| *** han-guangyu is now known as Guest3829 | 03:30 | |
| *** Guest3829 is now known as han-guangyu | 03:32 | |
| *** han-guangyu is now known as Guest3830 | 03:34 | |
| *** Guest3830 is now known as han-guangyu | 03:35 | |
| *** Guest0 is now known as osmanlicilegi | 04:11 | |
| opendevreview | Sahid Orentino Ferdjaoui proposed openstack/nova master: compute: enhance compute evacuate instance to support target state https://review.opendev.org/c/openstack/nova/+/858383 | 09:34 |
|---|---|---|
| opendevreview | Sahid Orentino Ferdjaoui proposed openstack/nova master: api: extend evacuate instance to support target state https://review.opendev.org/c/openstack/nova/+/858384 | 09:34 |
| sean-k-mooney1 | sahid: ill try to summerise the ptg discussion on the spec review but when we discussed you feature we agreed on two high level things. | 09:54 |
| sean-k-mooney1 | 1.) we wont ask you to fix all the drivers to not start the vms, we will file that as an existing bug and adress that later although your free to do that too once your feature is complete | 09:55 |
| sean-k-mooney1 | 2.) we dont wat to add a new api parmater. instead we woudl prefer the new micro verion to just make evacuate always evacuate to powered off | 09:55 |
| *** sean-k-mooney1 is now known as sean-k-mooney | 09:55 | |
| sean-k-mooney | sahid: what that means for you is all the rpc changes you made are still correct and can be resused but you do not need to add a new api parmater, the new microversion means set the target state to stopped and old means dont pass it for the rpc call. | 09:57 |
| sean-k-mooney | ill summerise this in the spec later today | 09:58 |
| sean-k-mooney | sahid: it was basicaly the last topic of the ptg https://etherpad.opendev.org/p/nova-antelope-ptg#L451 | 09:58 |
| sahid_ | sean-k-mooney: o/ I had some connectivty issues, sorry for that, could you copy/past your last messages addressed to me? | 09:59 |
| frickler | sahid_: unless you are referring to private messages, you could also check the channel log at https://meetings.opendev.org/irclogs/%23openstack-nova/latest.log.html | 10:17 |
| sahid | frickler: ack thanks | 12:18 |
| sahid | sean-k-mooney: i see, so now the prefererenc for evacuate would be to always evacuate to shutoff, right? | 12:24 |
| sahid | btw thank to give me the chance to not update all the virt drivers :-) Should I open the bug report or have you already assigned someone during PTG to open it? | 12:26 |
| rm_work | hey, hoping someone can help me understand how the fault reporting system works when retries are enabled... it seems like the "fault" field always just says "MaxRetriesExceeded" and gives that traceback, when I'd hope to be able to see the REASON for the failure (at least the latest failure)? | 12:45 |
| rm_work | I am not sure how that would work, but I'm sure there must be some way to make that happen (and maybe we messed it up internally) because otherwise the "fault" field seems like it'd be pretty useless? | 12:45 |
| rm_work | or maybe I'm expecting the wrong thing here, maybe that's right depending on the type of error? | 12:48 |
| rm_work | I guess it's storing each fault in the DB on the server object somehow when it happens, so it'd be the last failure recorded that gets shown in the `server show` return? | 12:52 |
| sean-k-mooney | sahid: we kind of felt it would be unfair to ask you do do that since its not actully required for your feature. its more just existing tech debt. sure please feel free to file one if not i can do it later. and yes the prefecne is to always evacuate to off when using the new microversion | 13:01 |
| sean-k-mooney | baiscally because of a data integrety, power usage and encypted voluems we felt it would be better to alwasy evacuate as off | 13:02 |
| sean-k-mooney | since evacuate is an admin only opeartion its not reasonable to assume the admin knows if its safe to restart the workload or not escially since there may be a posiablity of data curruption depending on howt eh host fails | 13:03 |
