498468 - Reduce l10n polling time

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[Chris AtLee [:catlee]]

Attachment: Reduce l10n polling interval to 15 minutes

Polling each of the per-locale hg repos can take a while, and add unnecessary load to production-master.

This patch backs reduces the polling interval to 15 minutes (from 5 minutes currently)

Comment 1

[Axel Hecht [:Pike]]

IMHO, 15 minutes is overdoing it, and not really "unnecessary". If the load on the master is too high, it might be a better option to spread out the load. IMHO, we should also check how much we can optimize the actual poller in bug 498055.

When trying to fix bugs in a localization, cutting down from 12 possible builds per hour to 4 is pretty tough.

Loadwise, it might run more l10n builds at the same time in turn, too.

I should revisit my thinking on getting the cross-repo pushlog system upstream, I guess.

Comment 2

[[e-mail] (:bhearsum)]

IMHO this is a short term fix. Is it really big burden for a localizer to wait an extra 10 minutes?

We should fix this better, and in a way that is good for everyone, but IMHO this is fine for a short term solution.

Comment 3

[Axel Hecht [:Pike]]

What is it a short term fix for, if I may ask?

And the extra 10 minutes may not matter for a single landing, but in the test sprints that I was in, errors in the localization masqueraded each other, so you fix a bug, you wait for the build, see the menu, fix another bug, etc. So you end up with various landings during that sprint. Which I mentioned that you get at max 4 builds per hour. For the low-technical l10n folks and their peer testers, that's not a whole lot.

Comment 4

[[e-mail] (:bhearsum)]

To help the load situation on production-master.

Comment 5

[Armen [:armenzg]]

Axel, I mentioned last night what you mentioned about having a cross-repo pushlog. I have filed bug 498641 to discuss the long term solution.

Comment 6

[Chris AtLee [:catlee]]

Comment on attachment 383346 [details] [diff] [review]
Reduce l10n polling interval to 15 minutes

changeset:   1203:44575ba71411

Comment 7

[Chris Cooper [:coop] (he/him)]

(In reply to comment #4)
> To help the load situation on production-master.

Yes, this is totally a short-term solution to help mitigate load on production-master. If we get a real downtime soon where we can restart the master to apply the logging patch (bug 493752), we might be able to revisit this. A unified pushlog (comment #5) is a better solution though.