Re: [Unbound-users] 1.4.1 crashing
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Hi, I have been trying to reproduce this, especially the bus signal that you reported. But have not been able to - I have created some extra portability stuff, but it is unlikely that this caused your troubles (its all in svn). Since bus errors are hardware and OS specific - and most tend to 'silently' handle them for you - I would like to know if (per chance) you both have a very specific hardware/OS combination. We are working towards reproducing but haven't succeeded. Trying different ideas and getting more info is it then. Best regards, Wouter On 01/26/2010 09:32 PM, Artis Caune wrote: 2010/1/26 Attila Nagy b...@fsn.hu: Hello, It still crashes for us. :( I thought about saving the traffic and trying to replay that to another (idle) server, but couldn't yet find the time to do that... same here. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.9 (GNU/Linux) Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/ iEYEARECAAYFAktm+WIACgkQkDLqNwOhpPgGwQCgq3h7INfVNzYeTKmFORz2f0mL C3cAoILI7bQtynlEOTCGAuWVQAdpZHKX =eJce -END PGP SIGNATURE- ___ Unbound-users mailing list Unbound-users@unbound.net http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users
Re: [Unbound-users] 1.4.1 crashing
Hello, It still crashes for us. :( I thought about saving the traffic and trying to replay that to another (idle) server, but couldn't yet find the time to do that... W.C.A. Wijngaards wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Hi, I committed r1959. This has a different structure alignment, which had changed in svn trunk (but is not in 1.4.0 or in the patch itself). This could help out with bus-errors, perhaps with alignment problems for malloc itself, especially on 64bit. I am not sure but it certainly changed and I think could create this sort of result. Could you try that? Perhaps the misaligned read only causes trouble on a page boundary or something like that and this is what trips it up only occasionally? Best regards, Wouter On 01/15/2010 08:46 PM, Artis Caune wrote: 2010/1/15 W.C.A. Wijngaards wou...@nlnetlabs.nl: Artis noted that it seems to be running fine with validator iterator module-config statement (not without the validator). Could this be the same for you? unfortunately one instance just crashed, it was running 10+ hours. Looks like with validator it just takes longer. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.9 (GNU/Linux) Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/ iEYEARECAAYFAktUVrcACgkQkDLqNwOhpPg+dACeN9C2f/bg4Pha9zewOMY9RAue PXcAnj5Qs/me/4c4zIsUfMyQjuJskSZf =Gj8j -END PGP SIGNATURE- ___ Unbound-users mailing list Unbound-users@unbound.net http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users ___ Unbound-users mailing list Unbound-users@unbound.net http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users
Re: [Unbound-users] 1.4.1 crashing
2010/1/26 Attila Nagy b...@fsn.hu: Hello, It still crashes for us. :( I thought about saving the traffic and trying to replay that to another (idle) server, but couldn't yet find the time to do that... same here. -- Artis Caune Everything should be made as simple as possible, but not simpler. ___ Unbound-users mailing list Unbound-users@unbound.net http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users
Re: [Unbound-users] 1.4.1 crashing
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 01/14/2010 04:51 PM, Attila Nagy wrote: Yes, we also have crashes (but with libev and on FreeBSD7 -8 crashes as well). Please try to obtain a crashdump, so we can see whether they are the same. Wouter wrote that I should run unbound in valgrind, but here it's not an easy task... The stacktrace is nice, but the lruhash code did not change between 1.3.4 and 1.4.1. Artis noted that it seems to be running fine with validator iterator module-config statement (not without the validator). Could this be the same for you? Other than that I have no clues except it seems to fail for both of you FreeBSD. But my tests on FreeBSD8 all work well. libevent/libev/builtin does not make a difference. Could check with/without-pthreads. Could try without a cache (0 sizes, the malloc behaviour should be very simple). valgrind is typically very good at spotting such heap-corruption or double free-s (if it works well enough for you). But if those stackdumps are the same that would be interesting... Best regards, Wouter -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.9 (GNU/Linux) Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/ iEYEARECAAYFAktQf2MACgkQkDLqNwOhpPj7sQCgkJpak07KGQRe2EKrd+lnbO4/ ssAAni3eqFJz0fJPH5859x7AgVCAstm1 =EN4K -END PGP SIGNATURE- ___ Unbound-users mailing list Unbound-users@unbound.net http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users
[Unbound-users] 1.4.1 crashing
Hi, we are switching to 1.4.1 version with this patch (FreeBSD8): http://www.freebsd.org/cgi/cvsweb.cgi/ports/dns/unbound/files/patch-fix-ipv6?rev=1.1 and unbound is crashing: pid 1597 (unbound), uid 59: exited on signal 10 pid 18943 (unbound), uid 59: exited on signal 11 I have tried to run with: - 1 and 2 threads - with/without ipv6 patch, no luck. now I'm running it without libevent (libevent-1.4.13). From statistics I can see, that it crashes when certiain amount of memory is eaten, maybe when it's starting cache cleaning ? Anyone have seen this? -- Artis Caune Everything should be made as simple as possible, but not simpler. ___ Unbound-users mailing list Unbound-users@unbound.net http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users
Re: [Unbound-users] 1.4.1 crashing
2010/1/14 Attila Nagy b...@fsn.hu: Hello, Yes, we also have crashes (but with libev and on FreeBSD7 -8 crashes as well). Please try to obtain a crashdump, so we can see whether they are the same. Wouter wrote that I should run unbound in valgrind, but here it's not an easy task... unbound without libevent is crashing too. Looks like 1.3.4 is working fine. -- Artis Caune Everything should be made as simple as possible, but not simpler. ___ Unbound-users mailing list Unbound-users@unbound.net http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users
