launchpad-dev team mailing list archive

[Tim Penhey <tim.penhey@xxxxxxxxxxxxx>]

On Wed, 14 Jul 2010 14:08:28 Stuart Bishop wrote:
> On Wed, Jul 14, 2010 at 12:19 PM, Michael Hudson
> 
> <michael.hudson@xxxxxxxxxxxxx> wrote:
> > UPDATE CodeImportMachine SET heartbeat=CURRENT_TIMESTAMP AT TIME ZONE
> > 'UTC' WHERE CodeImportMachine.id = %s took 8 seconds in this case.  I
> > think this must be a result of some kind of locking, but I can't really
> > imagine what as this is the only method I can think of that writes to
> > this table routine. Maybe many concurrent calls to the same method are
> > contending with each other?  Seems a tad unlikely though.
> 
> If job A on machine M does that update, other jobs on that machine
> cannot make the same update until job A commits or a aborts the
> transaction. Are there multiple processes on a machine trying to
> update the heart, and if so, are they running in autocommit mode or
> committing directly after updating the timestamp?

Yes there are multiple jobs, up to 10 on our fast box.  I don't remember 
enough to know exactly what is causing the slowdown.  Nothing changed on the 
code import worker over the last several months at least.

Tim