[08:22:21] <jynus>	 db1055 is having some strange behaviour: https://logstash.wikimedia.org/#dashboard/temp/AU8_5nIpOkQDz4dScnIo
[08:22:43] <jynus>	 with 10-second spikes of refused connections
[08:31:11] <jynus>	 similar behaviour by db1060
[08:35:51] <jynus>	 seems to correspond with some extra innodb locking/contention, without very few extra concurrent connections
[10:15:35] <springle>	 jynus: some reviews will come to you this week, from me, about T108850. probably just some pt-archiver wrapper
[10:16:32] <jynus>	 nice
[10:17:01] <jynus>	 springle, do you know something about dbproxy1003 (phab) not being in use?
[10:18:03] <springle>	 yeah, we setup dbproxy1003 for phab but iirc chasemp and I put it on hold while early phab issues were being resolved, to keep variables low
[10:18:19] <jynus>	 ok, just needed that
[10:18:25] <springle>	 so, do with it whatever you think
[10:18:45] <springle>	 more pressing need is (imo) to make phab use slave
[10:18:52] <jynus>	 yep
[10:18:55] <springle>	 or maybe it's better at that now
[10:19:12] <jynus>	 we had to increase the 500 max_connctions limit
[10:19:20] <jynus>	 because "normal traffic"
[10:19:27] <springle>	 hmm
[10:20:11] <jynus>	 I told them the typical thing
[10:20:41] <springle>	 yeah
[10:20:41] <jynus>	 but it seams that it has some connection pooling at client side
[10:20:58] <jynus>	 so they are not really 500 active connections
[10:23:15] <jynus>	 also, as it is a dedicated machine, it is not a huge issue- if the service dies, it dies on it own
[10:23:36] <springle>	 a clean death. a good death.
[10:23:50] <springle>	 a death of which to be proud
[10:24:14] <jynus>	 it is not like if I see max_connections on production I would increase it :-)
[10:24:30] * springle has started on beers
[10:24:42] <springle>	 hehe no doubt
[10:25:38] <springle>	 you prefer pt- tools over events, right?
[10:25:49] <jynus>	 wait, at the door
[10:25:59] <springle>	 k
[10:26:42] <jynus>	 well, I understand why you like events
[10:27:11] <jynus>	 not a huge preference, as far as any of them are puppetized
[10:27:32] <springle>	 fair enough
[10:30:02] <jynus>	 I do not like event because of master-slave and binlog issues
[10:30:41] <jynus>	 so on one way, they are tightly bound to a server, on the other way, they are tightly bound to the service
[10:32:26] <springle>	 nod
[10:36:23] <jynus>	 I talked to Tim yesterday (your today) about the SEx servers
[10:37:21] <jynus>	 mentioned your idea of using TokuDB
[10:38:10] <jynus>	 he is open about it, but I see to issues: 1) the home-brew compression is way better
[10:38:41] <jynus>	 and 2) T109069