[Pdns-users] SQLite3 problem during stress

[Christian Svensson]

The crash it self is not a segfault of what I can see.

I test it by running it in "monitor" mode and suddenly it just drops back to
the shell.
It might indeed be solved by your latest patch - I would gladly test it.

Greetings,

Christian "BC" Svensson
Codelead Systems - http://www.codelead.se


On Tue, Jul 21, 2009 at 10:27 PM, Norbert Sendetzky <
norbert at linuxnetworks.de> wrote:

> On Mon July 20 2009 11:01:05 you wrote:
> > Yes, updating only one zone at the time is very much acceptable - that
> the
> > initial transfer takes quite long time does not matter.
> >
> > We do not want to use an "active" database due to memory / CPU footprint.
> > If PowerDNS locks the table and let the other threads just wait for the
> > lock like any other mutex it would probably be just fine with
> > distributor-threads = 1.
>
> The problem of the SQLite database is that while a zone transfer is in
> progress, all other threads can't serve any records. This is because the
> complete table is locked by SQLite when records are updated and this also
> prevents reading records in other zones. Please keep this in mind for your
> installation.
>
> > I want to note that indeed PowerDNS _crashes_ / _exits_ in the end due to
> > the "Database is locked" error. Since I'm currently testing with OpenDBX
> I
> > don't know íf this is something that is regarded as an OpenDBX or
> PowerDNS
> > bug - but gsqlite3 did also crash during heavy load which (as stated
> > before) makes me believe that it's backend agnostic.
>
> What you've shown me in your logs is that the PowerDNS opendbx backend gets
> an
> error from the SQLite library that the database is locked and the opendbx
> backend forces the distributor thread to recreate the connection to the
> database. The problem is that the sqlite3 backend of the OpenDBX library
> returns a fatal error when the database is locked instead of a timeout.
> This
> is fixed in the SVN trunk and the fix will be also be included in OpenDBX
> 1.4.2. The next problem is that the PowerDNS opendbx backend doesn't make
> use
> of timeouts at the moment and treats them as errors. I've already enhanced
> the
> PowerDNS opendbx backend but I have to test it as soon as I can and see if
> the
> problem is gone. I will create a patch and send it to you so you can test
> yourself.
>
> What I haven't seen in your log file excerpt is that the opendbx backend
> crashed with a segfault.
>
>
> Norbert
> --
> OpenPGP public key
> http://www.linuxnetworks.de/norbert.pubkey.asc
>
>
>
> _______________________________________________
> Pdns-users mailing list
> Pdns-users at mailman.powerdns.com
> http://mailman.powerdns.com/mailman/listinfo/pdns-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.powerdns.com/pipermail/pdns-users/attachments/20090728/a5b7b645/attachment-0001.html>