| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-11-22 | |||
| 19:22:26 | melwitt | or rather, I would prioritize it a lot higher than I usually would | |
| 19:22:32 | mriedem | "especially now" is a bit heavy | |
| 19:22:34 | sean-k-mooney | adding a --delete to the event list might just be the simpelest option | |
| 19:22:46 | sean-k-mooney | mriedem: well we need it for resize not right | |
| 19:22:58 | sean-k-mooney | welli guess not for deleted instnaces | |
| 19:23:09 | mriedem | we changed one api from a call to a cast which might have failed with NoValidHost (MaxRetriesExceeded) after the api had already given you a 202 | |
| 19:23:20 | mriedem | if resize rescheduling fails | |
| 19:23:22 | melwitt | "especially now" is based on my own personal experience at the org where I worked and how openstack was used there. there was no need to look at action events unless you were the auditing | |
| 19:23:36 | melwitt | -the | |
| 19:24:02 | melwitt | I recognize that other orgs end users could/are already using the api heavily | |
| 19:24:26 | mriedem | or not, | |
| 19:24:50 | mriedem | my point is just that it's not like there was a sudden tectonic shift on how nova works that makes caring about instance actions a big deal now | |
| 19:25:00 | mriedem | *in how | |
| 19:25:15 | melwitt | maybe not, but it's an end user perception and I think that's what really matters | |
| 19:25:32 | mriedem | what is an end user perception? | |
| 19:25:44 | melwitt | and again, disclaimer, I don't know how every end user in the world uses nova | |
| 19:26:26 | melwitt | I was an end user of nova on a daily basis at my last company. and I never needed the actions api for anything. if a server action failed, I looked at the server fault. that was it | |
| 19:26:31 | mriedem | if users didn't care about instance actions before, nothing has really changed recently that makes them have to start caring | |
| 19:26:37 | melwitt | that was my perception as an end user | |
| 19:26:54 | mriedem | probably b/c back in havana/icehouse/juno we put the server into error status every time anything failed | |
| 19:27:10 | melwitt | I think there is. if a resize used to fail with a 400, then how would you know to look at actions suddenly to know your resize failed scheduling? | |
| 19:27:32 | melwitt | I think that's quite different. again, just MHO | |
| 19:27:55 | melwitt | as someone who has used nova. that would have surprised me | |
| 19:27:56 | mriedem | resize could have failed *before* after you got a 202 and the first host resize claim failed and rescheduled and got NoValidHost | |
| 19:28:01 | mriedem | to know that, you'd have to know about instance actoins | |
| 19:28:07 | melwitt | *shrug* | |
| 19:28:15 | melwitt | whatever, just saying my opinion | |
| 19:28:48 | melwitt | everyone else can think whatever they want | |
| 19:31:11 | dansmith | my feeling is that things should be as async as possible because that involves the least amount of tying architecture to api behavior, | |
| 19:31:15 | mriedem | i'm not trying to argue, i'm just saying nothing has dramatically changed - things have been this way for years wrt instance actions | |
| 19:31:20 | dansmith | which bit us in the ass with the cells conversion | |
| 19:31:39 | dansmith | right now we have a big mix, with some assumed behaviors from the users (i.e. that 200 from resize means something specific) | |
| 19:31:54 | dansmith | actions have been there a long time, with lots of useful information | |
| 19:32:02 | melwitt | no, I agree that more async is good. I'm just saying that to me, in my own perception, it's a shift from the past habit | |
| 19:32:26 | dansmith | if the users don't know about it, then we probably have some doc gaps, but I don't think that really changes anything | |
| 19:32:27 | melwitt | my habit was not to use instance actions. my habit will be to use instance actions from now on, now that I know | |
| 19:32:38 | dansmith | and I also don't think that actions are some obscure corner of the nova world | |
| 19:32:41 | mriedem | the shift happened years ago then, is my point | |
| 19:32:57 | melwitt | ok, mine didn't, and I figured maybe I'm not the only one | |
| 19:32:59 | dansmith | I think that some people might never go looking for it and just infer what they can from status, return codes, etc, but they're missing part of the story | |
| 19:33:01 | mriedem | after yesterday's meeting i have a todo to write a doc about using actoins | |
| 19:33:16 | mriedem | but i'm still working on this resize/cold migrate doc for artom | |
| 19:33:27 | dansmith | I totes don't see how this is a shift | |
| 19:33:39 | melwitt | I get that you guys don't | |
| 19:33:46 | dansmith | when actions became a thing, there was more information available, that's not a shift, that's just more | |
| 19:33:54 | artom | mriedem, I'll handle the sequence diagram, fwiw | |
