[Openstack-operators] Graceful Swift reloads
bcline at softlayer.com
Wed Sep 3 21:57:07 UTC 2014
Any insights from others using this command?
From: Brian Cline [mailto:bcline at softlayer.com]
Sent: Thursday 28 August 2014 12:04
To: openstack-operators at lists.openstack.org
Subject: [Openstack-operators] Graceful Swift reloads
There isn't much documentation about caveats of the 'swift-init <daemon> reload' command, which is meant to allow existing connections to finish up before restarting.
However, I've observed this creating a number of additional processes while those connections finish up (in some cases, nearly double the amount of normal processes). Since the documentation doesn't define in detail what occurs, and the code path behind this behavior is a bit rough, two questions:
1) Does it restart a single child process only after all connections to it are complete/closed, or does it fire up the normal amount of reloaded child processes, and let the existing processes finish up? (The latter would explain the bloat in process count)
2) Are any options in place to at least keep total connection count in check such that a proxy can't consume more backend storage connections than it normally would have the capacity to?
Ultimately I'm less concerned about process count and more concerned about total connections to backend storage nodes. Even though frontend connection capacity may not be bloated as a result of those additional processes, long running/keepalive connections may still be able to cause a single proxy to consume a higher than normal amount of connections to backend storage nodes.
More clarification and details on what exactly occurs with the reload command would be most appreciated.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the OpenStack-operators