Take 2: includes initial modularisation (untested; I'll test it at home
this weekend when I get my tproxy box back online) and configure magic
(with placeholders for tproxy4/freebsd.)
http://www.creative.net.au/diffs/20080408-tproxy-fix-2.diff
TODO:
* pull out the capabilities stuff, removing
mån 2008-04-07 klockan 23:41 -0600 skrev Alex Rousskov:
Can you post cache.log lines with full debugging enabled? I would be
happy to help with the queue review, but it would speed things up if I
do not have to go through the complicated steps required to reproduce
the problem.
The cache.log
Henrik Nordstrom wrote:
mån 2008-04-07 klockan 23:41 -0600 skrev Alex Rousskov:
Can you post cache.log lines with full debugging enabled? I would be
happy to help with the queue review, but it would speed things up if I
do not have to go through the complicated steps required to reproduce
the
tis 2008-04-08 klockan 23:50 +1200 skrev Amos Jeffries:
If its not easy and its not simple it means more bugs. I don't think 'c'
is a good choice here without a full ground-up recode. Nay on that from
me on general principles.
True. I view 'c' as more of a last resort if the other turn out
Hi all.
I'm interested in the performance of squid, and in particular, how do
the developers optimise/benchmark/stress-test the code - I'd like to
use similar ideas for evaluation.
Any hints/tips that you can recomend on a low budget? I've looked at
Web Polygraph, and in a lab environment this
tis 2008-04-08 klockan 13:23 +1200 skrev Amos Jeffries:
Is there a security problem with them being set?
IMO its just a cleanup otherwise (already voted those not to go back).
It's not even a source change, just a bzr attributes cleanup. But the
current situation may confuse some unexperienced
tis 2008-04-08 klockan 17:59 +1200 skrev Amos Jeffries:
Currently:
fde::flags::transparent == 'intercept/non-intercept'
fde::flags::tproxy == real-transparent/non-transparent
(new) COMM_TRANSPARENT == real-transparent
Their use is currently good for what they do. A small re-naming
tis 2008-04-08 klockan 14:28 +0800 skrev Adrian Chadd:
Take 2: includes initial modularisation (untested; I'll test it at home
this weekend when I get my tproxy box back online) and configure magic
(with placeholders for tproxy4/freebsd.)
http://www.creative.net.au/diffs/20080408-tproxy-fix
On Tue, 2008-04-08 at 11:39 +0200, Henrik Nordstrom wrote:
- There is two events that may happen, 'A' and 'B'.
- In the processing of one interest of the other is deregistered with
the silent assumption that after that it won't get invoked, but the
object it's related to still exists.
- Both
.)
http://www.creative.net.au/diffs/20080408-tproxy-fix-2.diff
Comments:
commTransparentRemote needs to move into comm_openex. You can't fit
tproxy4/freebsd in a reasnoable manner otherwise.
Yup, that'd be part of the follow up work. This first cut is to break out what
is there; the next
Polygraph takes a while to setup right. It depends on how hard you wish
to hit the proxy.
I also use httperf/apachebench in specific configurations to stress certain
parts
of the Squid codebase.
Adrian
On Tue, Apr 08, 2008, Bradley Kite wrote:
Hi all.
I'm interested in the performance of
On Tue, 2008-04-08 at 13:57 +0100, Bradley Kite wrote:
I'm interested in the performance of squid, and in particular, how do
the developers optimise/benchmark/stress-test the code - I'd like to
use similar ideas for evaluation.
Any hints/tips that you can recomend on a low budget? I've
On Tue, 2008-04-08 at 15:15 +0200, Henrik Nordstrom wrote:
tis 2008-04-08 klockan 17:59 +1200 skrev Amos Jeffries:
Currently:
fde::flags::transparent == 'intercept/non-intercept'
fde::flags::tproxy == real-transparent/non-transparent
(new) COMM_TRANSPARENT == real-transparent
tis 2008-04-08 klockan 13:23 +1200 skrev Amos Jeffries:
+1. I'm just waiting on you all to agree that its tested enough. If you
want to do the merge yourself Henrik, I'm okay with that.
