[sheepdog-users] zookeeper and more switches

[Bastian Scholz]

Sounds like some kind of network problem...

I assume that your three switches are normal
managementable switches?

Maybe you have (an old) Spanning Tree Protocol
enabled and the disconnected Cisco leads to a
recalculation of the paths where the uplink
activity is stopped (could be verified by ping).
Had similar Problems in the past, but only with
corosync, zookeeper seems more rugged against
this kind of failure...

But when the whole Network went bad under load
it could be a bad uplink cable, bad port or a
strange switch failure... Can you reproduce the
bad behavior with other traffic (netcat for
example)?

Cheers

Bastian


Am 2014-08-12 12:33, schrieb Valerio Pachera:
> I have a cluster with 3 nodes.
> 2 nodes are on the same switch (name it switch A).
> The 3th node (named 'vmserver006') has the I/O nic on switch A but
> zookeeper listen on the other nic, on switch C.
> 
> switch A --- switch B --- switch C
> 
> On switch B there's the firewall and a router cisco.
> 
> If I unplug the the network cables of the firewall, vmserver006 gets
> disconnected from the cluster!
> (Note: I also rebooted the cisco router).
> 
> It's the second time it happens.
> 
> What do you think about it?
> 
> Note: time ago I moved the I/0 nic from switch C to switch A because, 
> when
> rebuilding, the whole network was working bad.