| sean-k-mooney | so if we alwasy evacuate to off then the enduser who should know can decied if they want to start, rescue or delete the instace | 13:04 |
| sean-k-mooney | sahid: for the encypted volume case it would be niceif you could handel and ignore the excption we get during spawn but that can be a follow up patch | 13:05 |
| sean-k-mooney | sahid: evenutlaly we woudl like to split up rebuild/evacuate int 3 steps internally. 1 stop the instace, 2 rebuild the disk, 3 start it if requried. | 13:05 |
| sahid_ | sean-k-mooney: if I understand, making it stop by default (the current spec), then in future have the internal clean that you are mentionning, and finally give users ability to choice active/stop, right? | 13:09 |
| sean-k-mooney | no we dont want to give user a choice explictly. if they want the old behaivor they can use the old microverion but going forward we want evacuate to always mean evacuate to powerered off with the new micorversion | 13:11 |
| sean-k-mooney | well its exiplcit in that its contorlled by the microversion but we dont want a new api parmater for it | 13:12 |
| sean-k-mooney | so short term (A cycle) just allow evacuate to powered off | 13:12 |
| sean-k-mooney | medeium term bug fix to not try and start the vm when the intended state is off as a non backporatble change to all drivers | 13:13 |
| sean-k-mooney | that can be done in the A or later cycles | 13:14 |
| sean-k-mooney | sahid_: does that sound ok to you. | 13:14 |
| sahid_ | sean-k-mooney: yes understood, sounds good | 13:26 |
| sahid_ | thank you for your help on it, I will update the spec to reflect that | 13:26 |
| sahid_ | also open an issue | 13:27 |
| sean-k-mooney | cool once updated ill be happy to re review | 13:27 |
| sahid_ | ++ | 13:27 |
| *** dasm|off is now known as dasm | 13:28 | |
| *** dasm is now known as dasm|rover | 13:28 | |
| sahid_ | are we agree with that I will still have to change the RPC and API to pass a targetState that will be based on the microversion, right? | 13:28 |
| * sahid_ thinking about the impl | 13:29 | |
| sean-k-mooney | yep | 13:29 |
| sean-k-mooney | so you can keep all that code that you already wrote | 13:30 |
| sean-k-mooney | its jsut that instead of having a parmater in the payload to set targetState | 13:30 |
| sean-k-mooney | you will set it based on the micoversion | 13:30 |
| sean-k-mooney | and with the new micorverison it will hardoced to off | 13:30 |
| sean-k-mooney | with the old micorversion you dont need to pass it at all | 13:31 |
| sahid_ | yeah i understand the whole process, perfect | 13:31 |
| bauzas | (just typing here again, I was in the wrong chan) | 14:22 |
| bauzas | man, my fingers bleed | 14:22 |
| bauzas | just a bit of warning, I tried to summarize as much as I can for the PTG recap, but this will be a very long thread | 14:22 |
| bauzas | I'm at 80% of the writing but I need to caffeinate | 14:22 |
| sean-k-mooney | :) | 14:26 |
| sean-k-mooney | well hopefully that means it was productive | 14:26 |
| dansmith | bauzas: we already have --purge for archive that does the purge immediately, right? | 17:26 |
| sean-k-mooney | dansmith: yep | 17:27 |
| sean-k-mooney | dansmith: but we would like to have --deleted for pruge | 17:28 |
| sean-k-mooney | so we can eventrully remove the use of arcive and make our purge work like all the opter project | 17:28 |
| sean-k-mooney | the other service dont have shadow tables so there perge remove the deleted rows | 17:28 |
| dansmith | so instead of archive --purge, purge --deleted ? | 17:28 |
| sean-k-mooney | yep basically that | 17:29 |
| sean-k-mooney | thats a lower priority item but purge everywhere else is the command to remove the deleted rows | 17:29 |
| dansmith | meh, but okay | 17:29 |
| *** dasm|rover is now known as dasm | 22:13 | |
| *** dasm is now known as dasm|off | 22:14 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!