RabbitMQ annoying disconnections [kolla-ansible, rabbitmq]

Adam Tomas bkslash at poczta.onet.pl
Thu Sep 23 07:53:25 UTC 2021


Well, then see what I have in impl_rabbit.py (kolla-ansible 11, newest container images, Victoria):

cfg.BoolOpt('heartbeat_in_pthread',
                default=False,
                help="EXPERIMENTAL: Run the health check heartbeat thread "
                     "through a native python thread. By default if this "
                     "option isn't provided the  health check heartbeat will "
                     "inherit the execution model from the parent process. By "
                     "example if the parent process have monkey patched the "
                     "stdlib by using eventlet/greenlet then the heartbeat "
                     "will be run through a green thread."),


1. It is disabled by default, so heartbeats run through green thread?
2. Why enabling it didn’t change anything?
3. heartbeat = 600 for rabbit resolved problem for now (but I’ve seen 1-2 timeouts after 2 days anyway), but 10-minute timeout is kind of big, so I would like to use another solution

Best regards
Adam Tomas

> Wiadomość napisana przez Sean Mooney <smooney at redhat.com> w dniu 22.09.2021, o godz. 21:05:
> 
> On Wed, 2021-09-22 at 11:22 -0500, Ben Nemec wrote:
>> 
>> On 9/21/21 2:44 AM, bkslash wrote:
>>> So there are two questions:
>>> 
>>> 1. Why heartbeat_in_pthread didn’t change situation? And if it’s 
>>> deprecated it was kind of risky to use it long term (after upgrade to 
>>> wallaby it would stop working anyway?)
>> 
>> It's deprecated because it became the default behavior in subsequent 
>> releases. The functionality is not going away, just the ability to turn 
>> it off.
>> 
>> The only reason it was ever configurable was that we had concerns about 
>> mixing pthreads and greenthreads and wanted the people affected by the 
>> original bug to test the combination for a couple of releases before we 
>> turned it on for everyone.
> actully the ablity to trun it off is not going away either
> turing it on can break the nova comptue agent by breaking eventlet.
> https://bugs.launchpad.net/oslo.messaging/+bug/1934937
> here is the patch to undeprecate it https://review.opendev.org/c/openstack/oslo.messaging/+/800621
> based on the bug above this looke like it should be turned off unless you runing under mod_wsgi or uwsgi
> where the pthread is needed to escape there treadign lifecycle.
> 
>> 
> 
> 




More information about the openstack-discuss mailing list