[08:14] Hi All, [08:16] Hey [08:22] Hi, [08:22] is there anyone there ? [08:24] Helo [08:29] Helloooo [08:37] people here, some just very async [08:49] Is there anybody there? === frankban|afk is now known as frankban [08:59] A lot of people are here. :) [09:08] when issuing a `juju run`, what user is the command run under? I thought that it was root but I don't have permissions to create some directories, so now I'm curious :) [09:09] icey: ubuntu - same user as if you run 'juju ssh'. If you want root, you can use sudo. [09:10] blahdeblah: yeah, that's what I'm doing now :); thanks! [15:48] arosales: happy valentines day: bug 1749302 [15:48] Bug #1749302: manual add-cloud should be more descriptive about the endpoint requirements [15:48] better late than never ;) [16:07] rogpeppe: re: my PR, I know; That's what I did at first, there's a deeper problem there with Mongo not joining replicaset (sometimes..) === frankban is now known as frankban|afk [16:17] wpk: sorry, i don't understand [16:17] wpk: how are the API addresses related to mongo replica sets? [16:18] wpk: it's just that the code that i saw just looked fragile to me. [16:19] rogpeppe: At first I did it your way and it failed (with mongos on HA nodes not joining replicaset), I got to think that those addresses are used for mongo too. Then I did it the way in PR and it worked. Then I tested it again and it didn't [16:19] there's a weird race somewhere, currently working on it [16:20] wpk: hmm, looks like there's a deeper issue there that you're papering over... :) [16:20] wpk: i don't understand why you can't just use all the addresses anyway [16:21] wpk: ah, because you don't want to accidentally connect to a remote node i guess [16:21] rogpeppe: Yes [16:21] rogpeppe: since that caused problems before - that's why we're forcing localhost to be first on the list, but it doesn't help in all cases as apicaller can be up before apiserver [16:22] wpk: you could filter the addresses rather than just choosing the first one (filter all addresses that aren't localhost addresses) [16:22] wpk: that would be more robust (and i think it would make sense in the code too) [16:22] rogpeppe: For controller I'll just give one address in APIInfo.Addrs [16:23] (just like you suggested, as this was my first thought too) [16:23] But for now I need to figure out the deeper issue [16:23] wpk: that's fine too. [16:23] wpk: yeah. replica sets can be unreliable [16:24] wpk: although i don't think we derive API addresses from mongo replicaset addresses do we? [16:24] rogpeppe: we don't [16:25] wpk: ah, i guess that if the API server can't connect to its local mongod, it won't be up and therefore nothing will be able to connect to it [16:25] rogpeppe: the final symptom is that we can't connect to local mongo (as it's not in replicaset), so we can't start state, and since we're only connecting to ourselves we can't start apicaller and the rest of stuff [16:25] rogpeppe: and I guess there's something to prevent it if we can connect to other controllers somehow [16:25] wpk: in that case, i think just filtering for localhost addresses should work fine (and it's a fairly local change) [16:29] wpk: network.SelectInternalHostPort(addrs, true) might do the job [16:29] wpk: or probably SelectInternalHostPorts, although there shouldn't be a difference [16:29] * rogpeppe isn't very keen on the juju/network package [16:45] rogpeppe: the problem is somewhere else [16:45] Doesn't work: juju bootstrap aws --no-gui && juju switch controller && juju model-config logging-config="=DEBUG" && juju enable-ha [16:45] Works: juju bootstrap aws --no-gui && juju switch controller && juju model-config logging-config="=DEBUG;juju.mongo=TRACE" && juju enable-ha [17:29] kwmonroe: you know just how to sweet talk me. [17:29] heh [19:44] hi, im facing neutron-gateway/0: hook failed: "config-changed" ... im using openstack-telemetry bundle with MAAS. Any advise how should i resolve the charm [19:58] ejat: how is it failing? juju debug-log -i neutron-gateway/0 --replay | pastebinit