| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-02-10 | |||
| 19:28:52 | superdan | mordred: like you booted this instance with flavor foo, and that flavor had 24TB of disk, but foo now has no disk and 64MB of ram, but we report that the instance's flavor is "foo" ? | |
| 19:29:12 | mordred | yah. I mean, the operator who did that is an operator who hates their users | |
| 19:29:30 | sdague | superdan: in the conversation the other day we said we'd document it as "potentially highly fictional" | |
| 19:29:31 | mordred | but as a user who booted the server in the first place, I booted a "foo" server | |
| 19:29:43 | superdan | mordred: for what it's worth, horizon makes the above trivial to do | |
| 19:29:54 | mordred | superdan: that's really sad for everyone | |
| 19:30:05 | sdague | superdan: it does, but there are also lots of people that don't change the meanings of their flavors | |
| 19:30:06 | mordred | I think of flavors sort of like server model numbers | |
| 19:30:16 | mordred | when you order a DL380 from HPE, you get a DL380 | |
| 19:30:19 | superdan | mordred: right, horizon says "you can edit flavors" but what it really means is "I will delete and recreate it for you" | |
| 19:30:19 | mordred | it doesn't stop being a dl380 | |
| 19:30:31 | mordred | but next year, hpe will sell you a new thing called a dl380 | |
| 19:31:01 | sdague | so, it feels fine to include it, but say that the underlying definition may have changed, hence why we give you the definition of what it once was as well | |
| 19:31:09 | leakypipes | mordred: even though you can order 10 DL380s and you'll get seven different motherboard manufacturers? ;) | |
| 19:31:09 | mordred | ++ | |
| 19:31:25 | mordred | (to both of you) | |
| 19:31:30 | sdague | I also think mordred should open a bug with horizon and get them to remove that feature | |
| 19:31:43 | sdague | because, it's terrible as an easy thing to tell users to do | |
| 19:31:51 | mordred | sdague: can we just make flavors undeletable instead? | |
| 19:32:00 | superdan | we've had bugs reported because they changed the size of the flavor foo and the instances didn't change and that was surprising to them | |
| 19:32:10 | mordred | _wow_ | |
| 19:32:30 | sdague | superdan: right, given the way horizon exposes it, I'm not surprised | |
| 19:32:34 | superdan | yeah | |
| 19:32:38 | sdague | it's actually extra terrible UX | |
| 19:33:24 | superdan | leakypipes: before we go expose flavors in their current form, are we not supposed to at least consider reporting them in an RP-friendly way? like so that when we get rid of the current notion of a flavor we're not pinned to this format? | |
| 19:33:58 | superdan | leakypipes: like, does it matter if we just tell them how much DISK_GB they got and not necessarily report the split between swap and root and ephemeral? | |
| 19:34:02 | sdague | mordred: so even if we decided to make flavors undeletable (which I think has other fallout) it would have to be in a microversion, which means it would be a very long time before that was fundamental behavior | |
| 19:34:15 | superdan | we can't make them undeletable, IMHO | |
| 19:34:20 | sdague | mordred: honestly, it is better to get the horizon expectations reset | |
| 19:34:21 | mordred | yah. I see that ship as having sailed | |
| 19:34:23 | superdan | that's also punching ops in the face | |
| 19:34:39 | mordred | fwiw, I do not feel _strongly_ about name or id. I have a contract with shade users that I'll always be providing an id key in the flavor dict of a server, but I can easily just plop a none in there if it goes away from the server record | |
| 19:35:08 | superdan | mordred: in some sense, your flavor id becomes the uuid of the instance itself after boot | |
| 19:35:09 | mordred | so, in the realm of "mordred will scream and yell and rant for years" and "mordred will just sigh and have a beer" 0 this is WAY much more in the beer stage than the other | |
| 19:35:19 | superdan | mordred: if we later allow you to "buy an extra gig of ram".. that's what it'll be | |
| 19:35:22 | mordred | superdan: I can see that argument | |
| 19:35:57 | sdague | superdan: so, while it could some day | |
| 19:36:02 | sdague | it really doesn't today | |
| 19:36:14 | superdan | sdague: what doesn't? | |
| 19:36:35 | sdague | it feels like there is definitely a common case of "I want another like that one" | |
| 19:36:52 | superdan | sdague: that's even worse, because we know that id can be wrong | |
| 19:36:57 | superdan | if you teach people to ask for another just like it, | |
| 19:36:58 | sdague | and if people didn't rename all their flavors, or hot resize | |
| 19:37:02 | sdague | then it can work | |
| 19:37:03 | superdan | they're going to get a 0disk 64mb instance | |
| 19:37:22 | superdan | didn't johnthetubaguy tell me that RAX has renamed flavors before? | |
