See http://build.squid-cache.org/job/3.HEAD-amd64-CentOS-5.3/2228/changes
Changes:
[Amos Jeffries] Fix ConnOpener IPv6 awareness
When updating IPv6 support for split-stack one USE_IPV6 wrapper was
omitted conversion to the EnabledIpv6 stack auto-detect mechanism.
This resulted in IPv6 addresses
On Tue, Jan 15, 2013 at 6:07 PM, Rainer Weikusat
rweiku...@mobileactivedefense.com wrote:
I have no real idea what is necessary to comply with this requirement
but I'll try something sensible: I'm working for a US/UK based company
which provides a comprehensive enterprise security solution for
See http://build.squid-cache.org/job/3.HEAD-amd64-CentOS-5.3/2229/changes
Yesterday one of my busy 3.2.6 servers died after a rapid series
of 'isEmpty()' asserts in store.cc. The servers use one AUFS
cache directory and do not use workers.
Here is part of the cache.log leading up to the assert:
2013/01/15 16:53:47 kid1| WARNING: 1 swapin MD5 mismatches
2013/01/15
With the help of some more testing, I managed to capture this
again. Below is cache.log output at log lever 5 for an attempt to
connect to a HTTP server which didn't exist (http://76.31.53.108 in
this case):
The 'shutdown' sequence starts with the client closing its connection
to the server:
Hello,
clientProcessRequest() assumes that a CONNECT request is always
tunneled and sets flags.readMore to false. However, if url_rewriter
redirects the CONNECTing user, Squid responds with a redirect message
and does not tunnel. In that case, we do want to read more. Otherwise,
On 01/15/2013 07:39 AM, Rainer Weikusat wrote:
internal method named 'stillConnecting' is introduced which
contains the check for an 'abort done by the parent' which is presently
in ConnOpener::connect,
// our parent Jobs signal abort by cancelling their callbacks.
if (callback_ ==
On 01/16/2013 09:36 AM, Rainer Weikusat wrote:
Argument against it: doneConnecting would call
ipcacheMarkBadAddr(host_, conn_-remote); for the destination address
if the timeout handler was executed in this case which looks as if it
might be desirable.
One more reason to remove
On 17/01/2013 7:31 a.m., Alex Rousskov wrote:
On 01/15/2013 07:39 AM, Rainer Weikusat wrote:
internal method named 'stillConnecting' is introduced which
contains the check for an 'abort done by the parent' which is presently
in ConnOpener::connect,
// our parent Jobs signal abort by
On 01/16/2013 09:58 PM, Amos Jeffries wrote:
Whet I was trying to point out is that we have three race conditions
between timeout() and connect() Calls...
* connect() happens first and runs re-schedules while timeout() is queued.
- this is the case we are now ignoring. I'm fine with that.
On 17/01/2013 6:53 p.m., Alex Rousskov wrote:
On 01/16/2013 09:58 PM, Amos Jeffries wrote:
Whet I was trying to point out is that we have three race conditions
between timeout() and connect() Calls...
* connect() happens first and runs re-schedules while timeout() is queued.
- this is the
11 matches
Mail list logo
