15:29:39 <colindixon> #startmeeting clustering
15:29:39 <odl_meetbot> Meeting started Tue Aug  2 15:29:39 2016 UTC.  The chair is colindixon. Information about MeetBot at http://ci.openstack.org/meetbot.html.
15:29:39 <odl_meetbot> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:29:39 <odl_meetbot> The meeting name has been set to 'clustering'
15:29:56 <colindixon> #topic note that there was a lot more stuff talked about befor I started taking notes
15:30:00 <colindixon> #topic weaklyup
15:30:46 <colindixon> #info TomP says that he looked at weakly up, and it only allows new nodes to come in without needing all nodes to come back up, not allowing existing nodes to come back until they're all up
15:31:06 <colindixon> #info so it doesn't do quite what we'd hoped, but it still probably should be on
15:33:11 <colindixon> #info we were hoping this would resolve the fact that we need auto-down on to be able to handle the situation when 2 of 3 nodes go down, and then only one comes back
15:33:35 <colindixon> #info we have to disable auto-down, to avoid having the issue of cluster "islands" on their own
15:34:02 <colindixon> #info the solution is to manually down nodes if you need to make progress while one node remains unreachable
15:36:31 <colindixon> #topic election converence time
15:37:14 <colindixon> #link https://git.opendaylight.org/gerrit/#/c/42969/ TomP submitted this pach with explanation in the commit log
15:39:19 <colindixon> #info TomP says this can take minutues to converge if the first person to come up is behind the second
15:42:40 <colindixon> #topic robert on traying to improve stability under load
15:43:01 <colindixon> #info the stuff aorund snapshots and payload applicat is in progress, but is somewhat complext
15:43:17 <colindixon> #info there are at least 2 snapshot application code-paths and 3 payload code paths
15:49:42 <colindixon> #info there is conversation about making sure that the SchemaContext is the same on all nodes
15:51:21 <colindixon> #info example if the features installed on every node aren't identical, so different models are loaded
16:00:50 <colindixon> #info TomP says he'd like to avoid trying to cram this in for RC0 of Boron, targeting early Carbon and maybe a cherry-pick to Boron-SR1
16:07:01 <colindixon> #endmeeting