15:00:08 #startmeeting manila 15:00:09 Meeting started Thu Aug 18 15:00:08 2016 UTC and is due to finish in 60 minutes. The chair is bswartz. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:00:10 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:00:12 The meeting name has been set to 'manila' 15:00:17 Hi 15:00:19 hi 15:00:19 hello 15:00:20 hi 15:00:21 hey o/ 15:00:22 o/ 15:00:25 hello all 15:00:25 hi 15:00:28 hi 15:00:38 hi 15:00:56 #topic announcements 15:01:14 feature proposal freeze is TODAY! 15:01:43 new features proposed after 2359 UTC will get -2 for newton 15:02:08 everything for newton should be in gerrit and with +1 from jenkins by tonight 15:03:26 feature freeze is R-5, between Aug 29-Sep 02 (basically as soon as we can merge all the features) 15:04:00 #agenda https://wiki.openstack.org/wiki/Manila/Meetings 15:04:08 #topic Driver private data admin API 15:04:15 no name next to this... 15:04:25 bswartz: I can lead it 15:04:33 * bswartz grumbles nad looks into the wiki history 15:04:40 This is the new API to retrieve private driver share data 15:04:52 thanks cknight 15:04:52 The first question is whether this should be in the REST API or in manila-manage 15:04:55 hi 15:05:04 It's admin only, for debugging purposes. 15:05:19 there are testing use cases for it too 15:05:22 cknight: we already talked about that question before 15:05:31 bswartz: True, and that argues for a REST API 15:05:34 debugging and testing doesn't seem like a valid need for an API 15:05:46 gouthamr: why not? 15:06:01 this particular API defeats the point of abstraction imho 15:06:24 the counterarg is that it is needed for tempest? 15:06:24 if you start being able to see what drivers do how they do, what's the point? :s 15:06:34 well the same testing use cases could be addressed with database queries 15:06:35 gouthamr, I agree. for testing and debugging, you can access the db directly 15:06:55 for debug & manual testing one doesn't need rest, right? 15:06:58 toabctl: +1 15:07:27 if we really do need to validate, maybe provide a way to access the database.. not an API, not even admin facing.. 15:07:34 xyang2: I know we've covered this topic before but I don't remember the conclusion, do you? 15:07:56 gouthamr: +1 15:07:58 last time when we talked about this, only Valeriy had strong objections 15:08:01 cknight xyang2: any reservations using manila-manage? 15:08:14 he has since changed his vote so it allows us to move forward with REST API 15:08:18 bswartz: I remember we decided to use manila-manage 15:08:22 I can support manila-manage for this 15:08:27 gouthamr: would that still support white-box Tempest tests? 15:08:38 why after so many rounds of addressing review comments, go back to the beginning now 15:08:40 cknight: yeah, that's the question i think 15:08:41 cknight: we need a way to invoke manila-manage via tempest.. 15:08:45 o/ 15:08:58 cknight: if we do manila-manage then there's no tempest test because there's no API 15:09:03 xyang2: every round of my review has the question in it 15:09:05 :) 15:09:09 xyang2, the process is not ideal, I agree 15:09:10 we'd have to cover it with "functional" tests 15:09:18 but it's also not an argument to add it. 15:09:48 xyang2: you're still in favor of REST API for this? 15:09:49 bswartz, why do you want a tempest test for driver private data? what would that test do? 15:10:02 bswartz: yes 15:10:08 so we're saying use manila-manage instead and get coverage in functional tests rather than tempest, right? 15:10:21 toabctl: it would enable white-box tests to ensure the driver was storing what it should 15:10:24 yes.. i'm not opposed to testing this, thoroughly, but REST API isn't an answer... 15:10:26 toabctl: vendor black box testing would want to know what the driver set for the private metadata 15:10:48 cknight: s/white/black/ 15:10:58 you can easily write a script to invoke manila-manage from your white box "tempest" tests if that's your goal 15:11:18 s/white/black/ and tempest should be black, right? 