perl client crash after timeout while connecting

[Brad Fitzpatrick]

Crazy.

Patch applied.  I'll try to contact Perl people who can look into this bug
in Perl.

- Brad


On Thu, 28 Jun 2007, Matthieu PATOU wrote:

> Dear list
> the current version (1.23 from SVN) of the memcached perl client crash when it once fail to connect an host and the
> success to connect to another host.
>
> On my workstation (Linux ubuntu 6.10), when the problem occurs i get :
>
> mat at hades:~/work/memcached-ref/api/perl$ perl mctest.pl
> using parser: Cache::Memcached::GetParser
> *** glibc detected *** perl: munmap_chunk(): invalid pointer: 0x081fd690 ***
> ======= Backtrace: =========
> /lib/tls/i686/cmov/libc.so.6(__libc_free+0x18a)[0xb7e02b4a]
> perl(Perl_sv_setsv_flags+0x5cb)[0x80d370f]
> perl(Perl_pp_sassign+0x56)[0x80c37e4]
> perl(Perl_runops_standard+0x13)[0x80bb273]
> perl(perl_run+0x2e1)[0x80639b4]
> perl(main+0x112)[0x8060012]
> /lib/tls/i686/cmov/libc.so.6(__libc_start_main+0xdc)[0xb7db18cc]
> perl[0x805fe81]
> ======= Memory map: ========
>
>
> The GetParserXS module is NOT installed
> After some sessions of debugging it appears that it is Perl which fools itself when executing :
>    $sock = $buck2sock[$bucket] ||= $self->sock_to_host($self->{buckets}[ $bucket ]
>
> When sock_to_host had returned undef (because it had been unable to connect to the desired server), the next execution
> of this line will crash when trying to set the return value of sock_to_host into $buck2sock[$bucket]
>
> The test case is in the mctest.pl (I simulate the timeout on connect by NOT having a memcache server listening on 11212).
>
> I fix the bug by simplifying the code and splitting into smaller pieces (see the patch attached to this mail).
>
> Cheers
> Matthieu
>
>
>