| meena | I've been invited to the canonical org again! | 13:28 |
|---|---|---|
| Odd_Bloke | meena: powersj: Yep, it was so we could request you as a reviewer; we're adding you back ATM. | 13:28 |
| Odd_Bloke | (Our IS department are introducing automation to sync LDAP and GitHub to ease starter/leaver tasks; you just need adding to an allowlist. :) | 13:29 |
| Odd_Bloke | https://github.com/canonical/cloud-init/pull/493 is no longer WIP, and is ready for review/landing. | 16:32 |
| Odd_Bloke | smoser: rharper: Given ds-identify should mean that we only use the OpenStack DS when we should expect a response, https://github.com/canonical/cloud-init/pull/501 seems like a reasonable change; any thoughts? | 17:42 |
| smoser | so current behavior is it will try once, with a socket timeout of 10 seconds ? | 18:02 |
| smoser | before ditching it? | 18:02 |
| smoser | non-intel is an issue | 18:04 |
| rharper | Odd_Bloke: I wonder if we should order the 90_dpkg.cfg datsource_list by increasing timeout? for non-intel, they are trying all ds ; NoCloud, ConfigDrive, OpenNebula, DO, Azure, AltCld, OVF, MAAS, GCE before openstack ; for intel they'll head right for it (and not trying anything else); | 18:24 |
| Odd_Bloke | Oh, because non-Intel don't (necessarily) have DMI data? | 18:34 |
| rharper | correct | 18:37 |
| rharper | so it's a maybe and you don't get the short list | 18:38 |
| rharper | beyond optimization; for non-ds id paths; the change seems fine; though going from 10 to 120 is a big jump for timeouts; but generally I don't think there's anything wrong with the change; | 18:39 |
| smoser | its hard... because the truth is, if your metadata service isn't there, its because you have a failed cloud. | 20:45 |
| smoser | in the past 10 years, i have never seen ec2 metadata service down. | 20:46 |
| smoser | but, imo because of bad design, every openstack has issues with terribly slow metadata service. | 20:46 |
| smoser | so one par tof me says "sure, change cloud-init to work better on your little toy" | 20:47 |
| smoser | and the other part says "fix your silly cloud." | 20:47 |
| smoser | but we do not want a larrge timeout on non-intel unless there is only one datasource in the list. | 20:48 |
| rharper | smoser: we had a similar discussion around ipv6 imds for openstack ... without a non-networky way to know whether it's worth waiting or not; you just end up waiting; | 21:06 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!