15:11:24 bswartz, cknight then we should make every DB model accessable via a rest api 15:11:34 tbarron: tempest is emphatically white box tests 15:12:02 tempest knows nothing of the driver or what it does under the covers -- tempest only tests the external API surface 15:12:21 bswartz: that's what i meant to be saying too, may have inverted my colors 15:12:40 there's a legitimate issue with testing of this feature if we do add a REST API which is that you can't write a reasonable test without knowing what the driver is doing 15:12:51 oh wait 15:12:53 DAMNIT 15:12:58 :) 15:13:01 I'm getting my white and black mixed up 15:13:07 * bswartz headdesk 15:13:28 anyways, tempest deals w/ behavior and not internal mechanisms 15:13:28 bswartz: got to apply a negative filter 15:13:37 lol 15:13:43 yes please reverse all my white/black comments 15:13:57 bswartz: black tests matter 15:14:04 blackbox is what tempest does, which is only testing the outside API and not knowing the inside 15:14:33 whitebox is what downstream vendors do, where they test the internals of their driver and storage controller 15:14:47 and tempest is just the vehicle 15:15:08 gouthamr: if I get my way tempest won't be the vehicle much longer 15:15:35 bswartz: yes.. :P 15:15:39 but let's avoid that rathole 15:15:43 * gouthamr continues to write tempest tests 15:15:45 xyang2: could you get the coverage you need with functional teests and manila-manage? 15:15:47 okay so back to the topic at hand 15:16:06 tbarron: functional tests? 15:16:26 yeah 15:16:30 the proposal on the table is to implement the driver private share data as a manila-manage command, and not to add any REST APIs or tempest tests 15:16:34 tbarron: manila functional tests are tempest tests 15:16:52 tbarron: sorry I didn't get what you meant 15:16:56 we've discussed this before, and it seems that support has grown for the manila-manage approach 15:17:38 who has a link to the current patch? 15:17:52 #link https://review.openstack.org/#/c/315346/ 15:17:53 bswartz: manila-manage should be just for db migrations 15:18:06 bswartz: at least in Cinder that is what we are trying to do 15:18:23 xyang2: sorry, learning my way around, i thought we were going to do something between unit and tempest, but I was in cinder context I guess 15:18:28 well cinder doesn't have this feature, so it's not a great comparison 15:18:52 tbarron: np, we are using different terms 15:19:07 I'm not for manila-manage but against another rest api. xyang2 why can't you just so something via the manila rest api and then access the db with whatever mechanism you want to test the result? 15:19:10 bswartz: Cinder has other commands that are not db migrate 15:19:20 you could just as well argue that cinder-manage/manila-manage are just for admin-only-db-things 15:19:20 bswartz: we talked about they should go away 15:19:25 s/so/do/ 15:20:19 xyang2, to access the db during your tests, you can use sqlalchemy or even the mysql http plugin to have a DB rest api. 15:20:42 toabctl: well, tempests test should use REST API 15:20:47 toabctl: you're arguing for dropping this whole patch and blueprint 15:20:55 bswartz, hm. I think yes 15:21:09 toabctl: for tempest to access DB directly is wrong 15:21:17 xyang2: no, proper tempest test would never look at the values returned by the API because tempest doesn't know driver specific stuff 15:21:37 xyang2, see bswartz answer 15:21:41 bswartz: so proper tempest should check db directly? 15:21:50 bswartz: how does it know what to check 15:21:53 if we added an API we would need to cover it with tempest tests, but that puts us in a bind because we don't know how to write those 15:22:50 xyang2, if that is for downstream tempest tests (as written in the commit message) you can create an extra tempest plugin and do whatever you want there (also access the db directly) 15:23:13 I feel bad now because xyang and accelazh have been working on this since May and we're still arguing about the right way to do it 15:23:41 however I'm inclined to agree with toabctl 15:23:52 * toabctl feels bad, too 15:24:06 I wish Valeriy has not removed -2 or someone else has put -2 on it earlier, rather than say it now 15:24:12 but adding another API for testing sound just wrong. 15:24:16 yeah 15:24:44 done, abandoned 15:25:15 If anyone has such a strong opinion, please put -2 on it earlier 15:25:29 xyang2: sorry this didn't get addressed earlier 15:25:34 i'm sorry, i thought my -1s were not addressed.. 15:25:38 xyang2, yes. sorry for that. 15:25:38 that's what -2 is for 15:26:15 I suspect we have a lot of problem of this type where ideas are proposed and someone is opposed but doesn't provide feedback 15:26:28 it's something we should work on as a community 15:27:04 because the result is wasted time for the contributor and reviewers 15:27:08 there is difference between -1 and -2. If you think it can't be merged at all, please use your core right 15:27:34 your point is well taken xyang2 15:27:46 we have 2 other topics though so I'll move on 15:28:00 #topic Scenario tests are broken 15:28:09 #link http://logs.openstack.org/86/309286/34/check/gate-manila-tempest-dsvm-scenario/610f96e/console.html 15:28:20 no name again 15:28:28 bswartz: me 15:28:29 wiki history says ganso 15:28:53 just pointing out that something broke scenario tests. I tried to look at recently merged gerrit changes, but haven't found a suspect 15:29:06 ganso: how recently? 