On Sat, 2008-03-01 at 16:27 +0900, Adrian Chadd wrote:
> On Mon, Feb 25, 2008, Alex Rousskov wrote:
> > The overall goal is robustness: make production Squid installations
> > crash less often. In this particular case, we are trying to kill the
> > transaction or job that hit an internal error (a.k.a., assertion) rather
> > than killing the whole Squid. If our attempts to isolate the failure
> > fail, Squid aborts as before.
> >
> > The assert_burst_max configuration knob controls whether Squid tries to
> > be robust and how hard it tries to do that. The logic is experimental
> > and is likely to change once we get more experience with this feature.
>
> Although I applaud the effort, I'm not quite sure this is at all a good
> idea, even if it seems so.
Adrian,
Thank you for the comments. During the London meeting a couple of
days ago, I believe participating folks (including you) reached a
compromise regarding this feature specifications. I will post the
details soon for further review and discussion.
Alex.
Received on Tue Mar 04 2008 - 10:43:19 MST
This archive was generated by hypermail pre-2.1.9 : Tue Apr 01 2008 - 13:00:10 MDT