| === freyes_ is now known as freyes | ||
| === tris- is now known as tris | ||
| === beisner- is now known as beisner | ||
| === zz_CyberJacob is now known as CyberJacob | ||
| stub | aisrael: users are just roles in PostgreSQL. When the PG charm generates the new user/role, we grant it the old user/role so it gets all the old permissions. | 06:55 |
|---|---|---|
| stub | aisrael: So we don't need to introspect the database and change all the ownership, which is good as it is a) non trivial and b) quite invasive. | 06:55 |
| ParsectiX | Hi jujuers. IS there a way to run a sandbox of juju on my laptop ? | 07:50 |
| === Guest19391 is now known as frobware | ||
| === CyberJacob is now known as zz_CyberJacob | ||
| sparkiegeek | gnuoy: dosaboy: wolsen: can I get some review ♥ for https://code.launchpad.net/~adam-collard/charms/trusty/swift-storage/guard-paused-unit-service-restarts/+merge/269860 please? | 09:10 |
| gnuoy | sparkiegeek, isn't pause_aware_restart_on_change going to be applicable to lots of your maintenance actions? | 09:12 |
| sparkiegeek | gnuoy: yes, I assume your next question is "why not charm-helpers?" | 09:12 |
| gnuoy | sparkiegeek, it is, but maybe that's the layout change | 09:13 |
| sparkiegeek | gnuoy: to which the response is - thought about it, but the is_paused() semantics could be different for each charm | 09:13 |
| gnuoy | sparkiegeek, merged | 09:17 |
| sparkiegeek | gnuoy: thank you! | 09:17 |
| gnuoy | np | 09:17 |
| stub | Is it still possible for juju to be running 2 hooks simultaneously in the case where you have two units smooshed onto one machine? Or are hooks serialized per machine now? | 10:40 |
| stub | I think they are serialized now, as my primary and my subordinate hooks no longer clash. | 10:40 |
| stub | And if so, anyone know what version this changed in? | 10:41 |
| stub | Thinking that apt failing to grab a lock may indicate a real failure now, rather than something intermittent requiring retries | 10:42 |
| === perrito667 is now known as perrito666 | ||
| elmo | stub: there's a broken assumption there .. brb | 12:37 |
| sparkiegeek | broken assumption? | 12:37 |
| stub | But the assumption is too large for this margin | 13:16 |
| elmo | haha | 13:17 |
| elmo | stub: the broken assumption is that only juju is invoking apt on a system | 13:18 |
| elmo | stub: the classic counterexample is landscape | 13:18 |
| sparkiegeek | elmo: stub: right! I'm wondering if aptdaemon has a role to play here | 13:19 |
| sparkiegeek | of course it's not a complete answer because there will always be something else invoking APT and taking locks | 13:20 |
| stub | I think we need a synchronous apt-get update that keeps retrying until it works or fails hard (screwed apt, timeout). And an apt-get install that just tries and fails (eg. trying to install an unsigned package) | 13:22 |
| stub | I know nothing about aptdaemon or if it can help. | 13:22 |
| stub | I've looked at the code and can adjust retries or add my own timeout easily enough, so I can work around for a particularly problematic charm like Cassandra | 13:23 |
| sparkiegeek | stub: well the idea of aptdaemon is that it's one process that talks to APT and clients just make requests of it to do things | 13:23 |
| sparkiegeek | stub: I'm curious, what is it about Cassandra charm that makes the current logic problematic? | 13:24 |
| stub | Sounds good if you can wait for it to report success, rather than fire and forget and hope the job is done before you need it | 13:24 |
| stub | The Cassandra charm can install DataStax Enterprise, which exists in a password protected apt archive. Remembering to add the URL in the first place, getting it right - it doesn't sound like much, but so far everybody has screwed up some part of the configuration. | 13:25 |
| sparkiegeek | so the issue is that it takes 5mins to report the screw up? | 13:28 |
| stub | With archive.ubuntu.com, a ppa, the us datastax.com - apt-get update takes a minute or two to run and fail. 30 times that is not 5 minutes. | 13:29 |
| stub | Now I realize it is retrying a fixed number of times, I can lower that number. I'd assumed it was retrying forever, because that is what it felt like :) | 13:31 |
| sparkiegeek | :) | 13:31 |
| stub | aptdaemon looks nice. I'm also drawn to a few nice features like partial updates (update only the sources in a given sources.list). | 13:35 |
| stub | I have a feeling some of this will need work for wily. aptdaemon may simplify things. | 13:37 |
| sparkiegeek | i'm not sure whether we ship policykit in cloud images, do you know? | 13:37 |
| stub | No idea. Does it matter for root? | 13:39 |
| sparkiegeek | well it seems that aptdaemon depends on PolicyKit, *shrug* | 13:42 |
| sparkiegeek | I mean depending on aptdaemon just gives you a bootstrap issue - how do we get aptdaemon in the first place :) | 13:43 |
| === verterok` is now known as verterok | ||
| stub | sparkiegeek: The bootstrap packages aren't as hard, as we know there is a working primary archive or we would not have gotten as far as an install hook. | 14:03 |
| stub | sparkiegeek: but yeah, it would need to retry there. | 14:03 |
| sparkiegeek | stub: right, that's what I meant - two units racing to install aptdaemon | 14:03 |
| stub | sparkiegeek: Until juju sticks charmhelpers on the units for us and the charm dependency bootstrap problem goes away. | 14:04 |
| sparkiegeek | stub: I Want To Believe | 14:04 |
| stub | Or we make aptdaemon a snap. | 14:05 |
| === wendar_ is now known as wendar | ||
| === l6unchpad is now known as l6unchpad|away | ||
| === lukasa is now known as lukasa_away | ||
| === lukasa_away is now known as lukasa | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!