Doesn't matter for me who merges the backport. It's a trivial bzr merge
with the backport already done..
tis 2008-04-08 klockan 22:17 +0800 skrev Adrian Chadd:
* make sure upstream/peer-forwarded requests aren't thrown to the tproxy
code.
I read the current code as don't do that; its the behaviour I'd like to
maintain. We can always add the functionality later.
Yes, it is not something
On Tue, 2008-04-08 at 17:04 +0200, Henrik Nordstrom wrote:
tis 2008-04-08 klockan 13:23 +1200 skrev Amos Jeffries:
+1. I'm just waiting on you all to agree that its tested enough. If you
want to do the merge yourself Henrik, I'm okay with that.
Doesn't matter for me who merges the
tis 2008-04-08 klockan 13:57 +0100 skrev Bradley Kite:
Any hints/tips that you can recomend on a low budget? I've looked at
Web Polygraph, and in a lab environment this seems excellent, but I
think it requires more hardware than I have available to me.
There is a number of other tools simpler
On 08/04/2008, Henrik Nordstrom [EMAIL PROTECTED] wrote:
tis 2008-04-08 klockan 13:57 +0100 skrev Bradley Kite:
Any hints/tips that you can recomend on a low budget? I've looked at
Web Polygraph, and in a lab environment this seems excellent, but I
think it requires more hardware than I
On Tue, Apr 08, 2008, Adrian Chadd wrote:
Take 3:
http://www.creative.net.au/diffs/20080408-tproxy-fix-2.diff
* restored the global flag which indicates tproxyness; renamed to
enable_spoof_srv
* moved the tproxy bind stuff into comm.c, with a flag to comm_openex()
* changed forward.c to try
tis 2008-04-08 klockan 08:57 -0600 skrev Alex Rousskov:
IIRC, one-bucket-at-a-time was too slow in environments that cared more
about hits than optimizing miss performance. Folks wanted a knob to
control the trade-off.
Ok. We should change that to be buckets based, and default to 1.
The
tis 2008-04-08 klockan 07:47 -0600 skrev Alex Rousskov:
I believe there is a better option that was added to handle specifically
to handle these kind of problems:
d) Cancel the call when it is no longer expected.
That's also an option, not at all far from 'a'.
This is already supported by
Alex Rousskov wrote:
On Tue, 2008-04-08 at 17:04 +0200, Henrik Nordstrom wrote:
tis 2008-04-08 klockan 13:23 +1200 skrev Amos Jeffries:
+1. I'm just waiting on you all to agree that its tested enough. If you
want to do the merge yourself Henrik, I'm okay with that.
Doesn't matter for me who
ons 2008-04-09 klockan 00:19 +0800 skrev Adrian Chadd:
On Tue, Apr 08, 2008, Adrian Chadd wrote:
Take 3:
http://www.creative.net.au/diffs/20080408-tproxy-fix-2.diff
using -3 instead of -2 looks better... but incomplete.
You need to add the new ipt_* files to the EXTRA_squid_SOURCES
Henrik Nordstrom wrote:
So any proposals on how we would go about fixing comm_read_cancel?
Hmm.. on reading that code commio_cancel_callback looks a bit odd in
trunk. Why is callback set to NULL twice, and what is really the purpose
of this function now that the actual callback is in an
On Tue, 2008-04-08 at 20:46 +0300, Tsantilas Christos wrote:
Henrik Nordstrom wrote:
So any proposals on how we would go about fixing comm_read_cancel?
Hmm.. on reading that code commio_cancel_callback looks a bit odd in
trunk. Why is callback set to NULL twice, and what is really the
tis 2008-04-08 klockan 20:46 +0300 skrev Tsantilas Christos:
AsyncCalls are refcounted so believed that with ccb-callback = NULL
the AsyncCall which the ccb-callback point released. But yes if it is
refcounted by an other object (for example in AsyncCalls queues,)
it is not released and
in commSetTimeout there is the following fragment which looks very odd
to me
if (callback != NULL) {
typedef CommTimeoutCbParams Params;
Params params = GetCommParamsParams(callback);
params.fd = fd;
F-timeoutHandler = callback;
}
tis 2008-04-08 klockan 12:55 -0600 skrev Alex Rousskov:
BTW, if somebody commits that patch, please polish the reason phrase in
the cancel() call to something more specific.
