733070 - reset try repo

Status: RESOLVED WONTFIX

(Release Engineering :: General, defect)

Description

[[e-mail] (:bhearsum)]

While testing something in staging, Erick found that we are now getting HTTP 41*3 errors (Request *Entity* Too Large) from slaves that don't already have a cached try repository. It seems that Mercurial moved the list of heads from the URI to somewhere else in that request, and that we're now hitting *that* limit too. We should reset the try repo at some point to "fix" this.

Comment 1

[[e-mail] (:bhearsum)]

09:52 < catlee> I remember there were some settings being tweaked on the proxy to work around that
09:52 < bhearsum|buildduty> oh, really?
09:52 < catlee> yeah
09:53 < catlee> iirc the error is coming from zeus or varnish, not hg.m.o itself
09:53 < catlee> maybe poke corey and see what he remembers

Comment 2

[Corey Shields [:cshields]]

Best thing to do here is reset try..  we've bumped up limits as far as we can all through the chain.

I'd like to finish the discussion of resetting try in a monthly maintenance window too..

Comment 3

[John Hopkins (:jhopkins)]

Was there a reason we couldn't simply merge the old heads together (say, daily or weekly) to keep the head count on the try repository to a reasonable number?

In local tests, this simple script I wrote accomplishes that:

 http://pastebin.mozilla.org/1507044

Comment 4

[John Hopkins (:jhopkins)]

I'm re-merging the heads (beyond the newest 200) on my local try repo with the commit message 'try: -b do -n -p none -u none -t none' (DONTBUILD doesn't work on try).  This will result in no builds being triggered by pushing these merge changes.

I can push my local repo to try tomorrow morning to reduce the number of heads - if there are no objections by then, I'll give a heads-up in #build before doing so.

Comment 5

[John Hopkins (:jhopkins)]

Pushed my merged-heads repo to try: added 1973 changesets with 23440 changes to 3824 files (-1973 heads)

Re-cloned the try repo: 936 heads currently

We just need to verify that this fixed the problem.

<bhearsum|buildduty> the symptom that led to the reset bug being filed was the fact that a full clone from a linux build machine kept getting HTTP 413/414 errors

Comment 6

[John Hopkins (:jhopkins)]

Attachment: script to merge mercurial repo heads

Comment 7

[[e-mail] (:bhearsum)]

I just tried cloning the full try repo on a linux machine with the same version of hg that erick was having trouble with. It worked, which means I don't think we need to reset the repo!