Re: [squid-users] clarification of delay_initial_bucket_level

On 28/09/2012 7:41 a.m., tcr_at_raynersw.com wrote:
>> (maybe tcr can explain how he came up with 100ms).
> [root_at_sierra ~]# time service squid reload
> real 0m0.014s
> user 0m0.007s
> sys 0m0.005s
>
> My configuration includes an 18,000 line ACL file, so it's pretty big.

Ah. That is just the time it takes one instance of Squid to start,
load/verify the config and send a HUP signal to the existing running Squid.

It *excludes* the time it takes the running Squid to notice the HUP,
pause all transactions, process the config file(s), reset any running
state which may be affected (delay pools, logs, helpers, garbage
management events, etc), then resume the paused transactions.

Is possible that running Squid may do it all in 100ms, but more likely a
few seconds. Jenny Lee mentioned the side effect; connection timeout is
N seconds more likely to happen during the reconfigure, lost new
connections etc. The shutdown truncation is not relevant to a
reconfigure (only to restart), but they can still trunate on other timeouts.

Amos
Received on Sun Sep 30 2012 - 10:29:56 MDT