On 19 Sep 2006 at 14:12, Gonzalo Arana wrote:
> On 9/19/06, Andres Kroonmaa <[EMAIL PROTECTED]> wrote:
> > > On Tue, Sep 19, 2006, Gonzalo Arana wrote:
> > >
> > > > There is a comment in profiling.h claiming that rdtsc (for x86 arch)
> > > > sta
e can find
> and porting over the remaining stuff from 2.6 into 3. We really
> need to concentrate on fixing up -3 rather than adding shinier things.
> Yet :)
How would you go on with adding call graphs without
adding too much overhead? I think it would be hard to
beat gprof on that one.
Andres Kroonmaa
Elion
worst case looks like probes have been
running for 8.9secs straight, seems unlikely.
So there seems to be a need to get hires profiling
uptodate with current squid code base.
Unfortunately, I can't participate for now, my company
has been restructured and caching has been thrown out, so
I don't have any suitable platform at the moment.. ;(
Andres Kroonmaa
Elion
> Henrik Nordstrom <[EMAIL PROTECTED]> writes:
>
> > This question got me thinking, and maybe we should restrict Squid to plain
> > refuse to start if access rules say "http_access allow all".
Wouldn't this kick in in accelerator configs?
--
On 23 Jan 2004, at 15:07, Henrik Nordstrom <[EMAIL PROTECTED]> wrote:
> On Fri, 23 Jan 2004, Andres Kroonmaa wrote:
>
> > This works with 3.0 only. But for fun, I tried to extract latest probe
> > trace from a core I had. After some scripting, you'd get something li
On 22 Jan 2004, at 14:14, Andres Kroonmaa <[EMAIL PROTECTED]> wrote:
> Also, regarding indirect hints, it crossed my mind that you could try
> check out state of *xprof_Timers array. Profiling probes are reset every
> second by event, thus probes that have stop timestamp z
On 22 Jan 2004, at 11:42, Henrik Nordstrom <[EMAIL PROTECTED]> wrote:
> On Thu, 22 Jan 2004, Andres Kroonmaa wrote:
>
> > If you know the function, then you can try create gdb script that
> > after breaking at some line of the function enables hardware write
>
> Th
us bugs in stackshield) but
> the resulting binary is not even a fully valid binary..
>
> Does any of you know about other tools which can be used to trap where the
> stack gets smashed?
>
> Regards
> Henrik
>
Andres Kroonmaa <[EMA
quot; --- "
else
echo "No running squid found"
fi
sleep 10
done
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 6501 731, Fax: 6501 725
Pärnu mnt. 158, Tallinn
11317 Estonia
On 20 Dec 2003, at 14:09, Robert Collins <[EMAIL PROTECTED]> wrote:
> On Fri, 2003-12-19 at 22:18, Andres Kroonmaa wrote:
> > Now I wonder if it makes sense to wrap it up into some form of
> > ifdef'ed debug option into mempools, or would that clutter source
> >
to wrap it up into some form of
ifdef'ed debug option into mempools, or would that clutter source
needlessly? It isn't perfectly portable I guess, so, I don't know.
--------
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 65
On 17 Dec 2003 at 17:22, Henrik Nordstrom wrote:
> On Wed, 17 Dec 2003, Andres Kroonmaa wrote:
>
> > Yes, but there remains issue with memory fragmentation, to solve which
> > there is sense in coordinating assumptions of both malloc and pools.
> > Chunks are longliv
ult
is limited to 64, meaning no more than 64 chunks would get mmapped.
That seems abit small.
Finally, dlmalloc latest version has thread_safety options.
Rest, internal tunables are probably indeed not needed.
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 6501 731, Fax: 6501 725
Pärnu mnt. 158, Tallinn
11317 Estonia
Hello,
No comments on this one. Would someone please review proposed fix
if it is correct one for CVS?
Seems same bug as:
http://www.squid-cache.org/bugs/show_bug.cgi?id=637
--- Forwarded message follows ---
From: "Andres Kroonmaa" <[EMAIL PROTECTED]&
On 15 Dec 2003, at 23:43, Henrik Nordstrom <[EMAIL PROTECTED]> wrote:
> On Mon, 15 Dec 2003, Andres Kroonmaa wrote:
>
> > So far dlmalloc has been very good malloc. Why is it not safe anymore?
