Re: [Bacula-users] Tape job with spooling die at the end of the process

[Arno L. <al...@it...>]

Hello,

On 2/13/2007 8:52 PM, Zeratul wrote:
> On Tue, 13 Feb 2007 20:22:35 +0100, Arno Lehmann wrote
> 
>>Hello,
>>
>>On 2/13/2007 6:37 PM, Zeratul wrote:
>>
>>>On Tue, 13 Feb 2007 18:21:27 +0100, Andrea Conti wrote
>>>
>>>
>>>>>Fatal error: Network error with FD during Backup: ERR=Connection 
>>>>>reset by peer
>>>>
>>>>>Before this error, there are normal messages regarding the
> 
> spooling/despooling
> 
>>>>>process.
>>
>>This looks like a network problem. With firewalls and some routers 
>>this seems to be more or less common, without any of these devices 
>>in between your backup server and the client it should not happen. 
>>Unless the OS on one of the computers involved is configured funnily...
>>
>>Anyway, in the case I'm thinking about you will need to add the 
>>'Heartbeat Interval' setting to the relevant daemon configs and 
>>things should work again.
>>
> 
> 
> I'm using "Heartbeat Interval" option since the beggining for all jobs that
> implies a large amount of data (over 5GB).

That's bad... in this case I'd run the jobs with lots of debug output 
enabled on both DIR and FD (but only to make sure you don't miss 
something important) and observe the network connection during the 
despooling phase.

A tcpdump (or ethereal/wireshark) dump which only looks at Bacula 
traffic could be a good start.

Now that I write this, I remember a discussion a few days ago... IIRC, 
someone stated that during despooling the heartbeat is not sent.

This is definitely a problem.

The workaround would be to make sure your DIR and client computers can 
handle "automatic" tcp keepalives and set those to reasonable short 
intervals.

You might want to search the list archives for details.

Arno


-- 
IT-Service Lehmann                    al...@it...
Arno Lehmann                  http://www.its-lehmann.de