| 19:34:18 | dansmith | inferring things from api behavior is.. a thing I guess, and over time those inferences will have to change because they're based on assumptions | |
| 19:34:45 | dansmith | so there is a shift in those inferences I guess, but not in like fundamental behavior of how one uses nova or anything | |
| 19:34:51 | melwitt | it was a shift for me as a user, that's all I'm saying. and the only reason this came up is that, it's a shift for me, I see how important it is, so therefore I have a new interest in fixing the osc command for it | |
| 19:35:43 | melwitt | it means I will change my habit in how I use nova and it might change others too | |
| 19:35:52 | mriedem | artom: i'm about to push it | |
| 19:36:00 | mriedem | artom: i don't want you putting rebuild in the diagram somewhere | |
| 19:36:04 | mriedem | ZING | |
| 19:36:10 | artom | *ded* | |
| 19:36:27 | artom | mriedem, you saw my review, btw? | |
| 19:36:31 | artom | Before you push | |
| 19:36:45 | mriedem | nope, looking | |
| 19:37:12 | artom | mriedem, nothing major, but probably better to see it before pushing again | |
| 19:37:41 | sean-k-mooney | artom: by the way ill adress your comments on my patches on monday i didnt get to them this week but i still plan to adress them shortly | |
| 19:37:58 | sean-k-mooney | but im ment ot be in limerick in 20 mins so i better go o/ | |
| 19:38:21 | artom | sean-k-mooney, drive safe | |
| 19:39:31 | dansmith | our docs should probably have something like this: | |
| 19:39:32 | dansmith | https://support.metacloud.com/hc/en-us/articles/115003922347-Using-the-Instance-Action-List-for-Troubleshooting | |
| 19:39:52 | dansmith | which doesn't even say anything about error state or fault that I know of, | |
| 19:39:52 | melwitt | let's copy it | |
| 19:40:03 | dansmith | but specifically is like "my instance is stopped, whyforisthat?" | |
| 19:40:14 | melwitt | again, maybe it's just me, I'm the lone person who never used instance actions. like a rebel | |
| 19:40:15 | dansmith | which is exactly what we're talking about I think | |
| 19:40:32 | dansmith | melwitt: I think it's entirely possible that it may just be you :P | |
| 19:40:42 | melwitt | uh huh | |
| 19:40:45 | dansmith | heh | |
| 19:41:09 | mriedem | dansmith: that's basically the kind of doc i was going to write, just cli though | |
| 19:41:21 | mriedem | and using resize as an example where the claim on the dest fails | |
| 19:41:22 | dansmith | mriedem: there's cli below | |
| 19:41:25 | dansmith | yep | |
| 19:41:30 | mriedem | using the nova clie | |
| 19:41:31 | mriedem | *cli | |
| 19:41:33 | mriedem | i'd use osc | |
| 19:41:53 | mriedem | anyway, same idea | |
| 19:41:56 | dansmith | I guess if we have no "how do I figure out why my instance is unhappy" doc, then people will make assumptions about how to (or not do) that, which might stop at the instance detail output | |
| 19:41:59 | melwitt | well, if everyone else knows, then that's ideal. I'm glad to hear it | |
| 19:42:10 | mriedem | what are instance actions, when are they useful, how to use them, give an example scenario using resize where the claim fails | |
| 19:42:29 | dansmith | mriedem: well, I think starting with the use-case of debugging an unknown state of an instance is useful, | |
| 19:42:42 | dansmith | because if people _don't_ know what actions are, they're not looking for a doc on actions | |
| 19:43:06 | dansmith | melwitt: amirite/ | |
| 19:43:46 | dansmith | maybe "unknown state" is the wrong term, but.. debugging why something failed or isn't expected on the instance, | |
| 19:43:52 | melwitt | dansmith: yeah, I think so. case in point, gregwork was here earlier and a volume attach failed and there was no indication. I told him to look at instance actions based on the resize/migration thing from yesterday | |
| 19:43:56 | dansmith | which might be a weird state and might be "resize never happend" | |
| 19:44:25 | melwitt | and in there, it showed the volume attach with 'error' as the result | |
| 19:44:38 | dansmith | melwitt: I think gregwork is just lonely and wanted to talk to someone and not read docs, so not sure that would have helped him | |
| 19:45:23 | melwitt | so there's already a doc? what did I miss | |
| 19:45:33 | mriedem | no | |
| 19:45:37 | mriedem | i was joking, | |
| 19:45:43 | melwitt | what | |
| 19:46:08 | mriedem | how do i say politely, if you guys want to write the doc(s), go ahead | |
| 19:46:45 | melwitt | I'll copy metacloud's | |
| 19:46:52 | gregwork | heh i was just reporting something strange that i observed. normally openstack is very vocal about errors that get thrown | |
| 19:47:06 | gregwork | i thought it was unusual that not a lot of fuss was happening | |
| 19:48:36 | gregwork | and ultimately heat would just fail to deploy the stack after some time (token expiry) | |
| 19:48:40 | KeithMnemonic | mridem: melwitt: i can help as well and then abandon my hack sitting out there ;-) | |