[00:35] babbageclunk: i updated the tests based on the merge build failures… do you want to take another look before i squash the commits? :-D [00:37] hml: sure [00:38] babbageclunk: you’re awesome! [00:41] hml: Still looks good to me! [00:41] babbageclunk: ty [02:20] wallyworld: I'm looking at your PR now - can you look at mine? :) https://github.com/juju/juju/pull/8380 [02:21] sure, after tim meeting [02:21] wallyworld: oh, looks like hml beat me to it - do I still need to look at it too? [02:25] babbageclunk: looks like some fixes were asked for so i'll do those and then ping [02:25] lunch first [02:25] okies [02:25] bon appetit! [02:26] * babbageclunk feels bad for momentarily forgetting how to type a é [02:26] *an [02:26] é [02:26] ^ like that [02:27] You just cut'n'pasted my one though, be honest! [02:27] babbageclunk: compose, ', e [02:27] * babbageclunk is chastened [02:27] á é í ó ú [02:27] :-P [02:28] ☃ [02:28] Apparently, ẃ ŕ ý ṕ ś ǵ ḱ ĺ ź ć ń ḿ all work, too [02:29] probably useful in wpk-land. [02:29] possibly - I never realised non-vowels could have accents [18:00] Bug #1749763 opened: remove an application not in bundle should be ignored [22:49] Could someone look at https://github.com/juju/juju/pull/8375 (and at #1 comment in bugreport, as it gives the background) and state if that solution makes sense to them? (it works) [22:51] wpk: i think i agree with roger's comment [22:52] hml: did we need to make a fix to the update series tool to fix the symlink? [22:52] wallyworld: refresh [22:52] wallyworld: right now it’d be part of the update series how to guide in the docs [22:52] wallyworld: I forgot to remove it. [22:53] wallyworld: trying to understand if there way a change… and if that’s the only gotcha [22:53] wallyworld: don’t remember having to do that at all before [22:53] wallyworld: the big change is in MongoInfo actually [22:54] hml: ok. i do think we need a fix to the tool ultimately. good that we have a workaround for now. i am surprised we didn't hit the issue before given that agent symlink is tied to series [22:54] wallyworld: that’s what’s bugging me [22:56] wpk: the "isController" flag - that means is the connection for *this* controller or *any* controller? I assume *this* controller? [22:57] wallyworld: isController flag means that we're a controller, and then - as a controller we should connect to self only. [22:58] wpk: makes sense. could you expand the comment a little to make it very clear and that we are deliberately only returning api info with local host for that reason [22:59] wallyworld: and the MongoInfo stuff? That's a 'hack' (as it assumes that mongos are on the same IPs as controllers). [23:00] still digesting that bit, reading bug [23:04] wpk: so with the patch we try local mongo first but will fail over to connect to a mongo on a different controller if needed. but i'm not sure that we shouldn't just be waiting until out own mongo becomes available. ie controller and mongo tied together. if mongo not available on a controller yet, wait for it. maybe need a second opinion [23:05] wallyworld: "waiting until out own mongo becomes available" <- that could never happen (see my comment on the bug) [23:07] right, we have a problem. but maybe the solution lies elsewhere. i'm not sure we're fixing the correct root cause. i am not sure it makes sense for a controller to talk to anything other than it's own mongo. but i may be wrong. i'd like to see what john thinks [23:07] it's not super critical to land this right away as 2.3.3 has gone out [23:08] wallyworld: we've had a discussion today and that was one of the proposals [23:08] oh, i see. john was happy with the approach taken? [23:08] anyone.. https://github.com/juju/juju/pull/8388 [23:09] wallyworld: (and the controller talks to mongo master, not only own mongo - it just has to know where the other mongos are and we're giving only localhost, then mongo itself tells about all other peers) [23:09] wpk: i guess it will be temporary and will correct itself and then the controller will stil use local host [23:09] wpk: this may be what jam and I discussed the other day [23:09] wallyworld: the other approach would be to limit connections to localhost only after the controller is really a controller (hasVote) [23:10] right, i would have maybe preferred that but if john is happy [23:10] I'm gathering opinions :) [23:10] i am just wary about hw doign this current approach may affect HA [23:10] Democracy and stuff [23:11] i am not an expert here, just have an opinion [23:11] wpk: I don't agree with wallyworld and roger [23:11] just saying [23:11] wpk: if we are going to filter differently based on controller or not controller [23:11] I'd rather do that in the connection space, rather than the writing config space [23:11] we need a filter either way [23:11] and I'd rather it was closer to the connect [23:12] thumper: my point was that i disagree with the implicit assumption that localhost was always first [23:12] thumper: My first approach was the way Roger wanted (in config space), but it didn't worked so I moved it to connect code, it worked, but then I realized I have a race and it's just random [23:12] unless that is made very clear that's part of the contract [23:12] thumper: So here I agree with Roger and Ian [23:12] I was talking about api connect [23:13] are you talking about mongo connect? [23:15] API [23:15] mongo is a separate thing, that came out of comment/1 on this bug. [23:16] I like what we have, where we explicitly say []string{localAPIAddr} [23:16] and just that [23:16] don't assume localhost first [23:18] wpk: how did it work before? [23:20] wpk: if we only ever returned localhost before, how did HA work? [23:22] thumper: we returned localhost and all other addresses [23:22] thumper: (localhost first) [23:22] not in the MongoInfo function on the left [23:23] thumper: but connecting to other addresses during upgrade caused problems (that we've never reproduced, but we had consistent reports of them) [23:23] thumper: ah, for Mongo it worked because apicaller could connect to other controllers [23:23] I'm not sure we are talking about the same thing [23:24] ah... [23:24] thumper: and then machiner started, reported to other controller that this machine is alive, the other controller told mongo to connect 'this' mongo to replicaset [23:24] If we put localhost at the start of the Mongo info, does it happen to choose that first? [23:24] thumper: that's what in the code, but then it'll connect to master eventually [23:25] that happens at the driver level right? [23:25] thumper: even if we set MongoInfo to 'localhost' only it'll connect to master as it'll get it from local mongo, so it's not 'info', it's just 'seed' [23:25] * thumper nods [23:27] wpk: I'm adding a comment [23:28] wpk: actually, no comment, this makes sense to me now [23:40] OK, I'll fix the tests tomorrow and ask for acceptance. [23:47] thumper: babbageclunk: team meet?..