00:02:25 #startmeeting azure_working_group 00:02:25 Meeting started Thu Jan 25 00:02:25 2018 UTC. The chair is Kylie_. Information about MeetBot at http://wiki.debian.org/MeetBot. 00:02:25 Useful Commands: #action #agreed #halp #info #idea #link #topic. 00:02:25 The meeting name has been set to 'azure_working_group' 00:02:49 Two topics from me today. 00:02:54 #topic bug triage 00:03:03 https://github.com/ansible/ansible/issues/17925 00:03:03 https://github.com/ansible/ansible/issues/33448 00:03:03 https://github.com/ansible/ansible/issues/28392 00:03:03 https://github.com/ansible/ansible/issues/26195 00:03:03 https://github.com/ansible/ansible/issues/34047 00:03:03 https://github.com/ansible/ansible/issues/31105 00:03:04 Any objection to close it? 00:03:04 https://github.com/ansible/ansible/issues/29580 00:03:05 https://github.com/ansible/ansible/issues/29585 00:03:17 which- all of them? 00:04:39 yes:) We could start from 29580 and 29585. 00:05:27 I think these 2 are already in 2.4. Could you please take a look? If yes,could we close them? 00:08:06 I don't think 29585 has been implemented 00:08:50 I assume he's talking about an arg that would allow for a storage account to be specified to receive VM metrics and/or boot diagnostics. 00:08:57 You quoted data disks 00:09:28 O, ok. I thought he means metrics of data disks. 00:09:49 ok. keep it open. 00:10:38 Hi all, could you please take a look https://github.com/ansible/ansible/issues/28392? 00:10:39 Closed 29580 though, that was implemented a LONG time ago 00:11:19 What do you suggest for the behavior? If fail to execute a playbook, should Ansible clean up others which already created? 00:11:27 https://github.com/ansible/ansible/issues/28392 00:11:40 I debated that when we first built it 00:12:13 I saw some different behaviors. 00:12:14 I personally believe it should cleanup errors 00:12:47 especially when this could end up costing the user money for running in the background 00:13:08 We could definitely add "best effort" cleanup- it's just unfortunate when you get all the way to VM and have an issue, then have to create/teardown everything in between. But I guess you could always not use the auto-creation stuff anyway in that case if you want maximum performance. 00:13:44 (eg explicitly create dependent resources) 00:13:46 i have the same idea, it should clean up. 00:14:07 i think vm module is just too large though. 00:14:53 Yeah- I was hoping to clean up a bunch of things by removing support for automatic storage account creation in favor of managed disks, but until Azure Stack supports managed disks... 00:15:23 I guess we could still rip all that stuff out and just say if you're on stack, we won't create the storage account for you 00:15:42 Then someday when stack supports managed disks, it works the same everywhere 00:15:43 maybe if it cleans up everything, somethimes it my clean up too much. i believe some sometimes some resources may be created beforehand... 00:16:01 We would only clean up things we'd explcitly created in that case 00:16:30 (ie, things that the module had auto-created) 00:18:29 I'm fine with changing the behavior to clean up better- just need to make sure we don't accidentally delete resources that we didn't create. 00:18:48 (so track internally which ones were created by us and only delete those) 00:19:42 Anything more to discuss on that one? 00:19:57 ok. I think there is an agreement - clean up errors. As for how, seems need to be reviewed further. 00:20:02 Zim, can you take this one? 00:20:10 yes, i can 00:20:15 Thank you. 00:20:23 could we switch to prs 00:20:25 ? 00:20:32 Yes, please. 00:20:59 i have this keyvault pr: https://github.com/ansible/ansible/pull/35150 00:20:59 And another topic for Yuwei and Matt is to see any question about API versioning. And schedule. 00:21:29 reviewed by jborean93 already but needs nitzmahone to check 00:21:49 nitzmahone could you check and possibly merge? 00:22:24 yes; might be Monday next week though (I'm off tomorrow and Friday) 00:22:48 If I have some time waiting for tests on something else I might be able to do today 00:23:38 I'll also update the PR for API versioning on Monday next week 00:23:44 well, it's a kind of critical, so if you could briefly look and perhaps jborean93 could merge when i fix your comments 00:23:58 i have 2 more modules depending on this. 00:24:03 I have two prs without any comment 00:24:24 i am fixing https://github.com/ansible/ansible/pull/33607 00:24:49 #35037 #35038 00:24:50 and this https://github.com/ansible/ansible/pull/33606 00:25:28 these 2 modules in addition need more variables from CI 00:25:53 yuwei: I'll look at your PR's soon 00:25:54 azure_client_id and azure_ 00:26:13 zikalino: The immediate thing that made me want to look further at keyvault stuff is this statement: "All identities in the array must use the same tenant ID as the key tenant ID." 00:26:19 so i think mattclay is right person to make such changes 00:26:30 If that's the case, why is tenant ID even allowed on the rules? 00:27:51 indeed, i will hide this param 00:28:23 i couldn't find any circumstance when this should be different 00:28:59 Also, hardcoding tenant ID on the vault itself in the Ansible content is probably not the way most people would want to use it, so we should probably default to the current connection's tenant ID. Good to be able to override for more complex scenarios, but otherwise things like CLI auth are problematic as a user would have to dynamically get tenant ID somehow. 