nsd users - Jun 2014 - Problems loading zone with nsd4

Hi!

I just tried to switch from nsd3 to nsd4 but nsd4 fails to load the
zone. The zone is approx 170MB (Bind text format).

NSD4 is configured is slave. The zone is transferred but failed during
loading with memory allocation error:

10:17:16 nsd[3849]: zonefile at.zone does not exist
10:17:16 nsd[3849]: nsd started (NSD 4.0.3), pid 3847
10:17:26 nsd[3847]: xfrd: zone at committed "received update to serial
1402473601 at 2014-06-11T10:17:26 from 83.136.34.4 TSIG verified with
key rcode0-distribution"
10:17:26 nsd[4059]: rehash of zone at. with parameters 1 0 5
b81fd4d081abe7a4
10:17:59 nsd[4059]: mremap(/var/lib/nsd/nsd.db, size 1743910912) error
Cannot allocate memory
10:17:59 nsd[4059]: could not add RR to nsd.db, disk-space?
10:17:59 nsd[4059]: bad ixfr packet part 2007 in diff file for at.
10:17:59 nsd[3849]: handle_reload_cmd: reload closed cmd channel
10:17:59 nsd[3849]: Reload process 4059 failed with status 256,
continuing with old database
10:17:59 nsd[3847]: xfrd: zone at: soa serial 1402473601 update failed,
restarting transfer (notified zone)


The server has 8GB RAM, 512KB swap and 9GB of free disk space, and there
is plenty of disk and ram left when nsd logs the memory error.

Thus, I suspect something else is going wrong. Any hints?

Thanks
Klaus

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Klaus,

On 06/11/2014 10:28 AM, Klaus Darilion wrote:
> Hi!
> 
> I just tried to switch from nsd3 to nsd4 but nsd4 fails to load
> the zone. The zone is approx 170MB (Bind text format).
> 
> NSD4 is configured is slave. The zone is transferred but failed
> during loading with memory allocation error:
> 
> 10:17:16 nsd[3849]: zonefile at.zone does not exist 10:17:16
> nsd[3849]: nsd started (NSD 4.0.3), pid 3847 10:17:26 nsd[3847]:
> xfrd: zone at committed "received update to serial 1402473601 at
> 2014-06-11T10:17:26 from 83.136.34.4 TSIG verified with key
> rcode0-distribution" 10:17:26 nsd[4059]: rehash of zone at. with
> parameters 1 0 5 b81fd4d081abe7a4 10:17:59 nsd[4059]:
> mremap(/var/lib/nsd/nsd.db, size 1743910912) error Cannot allocate
> memory 10:17:59 nsd[4059]: could not add RR to nsd.db, disk-space? 
> 10:17:59 nsd[4059]: bad ixfr packet part 2007 in diff file for at. 
> 10:17:59 nsd[3849]: handle_reload_cmd: reload closed cmd channel 
> 10:17:59 nsd[3849]: Reload process 4059 failed with status 256, 
> continuing with old database 10:17:59 nsd[3847]: xfrd: zone at: soa
> serial 1402473601 update failed, restarting transfer (notified
> zone)
> 
> 
> The server has 8GB RAM, 512KB swap and 9GB of free disk space, and
> there is plenty of disk and ram left when nsd logs the memory
> error.
> 
> Thus, I suspect something else is going wrong. Any hints?
I think it is the memory somehow, perhaps the memory overcommit kernel
settings in Linux are disallowing the allocation, even though there
seems to be enough memory (at current usage).  The most recent code
from the repository (not yet released, but passed regression tests)
has the option to use database: "" in nsd.conf and then the nsd.db is
not made, not mmapped, and thus a lot of disk and memory space is
freed.  That would likely make your system work.

Best regards,
   Wouter

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
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=3WRO
-----END PGP SIGNATURE-----

Quoting Klaus Darilion <klaus.mailinglists at pernau.at>:
> Hi!
>
> I just tried to switch from nsd3 to nsd4 but nsd4 fails to load the
> zone. The zone is approx 170MB (Bind text format).
> 10:17:59 nsd[4059]: mremap(/var/lib/nsd/nsd.db, size 1743910912) error
> Cannot allocate memory
[...]
> The server has 8GB RAM, 512KB swap and 9GB of free disk space, and there
> is plenty of disk and ram left when nsd logs the memory error.
How confident are you on the "there is plenty of disk and ram left"
statement?  In particular, you've got 8GB of memory available,
virtually no swap, and you're trying mmap a file that is 1.7GB.
I don't know how many processes are trying to mmap that file, but
you've not got a lot of head room there.

With respect to the other server working with fewer resources, do you
by chance have overcommit disabled on the broken box and enabled on the
working box?

Also, compare the sizes of /var/lib/nsd/nsd.db on the two machines.
Perhaps the working box is able to mmap because the db file is smaller on
that one.

Is allocating more swap on the broken box an option?  Other than using
up disk space, having lots of swap isn't a downside as long as your
paging rate is low.  That is probably true even if your swap is on
SSD/flash; with a low paging rate, you shouldn't run into flash wear
issues due to using swap.  Maybe add a swap file to test?

Devin