serf w/SurfShield crash in authz_tests.py#15

[Johan Corveleyn <jc...@gmail.com>]

Justin (or anyone),

With latest trunk (@1349944) I get a crash in authz_tests.py 15
("authz issue #3242 - access required at repo root") when running with
serf and my AVG SurfShield enabled. Possibly also other tests crash
(I'll do a full run later tonight). With SurfShield disabled, there is
no crash.

See dav-fails.log (containing the stacktrace, now with more useful
line and variable information than before, thanks to Bert) in
attachment.

-- 
Johan

Re: serf w/SurfShield crash in authz_tests.py#15

[Justin Erenkrantz <ju...@erenkrantz.com>]

On Thu, Jun 14, 2012 at 5:02 PM, Daniel Shahaf <d....@daniel.shahaf.name> wrote:
> I know pf(4) on openbsd has some specific "Be unreliable" options.
> Would they be useful for this?

Since we're using TCP, I expect that such pf filters would just cause
delays that TCP would essentially recover from and might not even be
visible to userspace.  It might affect timings slightly, but I'm not
sure.  The fun bit of Johan's laptop is that it messes up the
userspace timings and breaks up the packets in interesting ways.  --
justin

Re: serf w/SurfShield crash in authz_tests.py#15

[Daniel Shahaf <d....@daniel.shahaf.name>]

Justin Erenkrantz wrote on Thu, Jun 14, 2012 at 15:52:17 +0200:
> It was a good way of capturing some of the more subtle race conditions
> that we've never triggered reliably elsewhere.  -- justin

I know pf(4) on openbsd has some specific "Be unreliable" options.
Would they be useful for this?

For example, from pf.conf(5) on freebsd:

     probability <number>
           A probability attribute can be attached to a rule, with a value set
           between 0 and 1, bounds not included.  In that case, the rule will
           be honoured using the given probability value only.  For example,
           the following rule will drop 20% of incoming ICMP packets:

                 block in proto icmp probability 20%

Re: serf w/SurfShield crash in authz_tests.py#15

[Justin Erenkrantz <ju...@erenkrantz.com>]

On Thu, Jun 14, 2012 at 3:14 PM, Johan Corveleyn <jc...@gmail.com> wrote:
> Great! I can now successfully run the entire testsuite over ra_serf
> with AVG SurfShield enabled. For the first time ever :-).

Yay!

> Thanks for all the bugfixes.

No problem.  Thanks for providing the underpowered laptop with the
nutty anti-virus software!  =)

It was a good way of capturing some of the more subtle race conditions
that we've never triggered reliably elsewhere.  -- justin

Re: serf w/SurfShield crash in authz_tests.py#15

[Johan Corveleyn <jc...@gmail.com>]

On Thu, Jun 14, 2012 at 12:27 PM, Justin Erenkrantz
<ju...@erenkrantz.com> wrote:
> On Wed, Jun 13, 2012 at 11:28 PM, Johan Corveleyn <jc...@gmail.com> wrote:
>> With latest trunk (@1349944) I get a crash in authz_tests.py 15
>> ("authz issue #3242 - access required at repo root") when running with
>
> I have confirmed on your machine that the fix in r1350179 doesn't
> cause a crash any more.
>
> Thanks!  -- justin

Great! I can now successfully run the entire testsuite over ra_serf
with AVG SurfShield enabled. For the first time ever :-).

Thanks for all the bugfixes.

-- 
Johan

Re: serf w/SurfShield crash in authz_tests.py#15

[Justin Erenkrantz <ju...@erenkrantz.com>]

On Wed, Jun 13, 2012 at 11:28 PM, Johan Corveleyn <jc...@gmail.com> wrote:
> With latest trunk (@1349944) I get a crash in authz_tests.py 15
> ("authz issue #3242 - access required at repo root") when running with

I have confirmed on your machine that the fix in r1350179 doesn't
cause a crash any more.

Thanks!  -- justin

Re: serf w/SurfShield crash in authz_tests.py#15

[Justin Erenkrantz <ju...@erenkrantz.com>]

handler->handler_pool is somehow NULL when we invoke get_locations.  Weird.
 Perhaps you can throw some printf's before the get_locations call to see
if it is indeed NULL and not an artifact of the trace.  A quick glance at
the code shows handler_pool should be set in add_file correctly.  -- justin

On Wednesday, June 13, 2012, Johan Corveleyn wrote:

> Justin (or anyone),
>
> With latest trunk (@1349944) I get a crash in authz_tests.py 15
> ("authz issue #3242 - access required at repo root") when running with
> serf and my AVG SurfShield enabled. Possibly also other tests crash
> (I'll do a full run later tonight). With SurfShield disabled, there is
> no crash.
>
> See dav-fails.log (containing the stacktrace, now with more useful
> line and variable information than before, thanks to Bert) in
> attachment.
>
> --
> Johan
>