17:00:52 #startmeeting Ansible Testing Working Group 17:00:52 Meeting started Thu Oct 11 17:00:52 2018 UTC. 17:00:52 This meeting is logged and archived in a public location. 17:00:52 The chair is mattclay. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:00:52 Useful Commands: #action #agreed #halp #info #idea #link #topic. 17:00:52 The meeting name has been set to 'ansible_testing_working_group' 17:01:18 o/ 17:01:24 #info Agenda: https://github.com/ansible/community/issues/248 17:01:40 #chair gundalow pabelanger 17:01:40 Current chairs: gundalow mattclay pabelanger 17:01:56 There is nothing new on the agenda for today. 17:02:00 #topic Open Floor 17:02:21 * gundalow waves 17:03:07 only question I had was, do we know why https://github.com/ansible/ansible/pull/46686 broke ansible-network? It seems that ansible-test didn't catch it on stable-2.7 for some reason 17:04:06 * mattclay looks 17:04:19 trying to find log of failure now 17:04:44 pabelanger: Hi, got a link showing what it broke? 17:04:56 https://object-storage-ca-ymq-1.vexxhost.net/v1/a0b4156a37f9453eb4ec7db5422272df/logs/61/161/8811951469c29d37304936b16b095f076e107f40/check/ansible-test-sanity/1448c09/job-output.html#l672 17:05:37 I'm not yet up to speed on ansible-network sanity test, but that caused our jobs to fail and zuul wouldn't allow merging 17:05:52 what ended up happening was developers force merged the PRs, by passing tests 17:06:00 I don't think that PR you linked to is responsible for the failure. 17:06:16 mattclay: not the failure, but was the fix 17:06:35 Oh, sorry, I misunderstood then. 17:06:36 The Zuul ansible-test sanity test copies the PR under test on top of an Ansible checkout then runs `ansible-test sanity` 17:06:44 https://github.com/ansible-network/network-engine/pull/160 17:06:52 shows the actually PR in ansible/ansible fixing the job 17:07:04 via depends-on header (cross project testing) 17:07:17 OK, so what broke and where? 17:08:01 I think what happen was, the ansible-network modules were using iterate funciton, but stable-2.7 didn't have it 17:08:05 and sanity caught it 17:08:17 but, I guess this is the only module so far that was trying to use it? 17:08:43 OK, so where's the test that failed? Is that the vexxhost link you provided? 17:08:50 ya 17:08:52 https://object-storage-ca-ymq-1.vexxhost.net/v1/a0b4156a37f9453eb4ec7db5422272df/logs/61/161/8811951469c29d37304936b16b095f076e107f40/check/ansible-test-sanity/1448c09/job-output.html#l671 17:08:56 is an example of the pylint failure 17:09:00 That is where the failure was first seen? 17:09:32 I'll need to confirm with trishnag 17:09:48 but our jobs in ansible-network started failing because of it for some reason 17:10:24 lib/ansible/plugins/action/command_parser.py:347:45 I guess that is coming from ansible/ansible? 17:10:29 OK, I think I see what happened. 17:10:32 not the ansible-network role 17:10:59 I have to drop off 17:11:08 We added a new sanity test to ansible/ansible. The next time the ansible-network tests ran they used the updated test and started showing errors. 17:11:57 mattclay: okay, yah. Any idea why ansible/ansible didn't see the error on the update test? 17:12:48 pabelanger: No ansible-network tests were executed for that PR. 17:13:01 With "that PR" being the one that added the sanity test. 17:13:09 oh, yah. That will do it 17:13:20 is that something we can add moving forward? 17:13:58 I think the problem is that ansible-network is using HEAD from an ansible/ansible branch for testing. It needs to be pinned to a specific SHA. 17:14:23 Then the SHA needs to be periodically updated in a PR so that any necessary fixes can be made to ansible-network at the same time. 17:14:52 yah, we are pulling latest HEAD for our testing 17:15:04 how do we know which sha to use? 17:15:54 Pick the latest one that's passing now and use that. Then someone needs to regularly pick the latest SHA and update it using a PR. If the tests pass, great. If not, update the PR again with the necessary fixes so that the tests pass. That way you won't have spontaneous test failures. 17:16:27 It's the equivalent of doing a pip install in CI. If you use the latest release you'll encounter new failures from time to time, so you instead pin to to a specific version and update the version periodically. 17:16:54 doesn't that mean we'll also pin ansible to a specific sha? 17:16:58 If updating the version causes failures then you have an opportunity to fix those in the same PR that updates to a newer version. 17:17:06 Yes, that's the whole point. 17:17:36 Either you're using a known good version of ansible (by SHA or tag) or you're using HEAD and prone to periodic failures. 17:17:48 okay, that is a different workflow we have today. We are testing against latest HEAD, I am not sure how important that is for ansible-network. 17:18:42 could we in ansible-network comment back on a PR for ansible-test changes? eg: the PR you mentioned that changed tests 17:18:48 also, do you mind linking it? 17:19:32 https://github.com/ansible/ansible/pull/46557 17:20:39 okay, I'll review that shortly 17:21:32 You could report back results. However, then the issue we have is that the fixes can't be made in the PR since it's for a differentn repo. 17:23:39 mattclay: yah, I think the issue is, that was a break change for us. Which is fine, but we didn't get any heads up about it. Not sure how best to protect in the future in this case. You are right, we can provider feedback but if pylint in adding a new rule, we just need to be aware of that 17:24:15 thanks for info, that helps me understand why this broke 17:24:34 back 17:25:00 Yup, sanity in Zuul currently runs from HEAD of devel. The integration tests run from `devel`, `stable-2.7`, `stable-2.6` 17:25:02 I'm thinking we may need more than one version of the ansible/ansible repo for testing ansible-network. It may be appropriate to pin the version used for the sanity tests and use HEAD for the integration tests. 