| 19:37:32 | sdague | it feels like the moment you hot added anything to the instance, the flavor id flips to uuid of the instance | |
| 19:37:46 | sdague | so we actively decouple there | |
| 19:38:08 | superdan | sdague: but if you change/delete a flavor, it's the same thing, and we can't really query for any instances that need changing if you alter the flavor | |
| 19:38:13 | superdan | that's hugely expensive | |
| 19:38:24 | sdague | superdan: sure, and that's why we document it as "possibly fiction" | |
| 19:38:26 | superdan | bigly expensive I mean | |
| 19:38:29 | mriedem | SO | |
| 19:38:34 | mriedem | besides placement and cells v2, | |
| 19:38:45 | mriedem | what's the #3 newsworthy feature delivered in ocata | |
| 19:38:46 | mriedem | ? | |
| 19:38:52 | superdan | sdague: well, I just don't see the point in documenting a thing as "likely wrong" when we could just not expose it, but whatever | |
| 19:39:01 | mriedem | https://specs.openstack.org/openstack/nova-specs/specs/ocata/index.html | |
| 19:39:22 | superdan | mriedem: "leakypipes posted an inappropriate picture in IRC" is a pike event I guess | |
| 19:39:56 | cfriesen | superdan: I kind of feel like without either name or id, we don't really have a good starting point for "find me another flavor like this one" | |
| 19:40:01 | mriedem | there was a lot of random feature parity stuff that got in for hyper-v, xen and ironic virt drivers, but i can't really say that's one thing, unless i group them together | |
| 19:40:27 | superdan | cfriesen: you compare the resources not the id.. the id _is_ fiction | |
| 19:40:43 | mriedem | there is actually a lot in https://specs.openstack.org/openstack/nova-specs/specs/ocata/index.html which didn't make the release | |
| 19:40:54 | superdan | cfriesen: to put it in a way that mriedem and sdague will love: how do you know if your cloud provider doesn't dickpunch people by changing flavors? | |
| 19:41:03 | superdan | cfriesen: you don't, so you can never rely on it, thus we should never expose it | |
| 19:41:15 | superdan | er, I meant s/mriedem/mordred/ above | |
| 19:41:18 | mriedem | yeah... | |
| 19:41:25 | cfriesen | superdan: you'd still have to check it it matches, but it gives you something to check rather than starting from scratch and searching all the flavors. | |
| 19:41:26 | mriedem | i love a good dickpunch | |
| 19:41:29 | mriedem | makes me feel alive | |
| 19:41:48 | superdan | cfriesen: but you don't know if your provider is good, so you always have to check the resources anyway | |
| 19:42:21 | cfriesen | superdan: you can start by checking whether that particular flavor still matches, rather than starting from scratch. | |
| 19:42:24 | superdan | mriedem: we re-released device tags? :D | |
| 19:42:33 | mriedem | superdan: that was for attach | |
| 19:42:38 | mriedem | attach vif, attach volume + tags | |
| 19:42:40 | mriedem | which didn't happen | |
| 19:42:53 | mriedem | the original spec covered boot and attach | |
| 19:42:55 | superdan | mriedem: well, I was just trying to be funny anyway, so thanks for ruining it | |
| 19:43:05 | superdan | mriedem: no, I thought the re-release thing was for boot | |
| 19:43:05 | mriedem | oh | |
| 19:43:15 | superdan | mriedem: because get me a network stepped on us | |
| 19:43:22 | mriedem | i'm talking about this one https://specs.openstack.org/openstack/nova-specs/specs/ocata/approved/virt-device-role-tagging.html | |
| 19:43:31 | superdan | I'm talking about the bugfix | |
| 19:43:32 | superdan | anyway | |
| 19:43:33 | mriedem | yeah https://specs.openstack.org/openstack/nova-specs/specs/ocata/approved/fix-tag-attribute-disappearing.html | |
| 19:43:36 | mriedem | har har | |
| 19:43:59 | mriedem | nova-status upgrade check was a thing | |
| 19:44:04 | superdan | yeah | |
| 19:44:05 | mriedem | came in late, seems fun | |
| 19:44:09 | mriedem | no bp | |
| 19:44:14 | superdan | and it's good press | |
| 19:44:54 | mriedem | yeah, because no one cares about being able to soft reboot ironic instances :) | |
| 19:45:21 | superdan | heh | |
| 19:51:20 | superdan | mriedem: so this is the cellsv1 regression I found whilst removing the flavor compat, which I pulled out into a separate fix: https://review.openstack.org/#/c/432352/1 | |
| 19:51:27 | superdan | i think we need that back in newton | |
| 19:53:35 | mriedem | file a bug? | |
| 19:54:53 | mriedem | leakypipes: we don't support scheduling decisions based on shared storage in ocata right? we never leveraged the aggregates stuff yet from whati remmeber | |
| 19:54:55 | mriedem | *remember | |
| 19:55:04 | leakypipes | mriedem: that is correct. | |
| 19:55:55 | leakypipes | mriedem: this merged: https://review.openstack.org/#/c/407309/. but it's only the beginning of what is needed. | |