15:29:07 so it could be other project's commit 15:29:15 and is it consistent or random? 15:29:22 bswartz: consistent 15:29:29 bswartz: Aug 16th it was working 15:29:39 okay 15:29:40 I submitted a patch on Aug 16th, and it ran ok 15:29:48 ganso are you looking at it? 15:29:57 do we need others to jump in an help too? 15:30:13 I need it running to validate migration patches 15:30:15 if the gate is broken it will make it hard to achieve the deadline of +1 from jenkins by tonight 15:30:22 scenario job has a migration tempest tests 15:30:31 bswartz: gate is not broken, just scenario job 15:30:37 it is non-voting 15:31:25 I am not investing effort to implement a fix myself right now 15:31:36 okay well it should be voting 15:31:37 I still have to update my patches 15:31:42 that's something else to fix 15:31:59 bswartz: Valeriy was trying to improve its run time before making it voting 15:32:02 anyone have bandwidth to investigate failing scenario tests? 15:32:07 bswartz: it hasn't improved much 15:32:13 bswartz: and still fails from time to time 15:32:20 ganso: I'm working on that too, but it's on backburner 15:33:12 okay I get that everyone is working on stuff for the feature freeze, but failing tests are important 15:33:36 if this is caused by some other project's commit, then we can expect more of the same kind of problems in the next 2 weeks 15:33:50 if we want any features to get in we need to be quick about tracking down tests failures and fixing them 15:34:10 after I update my patches, I may sacrifice review time to try to debug and fix it, which if nobody does I will have to do it in the end. 15:34:39 normally vponomaryov is our hero on these kinds of problems, but he's out for another 11 days 15:35:25 okay last topic 15:35:28 the Joker usually wreaks havoc while Batman is on vacation 15:35:38 #topic Revert to snapshot 15:35:45 ganso: yep 15:35:52 ganso, Does that make you Robin? 15:35:56 #link https://review.openstack.org/#/c/340502/ 15:36:01 markstur: lol definitely not 15:36:02 #link https://review.openstack.org/#/c/356682/ 15:36:11 So we have discussed revert-to-snapshot before, and some review feedback has pointed out that there should be a required extra spec for that to enable us to preserve that attribute on the share and check for it in the API layer. 15:36:23 In considering that, we realized that 'snapshot_support' is overloaded. 15:36:31 It has always meant that a driver can 1) create snapshots, and 2) create shares from snapshots. 15:36:34 so the revert to snapshot is a "small" "major" feature which has been hanging around waiting for reviews for a while 15:36:45 So before we add new snapshot semantics, we have to fix this overload, because there will be drivers that can create snapshots but not create new shares from them. 15:36:56 We have code up that does this (#link https://review.openstack.org/#/c/356682/), with associated Tempest work underway. 15:37:09 The revert-to-snapshot patch is now atop the aforementioned one (#link https://review.openstack.org/#/c/340502/), again with needed Tempest work in progress. 15:37:26 yeah it was recently we realized nobody had broken apart the snapshot-support and create-share-from-snapshot capabilities which we agreed to in Tokyo 15:37:42 Unfortunately, the mountable snapshots patch will also need to be rebased on this. 15:37:55 we didn't NEED to split them until we had either revert-to-snapshot or mountable-snapshots, and neither of those happened in mitaka 15:38:22 so thanks to cknight for going ahead and doing the split now 15:38:32 it's a surprising large change because it affects most existing drivers 15:38:44 bswartz: glad to do it, and I had help from gouthamr 15:39:08 it's fairly easy to review though because all of the driver changes are 1 or 2 liners 15:39:36 is tpsilva here? 15:39:47 yep 15:40:06 tpsilva: you plan to rebase on top of this right? 15:40:14 bswartz: yes, working on it 15:40:19 okay 15:40:48 only other thing is to support this in first party drivers 15:40:57 vponomaryov plans to add ZFS support for this, I'll do LVM 15:41:06 tpsilva: I tried to design the code & unit tests so your new snapshot semantic would slot in pretty easily 15:41:17 tpsilva: holler if you have questions 15:41:24 bswartz: you mean revert or mountable? 15:41:35 ganso: revert 15:41:35 cknight: okay, I'm just finishing up some of the -1's before rebasing 15:41:48 ganso: I could add mountable snapshot support to LVM too if nobody is doing it 15:41:55 tpsilva: you still have tempest and LVM to go, right? 15:42:03 ganso: no guarantee on a timeline for that though 15:42:08 ganso: yes 15:42:30 bswartz: wouldn't those have to be up by today? 