17:25:58 That way when sanity tests change you have an opportunity to update the ansible-network code to the new standards when updating the pinned version, while still being able to run integration tests against HEAD to catch conflicts with changes to the core engine. 17:27:09 yah, that is fine. But this seemed to happen on a stable branch of ansible. I wouldn't have expected modules to have to change APIs in this case 17:27:19 devel, I would fully support updates for pylint there 17:28:06 I have to run to next meeting, will catch up on backscroll in a bit 17:28:30 pabelanger: It wasn't changed in stable, only devel. 17:30:04 gundalow: Are sanity tests always being run using the devel branch of ansible? 17:30:22 Oh, yeah, I see you answered that already. 17:32:10 gundalow: So if I understand correctly the ansible-network tests are running sanity using ansible devel, and then integration tests using ansible devel, stable-2.7 and stable-2.6 ? 17:32:32 Using HEAD from all of those branches. 17:36:28 mattclay: correct 17:36:36 hum, we pip install 17:36:42 sorry 17:36:56 we pip install @stable-2.x for integration tests 17:38:03 gundalow pabelanger: OK, so then all you need to do is use a stable branch or specific devel SHA for sanity tests and then update it whenever you're ready to update the ansible-network code to newer standards. 17:38:21 https://github.com/ansible-network/ansible-zuul-jobs/blob/master/zuul.d/jobs.yaml#L21 17:38:34 As long as you're using devel:HEAD for sanity tests you'll fail periodically when tests are added/updated, as would be expected. 17:39:21 Yup 17:40:24 No need to try running ansible-network tests every time we make changes to ansible-test. 17:41:04 yup 17:41:34 OK, I think that resolves that issue. Is there anything else to discuss? 17:43:05 Well, I guess there is one other thing to think about. Once you pin to a stable branch or specific SHA, when are you going to update it to keep up with newer sanity tests? 17:44:31 gundalow: How are you handling branches for ansible-network? Are there separate branches to match ansible releases or something else? 17:49:44 mattclay: correct, will find a link 17:50:00 devel: https://github.com/ansible-network/network-engine/blob/devel/.zuul.yaml 17:50:18 stable-2.7: https://github.com/ansible-network/network-engine/blob/stable-2.6/.zuul.yaml 17:51:40 gundalow: OK, so you'll want to use the matching branch in ansible for the sanity tests in network-engine. 17:52:11 stable*:HEAD makes sense for the stable branches for sanity tests. 17:52:19 That's all fine I think 17:52:31 For devel, you'll probably want to pin to a SHA rather than using HEAD. 17:52:44 But then someone will need to periodically update that SHA. 17:52:58 Otherwise you'll fall behind and have issues when cutting the next stable branch. 17:53:35 So you'll probably want to update the devel SHA several times during development for the next release, including around the first beta or RC. 17:54:08 Yup, that's a fair point 17:54:46 That way you don't create a stable-2.8 branch for network-engine, only to find out it's failing stable-2.8 sanity tests because you had been testing on an older SHA that was too far out-of-date. 17:56:45 Just wondering how many times during a release we need to update the SHA, and how we remember 17:57:09 gundalow: Is it just ansible-network/network-engine that is using ansible/ansible for running tests? 17:57:37 ansible-network/* 17:57:42 same setup 17:57:47 same setup as network-engine 17:57:58 OK, so they're all following the same stable branching strategy? 17:58:07 yup 17:58:26 Is the config for which version of ansible/ansible to use for testing done at the org level or the repo level? 17:59:53 mattclay: gundalow: I'll look at our jobs again, I think you are right, we might not be setup properly on our side 18:02:46 #action pabelanger to look at jobs and consider using SHA for Devel 18:02:52 Thanks 18:03:57 Having separate stable jobs for each stable branch should be easy. For devel, you may need to have something like ansible-test-sanity-{short-git-sha-here} so you can apply it to each repo individually, as you'll likely need to make changes to that repo. 18:04:30 Yah, we'll be able to do that on zuul jobs 18:04:30 If you have only a single ansible-test-sanity-devel job then when you update it you'll break repos which will run with the new tests but haven't been updated to fix the reported issues yet. 18:04:46 ++ 18:04:52 thanks for explaining the flow 18:04:57 Then you can retire old ansible-test-sanity-{unused-git-sha} jobs when they're no longer needed. 18:06:35 pabelanger: We're just figuring it out as we go. It helps to think of the ansible-test sanity tests as though they're pylint or pycodestyle. You need to upgrade to newer versions of those tools on a repo-by-repo basis so you have an opportunity to fix the newly reported issues. 18:07:09 But in this case instead of pip installing a tool by version you're installing from a git SHA. 18:07:36 pip installing for integration tests. Sanity tests currently pull HEAD via git clone 18:07:55 Yep 18:08:15 pabelanger: OK, does that give you everything you need to work with? 18:09:37 mattclay: it does, thanks 18:10:07 We're past our scheduled meeting end time. Is there anything else to discuss before the meeting ends? 18:10:47 Nothing else from me 18:10:54 mattclay: Thanks for the info and the SHA idea 18:15:09 #endmeeting