Re: store.log bloat possible fix?

On Wed, Oct 04, 2000, Henrik Nordstrom wrote:
> That is because it was I who threw it there.
>
> This section is a bit magic, and should perhaps be documented better.
>
> the storeReleaseRequest takes care of unregistering the swap file
> number, and the storeRelease actually releases the storeentry (or
> something like that. Was after all about a year ago so my memory is a
> bit dim on the details now...)
>
> If you don't call storeRelease here, then any cancelled objects will
> stay in memory. The full story should be available on my squid-2.2 patch
> page. (if I remember correcly there was two problems in this area.. one
> causing a huge temporary memory usage during rebuild, and one causing a
> storeentry leak during rebuild..)

oh yes, this is all coming back to me now. Uhm, evil. Evil. :-)

> The whole procedure of storeLog while reqbuilds needs to be carefully
> reviewed. The current design is flawed in more than one way.
>
> * Amount of entries logged
> * How to deal with crashes/restarts during the rebuild
>
> My proposal is to not log actions from the rebuild, but keep logging
> normal actions to the current swap.state file. When the rebuild is
> completed (original end of swap.state found) then write out a clean log
> file.

I'll attack this and make it sane somehow.

Adrian

-- 
Adrian Chadd			"If a butterfly flaps its wings in China,
<adrian@creative.net.au>	    will a woman get naked in Amsterdam?"
				      -- Ashley Penney on Chaos Theory