launchpad-dev team mailing list archive

[Danilo Šegan <danilo@xxxxxxxxxxxxx>]

Hi Rob,

У уто, 02. 11 2010. у 12:32 +1300, Robert Collins пише:
> We've put slightly higher timeouts in place for the following pages:
> 
> === Top 10 Time Out Counts by Page ID ===
> 
>     Hard / Soft  Page ID
>      266 /    0  Archive:EntryResource:getBuildSummariesForSourceIds
>      162 /  260  BugTask:+index
>      133 /  161  CodeImportSchedulerApplication:CodeImportSchedulerAPI
>      121 /   43  Person:+commentedbugs
>      107 /  282  Distribution:+bugs
>      100 /    0  MailingListApplication:MailingListAPIView
>       29 /   53  Archive:+packages
>       26 / 1956  Archive:+index
>       21 /   25  Milestone:+index
>       15 /   85  POFile:+translate
> 
> LP devs can see these on https://edge.launchpad.net/+feature-rules -
> sorry but its staff only.
> 
> While the reported hard timeout count should massively reduce now,
> this doesn't reduce the urgency with which we should be addressing
> timeouts. Remember, per the ZeroOOPsPolicy that they should be
> prioritised to the top of each teams kanban backlog :). I know many
> teams are successfully fixing timeouts, and I think thats fantastic -
> I don't mean to nag by saying this, rather I'm hoping to avoid any
> premature happiness when the *report* stops showing the timeouts : the
> underlying issue is still there, and the bandaids are strictly
> temporary.

While we do see as many as 15 POFile:+translate timeouts on a very bad
day, they are usually slightly lower.

For instance, top timeouts per day doesn't include this page the day
before (Oct 30th) nor in the previous report (Oct 28th), and I have to
go back to 26th to find it with a big number of occurrences (50), that
was an issue that happened post 8.4 upgrade (and that I hope I have a
fix for: after 26th, we've worked around it until we came up with a
proper fix).

> I'll be cross referencing the soft timeouts (which haven't been
> altered at all) to see when we can remove the overrides and also to
> ensure that we consistently have timeout bugs filed for poorly
> performing pages.

Filing a bug for POFile:+translate page won't help much.  It issues a
huge number of queries and we can only fix it by changing how it's doing
things.  That's on our agenda, but it will take a while.  For now, we
are able to keep it mostly in check with DB trickery.

Unless you've got strong reasons to want to keep POFile:+translate under
different rules, I'd like us to get rid of the exceptional timeout
threshold for it.

Cheers,
Danilo