pkgsrc-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

bind9 thread problems?

I have a mid sized server ( E5 with 12 cores )
running netbsd 6.1.4 amd64 with pkgsrc from the pkgsrc-2014Q2 branch
on which named hangs soon after startup. This happens with the bundled named
and with bind98 and bind99 packages. I pulled bind910 from pkgsrc HEAD and
it hangs as well. 

This server does have 19 master zones and 5 forward only zones. 
Some of the zones have GENERATE statements for full /24 subnets, 
so the number of A records is in the 1000s. By university standards
I know this is still small, but it's not a trivial install either.

I enabled the xml statistics server and up until it stopped
responding, nothing looked out of place. Using ps to look at the
LWP didn't reveal anything odd either, except that when it worked,
one process was in state kqueue, and when it didn't there were none in 
state kqueue. 

Bind 9.x seems fine on my smaller Netbsd 4 systems using the same 
config and zone files. ( modulo a few tweaks to the ip addresses for listening )

I compiled 9.10 from source with threads disabled and it doesn't seem to hang. 
It's also noticably faster. 

> Statistics:
>   Parse input file:     once
>   Ended due to:         reaching end of file
>   Queries sent:         541729 queries
>   Queries completed:    541729 queries
>   Queries lost:         0 queries
>   Percentage completed: 100.00%
>   Percentage lost:        0.00%
>   Started at:           Wed Jul 30 15:14:47 2014
>   Finished at:          Wed Jul 30 15:14:59 2014
>   Ran for:              11.729994 seconds
>   Queries per second:   46183.229079 qps

I'm not sure if this is a netbsd 6 thread issue or a bind issue at this point,
but if I had to guess, the former seems more likely. 

It did seems like 9.10 hung sooner than 9.9, which in turn hung sooner than 
9.8 can run for several minutes, or even a few days before it hangs. 9.9 only 
works for 
a few minutes ( or about 3000 queries ), and 9.10 hangs almost immediately. ( 
in less than a minute )
So whatever the problem is, it seems to be getting worse with newer versions of 
Regardless of where the problem is, I think it might make sense to modify
the bind 9.x packages to allow threading to be disabled cleanly. 
Right now threading is a default option. 

> ===> Installing dependencies for bind-9.10.0pl2
> ==========================================================================
> The supported build options for bind are:
>         bind-dig-sigchase bind-xml-statistics-server
>         dlz-filesystem inet6 ldap mysql pgsql readline
>         rrl threads
> The currently selected options are:
>         inet6 readline threads

Is there a non-hackish way to disable threads other than redefining 
PKG_DEFAULT_OPTIONS in /etc/mk.conf? 

Is there any other data I should gather before filing bug(s)?
Any suggestions would be appreciated.

Thanks in advance, 

Harry Waddell 

Home | Main Index | Thread Index | Old Index