>
> It has never been really safe and this has bitten us to various degree
>
ze(Void_t*);
-voidmalloc_stats();
+voidmalloc_stats(void);
int mALLOPt(int, int);
struct mallinfo mALLINFo(void);
#else
----
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 6501 731, Fax: 6501 725
Pärnu mnt. 158, Tallinn
11317 Estonia
eader, HDR_CONTENT_MD5);
353
354 two = httpHeaderGetStrOrList(&otherRep->header, HDR_CONTENT_MD5);
355
356 if (strcasecmp (one.buf(), two.buf())) {
357 one.clean();
358 two.clean();
359 return 0;
360 }
(gdb) p one
$1 = {size_ = 0, l
On 17 Nov 2003, at 22:09, Henrik Nordstrom <[EMAIL PROTECTED]> wrote:
> On Mon, 17 Nov 2003, Andres Kroonmaa wrote:
>
> > gcc (3.3.1 and 3.3.2) produce error whenever there is variable
> > definition with all uppercase:
> >
> > protos.h
> > -SQUIDC
sFree(FwdServer ** FS);
+SQUIDCEXTERN void fwdServersFree(FwdServer ** fs);
with uppercase compile produces error: parse error before numeric constant
As there are quite many such places I suspect something is wrong on my end.
anyone point me to right rtfm?
----
A
unkillable process means problem blocks inside kernel.
My bet would be buggy kernel threads implementation, especially
that fsck after reboot hints that ext2 was corrupt already when
squid hangs. Try diskd instead of async-io on linux.
Other bet would be faulty hardware or non-ECC memory get
voiding OS support, so, imo, it would work as well
on plain FS as on any other approach. The only reason to go for raw disk
or O_DIRECT would be if we want to keep ram usage to very minimum, ie for
very lowend boxes.
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 6501 731, Fax: 6501 725
Pärnu mnt. 158, Tallinn
11317 Estonia
afaik all network ops should be accounted by profiler, by means of
w/read_handlers in comm_poll at least. But I haven't looked after 3.0
How large is prof_unaccounted during normal high load? If its more
than 1-5% then something substantial is unaccounted for.
Is it possible that more than 2
Prof_unaccounted gets assigned time during which no timers are active.
Most probes that are there have been added ad hoc to cover most busy
parts of squid. Thus we can assume that cpu is hogged by some section
of code thats not normally very cpu intensive.
You could try either add more probes
same url have to go to
origin?
> Sounds odd that a reconfigure makes Squid snap out of the condition. I
> can understand if restart helps, but not a reconfigure..
All connections forcibly closed?
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Mi
On 20 Feb 2003, at 9:41, Henrik Nordstrom <[EMAIL PROTECTED]> wrote:
> On Thu, 20 Feb 2003, Andres Kroonmaa wrote:
>
> > Hey, watch your words, i'm Solaris shop ;)
>
> I am free to express my opinion. Any my firm opinion in this matter is
> that the Solaris 32 b
even though its damn sure outcome
will be exactly same every time. Depending on acls thats not always the
case, but for some most commonly used setups there could be some nice
shortcut.
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 6501 731, Fax: 6501 725
Pärnu mnt. 158, Tallinn
11317 Estonia
ially considering the OS vendors today have
> alternative interfaces which is not crippled like this. For Solaris,
> just compile applications 64bits and be done with it.
>
> Regards
> Henrik
>
> On Wednesday 19 February 2003 18.24, Andres Kroonmaa wrote:
> > hi,
> >
&
calls / % -
> which is very high for headersEnd.
>
> Rob
>
> --
> GPG key available at: <http://users.bigpond.net.au/robertc/keys.txt>.
>
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 6501 731, Fax: 6501 725
Pärnu mnt. 158, Tallinn
11317 Estonia
hi,
we should really avoid fopen() in any place. it may fail when
first 256 FDs are in use. Cause of failures during reconfig.
Andres Kroonmaa <[EMAIL PROTECTED]>
CTO, Microlink Data AS
Tel: 6501 731, Fax: 6501 725
Pärnu mnt. 158, Tallinn
29 matches
Mail list logo