Re: [squid-users] clarification of delay_initial_bucket_level

From: Amos Jeffries <squid3_at_treenet.co.nz>
Date: Sun, 30 Sep 2012 23:36:50 +1300

On 28/09/2012 7:59 a.m., tcr_at_raynersw.com wrote:
>> Eli, sockets are being closed and reopened on reconfigure as well
> I just tested, and this is false. I opened a 100MB test file download through a proxy, reloaded the proxy multiple times, and the download kept chugging.
>
> If I'd blacklisted my IP as part of that reload, I don't know if it would have kicked me off or let the open socket finish, but that's primarily an academic question.

Depends on whether the access control you blacklisted yourself in was
already tested or not. The time ACL displays the expected behaviour bet
here, ACL only apply when they are tested so existing transactions
continue, new ones are blocked.

> Second test: logged into a server that's currently doing 50Mb/s and probably around 100 requests/sec. Watched traffic in dstat. Did a squid reload. No drop.
>
> Reloading is fine.
>

That statement depends on what type of system and configuration you
have. All the helpers are restarted, which on some OS means UNIX pipes
and others TCP socket closures and reopenings. They overlap and are
scheduled while Squi dis running so it is not overly harmful, but will
face a few connections with extra lag at they wait for the new set of
helpers. YMMV.

Amos
Received on Sun Sep 30 2012 - 10:37:03 MDT

This archive was generated by hypermail 2.2.0 : Sun Sep 30 2012 - 12:00:15 MDT