Re: [HTCondor-users] Multiple CCB on a single server

[Todd L Miller <tlmiller@xxxxxxxxxxx>]

> We have some execute nodes which don't have a public ip address in the
> private network. We would like them to join the condor pool in the public
> network.  CCB seems to be a good solution for us to expose the execute
> nodes in the private network to the collector in the public network. We
> would like to make the system more robust by running multiple CCBs. Right
> now we have already had multiple submit nodes (each has a running schedd)
> in the public network. So we decided to colocate each schedd with a CCB.

	OK, I understand your architecture now.

> So, in the configuration of the execute nodes, we specify the list of CCBs
> that are colocated with the schedds in the public network. We hope that the
> shadow daemon on submit node could heartbeat with startd on the execute
> node in the private network through the local CCB. But it seems that the
> schedd could connect to a remote CCB on other submit nodes to communicate
> with the startd on the execute node in the private network.

	The way we implemented fail-over is by choosing randomly between 
the CCBs for each connection, so yes, this is a possibility.  On the other 
hand, requesting a connection via CCB is not very expensive, so it's 
not just reasonable to make all CCB requests through a third machine, but
actually the conventional set-up.

> I am not sure if my understanding of CCB and the interaction between the
> daemons is reasonable or not. Feel free to correct me or make any
> suggestions on the setup.

	The conventional set-up has been to use the central manager(s) as 
CCBs, because those collectors are already so important to the functioning 
of the pool, and outbound traffic from the execute nodes has to be able to 
make it there already.  However, I think it should be fine to run a CCB on 
each schedd.

- ToddM