| [02:47] <teward> cjwatson: got an email from the build bot that it failed, looks like the timeout was RabbitMQ related, is this a failure in buildbot? | |
| [03:03] <wgrant> teward: Yeah, known flakiness. I've retried it. | |
| [05:46] <teward> wgrant: ah, thanks. So not my fault :) | |
| [05:46] <teward> looks like it passed. | |
| [05:47] <teward> (and yes I can't sleep >.< Time to try again with melatonin now) | |
| === jamesh_ is now known as jamesh | |
| [09:19] <tomwardill> Fix an OOPS if you search for OCI projects with no term: https://code.launchpad.net/~twom/launchpad/+git/launchpad/+merge/383773 | |
| [09:25] <cjwatson> r=me, one comment | |
| [09:31] <tomwardill> oh, it's not different, I just did a silly | |
| [11:31] <cjwatson> Hmm, looking at the perennial problem where you need to manually clear launchpadlib's wadl cache | |
| [11:31] <cjwatson> I don't think it's a client-side problem | |
| [11:31] <cjwatson> At the server end we seem to be somehow changing the content without changing the Etag | |
| [11:31] <cjwatson> So no wonder the poor client is confused | |
| [11:35] <tomwardill> that would do it | |
| [11:36] <cjwatson> Hm, this only seems to be the case on dogfood though. I wonder if Apache is doing something in front of all the other instances? | |
| [11:36] <cjwatson> Though I'm sure I remember seeing this problem across production upgrades too | |
| [11:37] <cjwatson> Right now the etag on dogfood is literally the sha1 of '\0application/vnd.sun.wadl+xml' though | |
| [11:44] <wgrant> The etag config is customised to ensure consistency across the FEs | |
| [11:44] <wgrant> I forget what it is, maybe mtime and size | |
| [11:50] <cjwatson> I'll see if I have to do anything manual after the next deployment | |
| [11:58] <cjwatson> https://code.launchpad.net/~cjwatson/lazr.restful/fix-wadl-etag/+merge/383777 | |
| [12:04] <wgrant> That MP explains a lot. | |
| [15:08] <tomwardill> Looks like the librarian file upload deduplication either doesn't work or wasn't actually implemented | |
| [15:08] <tomwardill> (or I've lost a bunch of code somewhere) | |
| [15:13] <tomwardill> fixing | |
| [15:38] <tomwardill> oh no, I think I'm just being silly | |
| [15:39] <tomwardill> cjwatson: remind me where the code that pulls files from the builder lives? | |
| [15:46] <tomwardill> aha, buildbehaviour, of course | |
| [15:50] <tomwardill> ah, right,that code does exist, I was just looking in the wrong place | |
| [15:50] <tomwardill> (for future travellers, it's in the buildbehaviour, which ensures that all the files exist from the librarian on disk, ready for the uploadjob) | |
| [15:55] <tomwardill> although the scope it's checking for existing files might be too limited there | |
| [15:55] * tomwardill stares at it a bit harder | |
| [15:56] <cjwatson> lp.buildmaster.interactor too IIRC | |
| [15:56] <cjwatson> Well, in fact that's more about causing the builder to pull files from LP | |
| [15:56] <cjwatson> There's also lp.archiveuploader.ocirecipeupload | |
| [15:57] <tomwardill> yeah, that's where I was originally | |
| [15:58] <tomwardill> by the code in ocirecipebuildbehaviour, we're only looking for existing layer files within the same build (L13( | |
| [15:58] <tomwardill> *L139 | |
| [15:58] <tomwardill> which isn't correct, or useful | |
| [16:02] <tomwardill> cjwatson: so, the buildbehaviour grabs the (existing) file from the librarian and puts it on disk, so the uploadjob can reupload it. This seems a bit weird | |
| [16:03] <cjwatson> I think this was partly in case of racing with GC, but I'm sure it's possible to do better, yes | |
| [16:04] <tomwardill> yeah | |