[05:44:16] I will enable query logging on zarcillo while monitoring its output size for a while [08:06:10] open question: is there a reason for us to keep zarcillo and orchestrator on bare-metal hardware instead of ganeti instances? [08:08:36] well, DBs and DRBD doesn't generally get very well [08:09:25] in i/o intensive case, I totally agree, but those 2 specific instances are quite light afaict [08:09:41] plus I would dare to say that the uptime and reliability of a master-replica setup is higher than that of ganeti- it works well for a stateless system, not for a stateful one [08:10:45] challenge accepted :P [08:10:53] I think your issue may be you are using the wrong tooling [08:11:20] oh no I don't have issue, I just checked the CPU frequency while I enabled query logging [08:11:21] if you had enabled p_s/sys you would have almost already finished your objective 0:-D [08:12:20] will get to it on that angle as well! [08:13:18] I was meaning: CPU is at 800MHz on db1215, I think it would be even slower if it could to sustain those 2 dbs activity [08:14:59] my take was: given there is little activity on those db, and having hosted that kind of mariadb workload on drdb backed storage on ganeti, I was curious if it was ever considered ^^ [08:26:24] Emperor: I think you are fighting worse issues right now, but I wonder if you saw T377853 [08:26:25] T377853: RAID monitoring on new hardware spec requires new or updated user space cli tool - https://phabricator.wikimedia.org/T377853 [08:40:37] fwiw netbox's postgres runs on VMs since the beginning :) [08:41:39] jynus: I've not got that far (and don't actually use RAID disks any more, just JBOD); I dunno if megacli works on your host, it's what I use on other ms backends [08:41:55] no, it doesn't [08:42:11] they are all technically the same tool, but vendor-dependant [08:42:33] and in fact, the same chipset from one we had [08:42:53] :sadpanda: it's a similar chipset to the other ones, so I thought it might [08:43:07] no no, it is exactly the same vendor id [08:43:16] but the user space tool seems to ignore it [08:44:12] either the host must be broken or it is just an on-purpose ignore based on the board of the card/vendor id [08:44:40] because when I use the chipsect manufacturer tool, it works for both [08:47:57] arnaudb, jynus: ok for me to test the depool/pool cookbook again today on db1185? nothing on the map of database maintenance [08:48:14] I am not a blocker [08:51:21] same here volans! happy testing [08:52:00] thanks! [20:18:06] Amir1: thanks a lot for deploying. I checked and unfortunately there is more needed (in addition, patch was fine but not enough). I left comments on Gerrit and Phab and linked them.