15:42:42 ganso: for revert it will be 15:42:43 bswartz: or are you including in the same patch? 15:42:47 bswartz: oh, ok 15:42:55 I was going to ammend cknight's patch 15:43:14 it's easier to test as 1 patch 15:44:01 the main reason I put this topic on the agenda is because we want drivers authors to review cknight's patch as it affects all existing drivers 15:44:16 make sure the new capability being advertised is correct, and if not, fix it 15:44:35 it will now be possible to support snapshots without also supporting create-share-from-snapshot 15:45:09 and we know of some drivers which have that restriction, so those will be able to add snapshot_support=True 15:45:38 #topic open discussion 15:45:40 bswartz: I suspect some driver require more drastic changes 15:45:40 i've been reading it and it looks good so far, waiting for the outstanding tempest stuff 15:45:49 bswartz: like, ceph driver, IIRC 15:46:03 bswartz: it had snapshot_support disabled because it could not support create_from_snapshot 15:46:03 * bswartz looks around for jcsp 15:46:22 * bswartz doesn't find him 15:46:42 ganso: yes, but I'd rather have the driver maintainer support those patches rather than cknight 15:46:44 i will follow up with jcsp and rraja 15:46:49 s/support/submit/ 15:47:22 bswartz: I assume separate patches won't be accept after today 15:47:26 cknight's patch just preserves the existing behavior after the capability split 15:47:30 * tbarron thinks jcsp may be on vacation 15:47:42 i'll help out 15:47:45 ganso: we can make exceptions for small things 15:47:48 rraja: ++ 15:47:53 if it's a 1 line to correct a capability then it's fine 15:47:56 1-liner 15:48:11 if it's adding support for snapshots where none existed before, that should wait for ocata 15:48:16 * tbarron thinks it would be a bug that the capability is reported wrongly :) 15:48:21 bswartz: 1-liner could be included in cknight's I guess. But "more drastic" as I said, would be left out 15:48:49 any other topics? 15:48:54 bswartz: yes.. 15:49:02 ganso: yes, I'm glad to make trivial updates during the review process 15:49:21 cknight: as long as the driver maintainer is involved 15:49:24 so, if you see the patch - you'll see the capability is called "create_share_from_snapshot_support" - that's a mouthful 15:49:33 I was hoping someone here has a better name for it 15:50:08 gouthamr: +1, the long name is crystal clear, but it makes for some ugly code wrapping 15:50:19 cknight +1 15:50:29 I like the self-documenting name 15:50:42 maybe we could drop the 'support' from it 15:50:42 i proposed "clonable" -> but bswartz and cknight say that it is suggestive of a potential feature that we may have down the road 15:50:44 but if it's too long for pep8 then that's a problem 15:51:09 *would be very glad to increase the 80-character limit* 15:51:13 tpsilva: I think the _support suffix is essential and fits with the existing snapshot_support capability 15:51:14 yeah, i'd hate to misspell it. 15:51:31 gouthamr: add bash completion for manila then... 15:51:37 we already have DHSS, how about CSFS 15:51:48 * bswartz barfs 15:51:53 tbarron: lol 15:51:53 tbarron: we don't have DHSS 15:51:53 :D 15:51:53 I like clonable, it fits perfectly with mountable and revertable 15:52:01 * tbarron doesn't like acronyms but there are reasons they exist 15:52:08 gouthamr: not officially 15:52:15 it's not a clone though 15:52:22 you are makiing one thing from another 15:52:26 can_create_share_from_snapshot 15:52:38 one kind of thing from another 15:52:41 tbarron: you're right 15:52:49 IMO, "clone" means create-share-from-share 15:52:59 bswartz: +1 15:53:05 and a snapshot may or may not be involved 15:53:22 it's a feature cinder has that I think we should copy, but it's never been very important 15:53:23 i wish it were "snapshottable", "clonable", "mountable", "revertible" 15:53:27 and if it's involved it should be as behind-the-scences mechanism 15:53:32 yeah, something better than "clonable" in this case then 15:53:59 clonable_from_snapshot ? 15:54:04 :D 15:54:06 CSFSble 15:54:12 okay well let's not debate this one forever -- if someone has feedback on this topic please take it to the review 15:54:14 cknight: but then are you cloning a snapshot to create another snapshot? 15:54:22 clone_from_snapshot_support? 15:54:40 bswartz: same as above ^ 15:54:40 * bswartz -1s is own idea 15:54:46 bswartz: lol 15:55:38 I suspect we may have to just shake our fists at the PEP8 gods and stick with the long name here 15:56:16 if it's too painful inside the code we could use an abbreviated variable name 15:56:27 but keep the full name for the API 15:56:37 any late things? 15:56:44 we're almost out of time 15:56:57 okay thanks everyone 15:57:03 #endmeeting