yahoo-eng-team team mailing list archive

[Li Ma <skywalker.nick@xxxxxxxxx>]

Public bug reported:

After upgrading from icehouse b2 to b3, neutron-server using zmq was
crashed silently. No core dump, no exception, no error, no warnings.
After I switched to rabbitmq, everything went well. It really drove me
crazy.

Then I tried to debug, it was really hard to do it, because there were
no useful messages provided.

Then I compared source codes between b2 and b3, and finally I found out
the main difference between the two versions are that b3 enables a new
rpc_worker mechanism.

I removed all the related code back to the way b2 goes.

Then the problem is immediately solved.

I just guess that rpc_worker uses greenpool which has conflicts with the
thread pool from zeromq. I suggest to claim in neutron.conf that it is
incompatible with zeromq, and when rpc_worker == 0, just disable it.

** Affects: neutron
     Importance: Undecided
     Assignee: Li Ma (nick-ma-z)
         Status: New

** Changed in: neutron
     Assignee: (unassigned) => Li Ma (nick-ma-z)

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1298803

Title:
  rpc worker makes neutron-server crash

Status in OpenStack Neutron (virtual network service):
  New

Bug description:
  After upgrading from icehouse b2 to b3, neutron-server using zmq was
  crashed silently. No core dump, no exception, no error, no warnings.
  After I switched to rabbitmq, everything went well. It really drove me
  crazy.

  Then I tried to debug, it was really hard to do it, because there were
  no useful messages provided.

  Then I compared source codes between b2 and b3, and finally I found
  out the main difference between the two versions are that b3 enables a
  new rpc_worker mechanism.

  I removed all the related code back to the way b2 goes.

  Then the problem is immediately solved.

  I just guess that rpc_worker uses greenpool which has conflicts with
  the thread pool from zeromq. I suggest to claim in neutron.conf that
  it is incompatible with zeromq, and when rpc_worker == 0, just disable
  it.

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1298803/+subscriptions