[Openstack] problem with nova retry to neutron under load
don.waterloo at gmail.com
Sun Aug 17 19:07:48 UTC 2014
[icehouse on ubuntu 14.04]
in neutronclient/v2_0/client.py, in __init__, it sets self.retries = 0.
Later, the logic is max_attempts = self.retries + 1, so effectively 1 retry.
In my setup, when i do a mass delete (e.g. 40 heats stacks that each
have 5 instances w/ 10 networks), once in a while I will get an error
on delete, indicating neutron unavailable.
At first I thought it was https://review.openstack.org/#/c/89645/ but
that fix is in the codebase already.
It seems to me that there should be some better backoff/retry logic,
e.g. use the Ethernet algorithm of pick a random delay, wait that
amount, then double each time if there is a problem connecting.
Is anyone else seeing sporadic failures on delete when under load like this?
More information about the Openstack