In an HA deployment, a 60 seconds delay between reconnects can be quite
problematic. This patch changes the delay calculation by setting the max
delay to 5s and by changing the way it is increased.
Unfortunately, this is one of the places where both our main drivers are
not consistent. Rabbit's driver uses configuration parameters for this
whereas qpid's driver has never had one. However, I would prefer not
adding configuration paremeters to qpid's driver for the following
reasons:
1. Most of OpenStack services depend on the messaging layer, hence
they need it to be available. A 5s delay seems to be reasonable and
I could argue the need of tune it further. Although so frequent
reconnects can add load to the network, that wouldn't be the main
issue if one of the brokers go down.
2. We're trying to move away from configuration options towards using
transport URL. This path is still not clear and I would
prefer avoiding adding new options until we clear it out.
Reviewed: https:/ /review. openstack. org/79304 /git.openstack. org/cgit/ openstack/ oslo-incubator/ commit/ ?id=8b628d1e024 f787dbb93d50811 7d9148388c0590
Committed: https:/
Submitter: Jenkins
Branch: master
commit 8b628d1e024f787 dbb93d508117d91 48388c0590
Author: Flavio Percoco <email address hidden>
Date: Tue Feb 18 10:56:21 2014 +0100
User a more accurate max_delay for reconnects
In an HA deployment, a 60 seconds delay between reconnects can be quite
problematic. This patch changes the delay calculation by setting the max
delay to 5s and by changing the way it is increased.
Unfortunately, this is one of the places where both our main drivers are
not consistent. Rabbit's driver uses configuration parameters for this
whereas qpid's driver has never had one. However, I would prefer not
adding configuration paremeters to qpid's driver for the following
reasons:
1. Most of OpenStack services depend on the messaging layer, hence
they need it to be available. A 5s delay seems to be reasonable and
I could argue the need of tune it further. Although so frequent
reconnects can add load to the network, that wouldn't be the main
issue if one of the brokers go down.
2. We're trying to move away from configuration options towards using
transport URL. This path is still not clear and I would
prefer avoiding adding new options until we clear it out.
Closes-bug: #1281148 0acba41fdedfe22 1628f52a920
Change-Id: I537015f452eb77