On 18/04/11 19:35, Eugene M. Zheganin wrote:
> Hi.
>
> Around 6 months ago I switched from 2.7 to 3.1 for its IPv6.
> I may be wrong, but after that I noticed that 'squid -k reconfigure' (I
> use my own custom quota manager, which web-interface issues reconfigure
> request when quotas are changed) now breaks existing connections and
> reopens listening sockets (and it says that in its cache.log). During
> this socket reopening a packet can be received from browser and if there
> is no listening socket on the server, a client then receives RST from
> operating system network's stack and then its browser shows 'The browser
> is configured with proxy which is refusing connections'. And this is
> sad, user start to think that this is a crash and starts ticketing my
> support staff.
>
> Is this a 3.x-only behaviour or was 2.7 behaving identically ?
The behave identically in this regard. I suspect something is causing
3.1 to resume service much slower than 2.7 did. Which particular 3.1
release is doing this?
> Anyway, is there a way to do a 'soft reconfiguration' ? Without closing
> HTTP/ICP/SNMP connections (or at least not breaking client
Sadly not yet. We are working towards it for future releases.
> At this time it looks like '-k reconfigure' is just quite similar to
> fast '-k kill' and restart.
If that is true, then I suspect you are using one of the early 3.1
releases with broken LFS support. Or something is breaking/corrupting
the swap.state journal during a reconfigure.
Does your cache.log contain a warning about "version 1 LFS" detected
or mention a DIRTY load during reconfigure? (may need ALL,1 debug level).
Amos
-- Please be using Current Stable Squid 2.7.STABLE9 or 3.1.12 Beta testers wanted for 3.2.0.6Received on Mon Apr 18 2011 - 07:47:22 MDT
This archive was generated by hypermail 2.2.0 : Wed Apr 20 2011 - 12:00:03 MDT