one unexpectedly received two continuous service map update. 
  The upstream fix has relaxed the assert condition to allow the new active mgr to receive multiple service map update in a fail-over scenario. 

  [Test Case]

  1. Deploy a 15.2.16 ceph cluster

  2. upgrade it to 15.2.17, inject multiple service map to the monitor

  3. stop the active mgr

  4. observe the new active mgr will hit the assert condition

  
  [Potential Regression]
  The new active mgr would be required to process multiple service map, it might slow down a little bit on the fail-over process, but still much better than crash.


  [Other info]

  Upstream bug tracker: https://tracker.ceph.com/issues/51835
  Upstream PR: https://github.com/ceph/ceph/pull/45984
  we need to backport it to octopus

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/ceph/+bug/2019293/+subscriptions