On 21/04/2013 9:01 p.m., babajaga wrote:
> Amos,
>
> a bit off-topic, but anyway: How "transparent" is the "squid -k reconfigure"
> for the user ?
Depends on the changes that have been made to the config file and speed
of loading it.
Squid *pauses* the TCP accept and I/O handling while parsing the config
file. Then when configuration is all processed successfully it un-pauses
and keeps going.
Normally this whole process should be very fast and users only notice a
few hundred extra milliseconds of delay.
> Is squid simply refusing new connections, and waiting, until busy conns are
> finished, then doing the reconfig ?
No, squid is simply queueing all I/O events, new TCP connections and
packets pile up in the TCP buffers.
> When doing single-user test of my simple method, I did not notice any glitch
> during brwoser session, when the "reconfigure" was done. However, with many
> users, this might be different.
Yes you should not notice it unless there is a lot of traffic throughput
at the time or there are a lot of big changes to be done (ie loading MB
of a ACL blacklists every hour could be problematic).
Still the pause is a periodic impact on *all* traffic, however small.
Whereas using an external_acl_type helper to check things provides
real-time ACL changing and no queuing at all for any of the accept() or
I/O operations.
Amos
Received on Sun Apr 21 2013 - 13:30:54 MDT
This archive was generated by hypermail 2.2.0 : Sun Apr 21 2013 - 12:00:04 MDT