[Samba] Internal DNS stops forwarding

[Michael Wood]

Hi

On 8 October 2012 20:02,  <felix at epepm.cupet.cu> wrote:
>
>>> Happened again with rc2 but found that at the same time this error shows
>>> every second:
>>>
>>> [2012/10/05 09:01:39,  0]
>>> ../source4/smbd/process_single.c:56(single_accept_connection)
>>>    single_accept_connection: accept: NT_STATUS_TOO_MANY_OPENED_FILES
>>>
>>> After restarting everything is OK, but it happened yesterday though I
>>> didn't notice it was at the same time the dns error and this too many
>>> opened files.
>>>
>> Next time when it happens can you do this:
>> netstat -anp | grep 53 to get the pid of the samba process that is
>> listening on port 53 then do
>> gdb -p <pid>
>> bt full
>> thread apply all bt full
>> info locals
>> generate-core-file /tmp/core_for_dns
>>
>> And send to the list the info, keep the corefile in a safe place and
>> send it upon request to one of the samba developer.
>>
>> Matthieu.
>
> Here we go:
>
> root at ad:~# netstat -anp | grep 53|grep samba|wc -l
> 1003

I think this is the direct cause of the "too many open files" error.

The default limit for number of open files is 1024.

Of course I don't know what's causing samba to start so many instances
of itself in the first place.

Matthieu Patou <mat at samba.org> wrote:
> Somehow I missed this information, can you do a lsof and filter
> just the process that is handling the DNS requests (the one that
> you get from netstat -anp | grep 53).

e.g. try this:

# lsof -n -P -i :53

> It could be related to this bug:
> https://bugzilla.samba.org/show_bug.cgi?id=8878
>
> In a nutshell I suspect that our server sends forward requests to
> the forwarder that are never answered and the connections piles
> up, once we reached the limit (1024 ?) the server didn't accept
> any new connections.

Seems likely.

-- 
Michael Wood <esiotrot at gmail.com>