← Back to team overview

launchpad-dev team mailing list archive

Re: capturing backtraces on every 'expensive' operation

 

On Sat, Nov 19, 2011 at 12:16 PM, William Grant
<william.grant@xxxxxxxxxxxxx> wrote:
>> The file size on disk is 540K, which isn't tiny, but isn't dire either
>> - at that size ~5G a day before garbage collection.
>
> I'm confused -- how is that not dire? We historically kept at least the
> last 30 days, but now that will be hundreds of gigabytes. OOPSes with
> sensible query counts used to be 10-15KB.
>
> Particularly with oops-prune no longer running, we are likely to be in a
> pretty bad situation by the end of next week, even with all the disk
> space we freed on carob yesterday.

We need to get pruning running again, thats totally true.

The new pruner keeps for 1 week, allowing us one week to get a reference.

Some low hanging fruit to get a longer gc period, if we want one:
 - we can compress the oopses - they compress very very well.
   - this just needs a patch to amqp2disk -  oops-tools already
handles compressed oopses.
 - we can move oops-tools to a dedicated server (already planned)

And a longer term goal -
 - we can generate less oopses :)

-Rob


Follow ups

References