[collectd] Collectd clients will loose connection to collectd server if restarted

[Florian Forster]

Hi Giorgio,

On Thu, Apr 16, 2009 at 11:53:15AM +1000, Giorgio Mandolfo wrote:
> What happens is that if for any reason collectd running on the server
> restarts (let's just say, reboot, power failure, hardware upgrade),
> all the collectd daemons running on the 30 clients still continue
> sending data (I can see it with tcpdump), but the collectd server
> seems not catching it. After restarting collectd on every client, the
> server starts to collect data again.

the network plugin uses a unidirectional protocol, i. e. the client send
data but never receive (nor expect) an answer. This means, if the server
reboots, the client's system will get an ICMP message “host unreachable”
or similar, but the daemon itself doesn't notice. This may be an
interesting fact for your network setup though.

The server on the other hand, has absolutely no idea how long a client's
been running. (Unless you're using the Git master branch and the new
uptime plugin. But even then the server doesn't interpret this data but
handles it like any other data it receives.) However, if you restart the
client(s), the source port from which data is sent changes. I wouldn't
be surprised if this is the second interesting fact towards your
problem.

Other than that I have to admit I'm clueless. I've restarted a number of
servers with over 1 MByte/s of data being dropped/dismissed for a short
time, and they all picked it up cleanly when they were started again.
(Except one small and unlikely race-condition which has been fixed in
the meantime.)

Hope this helps :)
-octo
-- 
Florian octo Forster
Hacker in training
GnuPG: 0x91523C3D
http://verplant.org/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://mailman.verplant.org/pipermail/collectd/attachments/20090417/df94c1fe/attachment.pgp