00:29:57 ok, i can add this too 00:30:10 Maybe as a special non-GUID value like "current" or something 00:30:26 (so it can also be selected dynamically) 00:30:38 i think i will just default to current 00:31:20 Anyway, just saw those things on first glance and thought I should spend some time playing with it to see if there are other usability problems 00:31:32 (as I've not used a keyvault before) 00:32:17 It's not just a matter of code review 00:32:46 then major problem people encounter with keyvault is using proper object_id 00:33:05 (often confused with application_id) 00:33:35 guys, I need to leave for a customer meeting. I don't have other topics. Thank Matt to take care of API versioning. I will ask Yuwei check your PR next Monday. Thanks. 00:33:55 (well, her Tuesday since you're many hours ahead of us) 00:34:30 Usability-wise, it'd be nice if we could look up the user object ID by some other selector, but I don't think we want to hold it up for that 00:34:33 yes:) 00:35:10 Could be added later (eg, provide a dictionary with values that could be looked up and translated to an object id) 00:35:56 As-is, I don't think we have a facts module or anything that could do that translation, so it'd be very difficult to actually use this module in a real-world scenario 00:35:58 that could be done only via sth like "az ad sp show --id " 00:36:52 if we had ansible module to do this, user could first query object_id from and then use it in this module. 00:37:06 you want to make this easy for users, not everyone who uses Ansible is a programmer 00:38:26 Yeah, something like azure_rm_serviceprincipal_facts and azure_rm_aduserfacts 00:38:36 i think object id should somehow come from common implementation. we already have credentials there, we could obtain object_id or we could just require object_id to be provided along with credentials. 00:39:48 this is not only ansible problem, i run into that when testing keys and secrets modules, and i found our several people made the same mistake (when using Azure API from difefrent SDKs) 00:40:19 That's why we try to make the modules more than just a thin wrapper over the SDK 00:40:51 (idempotence, linked resource lookup, etc) 00:40:59 "what it" scenarios 00:41:03 *if 00:42:24 Anyway, we don't necessarily have to solve that problem right away, but the module is not very usable in the real world without it IMO 00:42:55 Nobody should be hardcoding AD GUIDs into their playbooks 00:43:14 (nor required to manually look up and pass a whole bunch in as vars to execute something) 00:45:27 true 00:46:20 but in this case you have to pass some guids somehow whether it's application_id or object_id 00:46:44 yep, though application_id could be looked up the same way 00:47:01 (eg, hypothetical AD facts module or something) 00:47:11 well, that's not necessarily the same application_id as your environment has 00:47:34 right, but if you were creating something reusable, there'd be a name associated with it 00:47:37 you may need to pass 5 different application_ids or (object_ids) 00:48:10 because you may be giving access rights to 5 different applications / people 00:48:30 right- so you'd want a way to look those up programmatically by name 00:49:39 yes, that should be possible, but single app may have several service principals 00:49:43 The whole intent of Ansible is to build reusable content that can be used across many environments, so hardcoding GUIDs anywhere in the content is a problem. Those kinds of things always need to be discoverable dynamiclly 00:50:44 Right, so you'd look them up by name/tag/whatever in a previous task, then pass them into this module (or allow for a selector of some sort in the module) 00:51:19 yes, something like that, or even previous module might create service principal 00:51:27 Yep 00:52:05 Like I said, we don't have to solve that part now, but something to think about 00:52:10 well, key vault actually is not a good example of module that can be used for "ordinary users". 00:53:20 ok, so right now i think we should: 00:53:42 1) fix tenant_id thing so it's populated by default if not specified 00:54:00 2) document object_id to make sure people don't make mistakes 00:54:30 sounds good 00:54:57 If the underlying SDK accepts tenant_id for the rules, I'd still make it overridable, but also default to "current" 00:55:04 afterwards i will look into creating active directory module. 00:55:35 (there are several things like that I've run into where it seems like maybe the SDK is set up for future capability that doesn't exist today, eg, cross-tenant keyvault access) 00:57:03 hmm, Kylie didn't chair anyone- I don't think we can add #action or even end the meeting 00:57:17 @Kylie_ can you chair someone? 00:57:47 #chair jborean93 00:57:52 nope 00:57:53 :( 00:58:52 OK, any others to discuss? We're about at time. @yuwei - you mentioned a couple PRs? 00:59:19 oh i just mentioned 2 which has no comment 00:59:33 just merged 35037 after testing it 00:59:41 going onto 35038 now 00:59:48 ah ok 00:59:48 Thanks 00:59:54 Anything else for today, then? 01:00:17 No from my side 01:00:53 OK. 01:01:30 @Kylie_ - can you please #endmeeting if you see this (since nobody else got #chair ed, otherwise the meeting will run on for a long time) 01:01:43 Thanks all 01:04:15 thanks! 09:03:38 hey! i am currently trying to create a nic with azure_rm_networkinterface without a public ip. documentation says: "public_ip_address_name: Name of the public ip address. None for disable ip address." but when running my playbook i get "ValidationError: Parameter 'public_ip_address_name' can not be None". anyone there for any help, please? 00:00:52 #endmeeting