The submitted pach didn't work for two reasons. In fact this
cancellation of comm scheduled calls doesn't work at all.
On Tue, 2008-04-08 at 22:23 +0200, Henrik Nordstrom wrote:
What worries me a bit is that this introduces a queue where invocation
was always immediate before. So we have a different problem space to
deal with in queued events where we before had to deal with recursive
events.. It's a subtle
On Tue, 2008-04-08 at 22:38 +0200, Henrik Nordstrom wrote:
in commSetTimeout there is the following fragment which looks very odd
to me
if (callback != NULL) {
typedef CommTimeoutCbParams Params;
Params params = GetCommParamsParams(callback);
tis 2008-04-08 klockan 17:55 -0600 skrev Alex Rousskov:
is worth the bugs and complexity we are _removing_. Also, the future SMP
support may operate within the same nothing-is-immediate problem space.
In other words, let's get used to it! :-)
Note: For SMP to work out you want as few hops
tis 2008-04-08 klockan 18:04 -0600 skrev Alex Rousskov:
Yes, it is setting the fd member of the call[back].
Yes, I realized that after sending the message.
The syntax is very ugly, but I do not have a better proposal at the
moment. I'll guess I get used to it eventually...
Regards
Henrik
On Tue, 2008-04-08 at 22:23 +0200, Henrik Nordstrom wrote:
I still think the cbdata model is generally preferrable. Much less to
keep track of. For pconn it would need one cbdata object per idle fd,
and having only that object invalidated when the fd is taken out of
the pconn queue for
On Wed, 2008-04-09 at 02:05 +0200, Henrik Nordstrom wrote:
tis 2008-04-08 klockan 17:55 -0600 skrev Alex Rousskov:
is worth the bugs and complexity we are _removing_. Also, the future SMP
support may operate within the same nothing-is-immediate problem space.
In other words, let's get used
On Wed, 2008-04-09 at 01:51 +0200, Henrik Nordstrom wrote:
tis 2008-04-08 klockan 12:55 -0600 skrev Alex Rousskov:
BTW, if somebody commits that patch, please polish the reason phrase in
the cancel() call to something more specific.
The submitted pach didn't work for two reasons. In fact
On Tue, 2008-04-08 at 15:15 +0200, Henrik Nordstrom wrote:
tis 2008-04-08 klockan 17:59 +1200 skrev Amos Jeffries:
Currently:
fde::flags::transparent == 'intercept/non-intercept'
fde::flags::tproxy == real-transparent/non-transparent
(new) COMM_TRANSPARENT == real-transparent
tis 2008-04-08 klockan 13:23 +1200 skrev Amos Jeffries:
Is there a security problem with them being set?
IMO its just a cleanup otherwise (already voted those not to go back).
It's not even a source change, just a bzr attributes cleanup. But the
current situation may confuse some
I've started getting this today:
cvs -z9 -d :pserver:[EMAIL PROTECTED]:/squid co squid
open /dev/null failed
Operation not supported
Has something broken on the CVS server?
--
Benno Rice
[EMAIL PROTECTED]
Tsantilas Christos wrote:
Alex Rousskov wrote:
On Tue, 2008-04-08 at 17:04 +0200, Henrik Nordstrom wrote:
tis 2008-04-08 klockan 13:23 +1200 skrev Amos Jeffries:
+1. I'm just waiting on you all to agree that its tested enough. If you
want to do the merge yourself Henrik, I'm okay with that.
On Wed, 2008-04-09 at 13:07 +0800, Adrian Chadd wrote:
On Tue, Apr 08, 2008, Alex Rousskov wrote:
Hi,
FYI: I will commit this tomorrow if there are still no objections.
I'd like to see the pconn stuff with AsyncCalls sorted out first before
more potentially disruptive stuff is
On Tue, Apr 08, 2008, Alex Rousskov wrote:
I'd like to see the pconn stuff with AsyncCalls sorted out first before
more potentially disruptive stuff is included.
Why wait? The two issues are unrelated, being worked on by different
folks, and the suggested changes are relatively benign
41 matches
Mail list logo