14:09 * cjwatson peers at the Contents thing
14:09 <cjwatson> We must have got very unlucky with hitting whatever this bug is on the last run before eoan released, but it doesn't seem to be a totally isolated incident. The same thing happened to focal on 2019-10-26
16:32 <infinity> cjwatson: Are we racing somewhere we shouldn't be with Contents? I didn't look into it.
16:32 <infinity> But I guess speeding up the publisher would make that sort of thing more frequent.
16:33 <cjwatson> infinity: I think so, yes. Timestamps don't quite line up perfectly for that but maybe there's some buffering getting in the way
16:34 <cjwatson> The cp -a of ubuntu-overrides at the start of generate-contents-files was while the publisher was generating overrides, and it doesn't do the proper "write to .new, move into place" thing for those
16:34 <infinity> cjwatson: I forget how it all works, but in essence, we run a-f in Contents mode out of band, and suck the results in when publishing, yeah? Could it be that we just need an interim step of "copy the final results from a-f to the interim location the publisher will look at".
16:35 <infinity> Jinx, ish.
16:35 <cjwatson> We have such an interim step, it's just not quite stitched together exactly correctly
16:35 <infinity> Oops?
16:35 <cjwatson> Yes
16:36 <cjwatson> I'm guessing slightly but it looks about right
From #launchpad-ops on 2019-10-29:
14:09 * cjwatson peers at the Contents thing contents- files was while the publisher was generating overrides, and it doesn't do the proper "write to .new, move into place" thing for those
14:09 <cjwatson> We must have got very unlucky with hitting whatever this bug is on the last run before eoan released, but it doesn't seem to be a totally isolated incident. The same thing happened to focal on 2019-10-26
16:32 <infinity> cjwatson: Are we racing somewhere we shouldn't be with Contents? I didn't look into it.
16:32 <infinity> But I guess speeding up the publisher would make that sort of thing more frequent.
16:33 <cjwatson> infinity: I think so, yes. Timestamps don't quite line up perfectly for that but maybe there's some buffering getting in the way
16:34 <cjwatson> The cp -a of ubuntu-overrides at the start of generate-
16:34 <infinity> cjwatson: I forget how it all works, but in essence, we run a-f in Contents mode out of band, and suck the results in when publishing, yeah? Could it be that we just need an interim step of "copy the final results from a-f to the interim location the publisher will look at".
16:35 <infinity> Jinx, ish.
16:35 <cjwatson> We have such an interim step, it's just not quite stitched together exactly correctly
16:35 <infinity> Oops?
16:35 <cjwatson> Yes
16:36 <cjwatson> I'm guessing slightly but it looks about right
However, Tom fixed the thing that we thought was the cause there in https:/ /code.launchpad .net/~twom/ launchpad/ +git/launchpad/ +merge/ 374930; so we're going to need to investigate further.