If we have more than one live-migration occurring at a time, some will fail. Basically the concluding handshake won't complete. It's not clear if this is a contention issue, and issue in Juno (which has some well-documented messaging problems), or something else.
These issues occur when trying to use l-m to evac a node (of say 8 VMs). There really isn't a load consideration, it seems to solely be a messaging issue. Workarounds: we were able to work-around this issue by doing a single live-migration at a time and ensuring it completes (is no longer marked as migrating) prior to initiating the next one. Resolved our issue completely but has the side-effect of taking a very long time. We have not bisected the problem to see if we can do more than that or do each migration for a given node singly but actually do more than one node at once. These are exercises we will undertake very soon. On Tue, Jun 30, 2015 at 11:50 AM, Auld, Will <will.a...@intel.com> wrote: > We’d like to get a better handle on issues around live migration and I > thought that all of you may be the most knowledgeable people. > > > > 1. In your own work do you see issues with live migration? What > issues and how critical are they? > > 2. How frequently do these issues occur? > > 3. What fixes or workarounds have you employed for these issues? > > > > Thanks, > > > > Will > > _______________________________________________ > OpenStack-operators mailing list > OpenStack-operators@lists.openstack.org > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators > >
_______________________________________________ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators