Re: [HTCondor-users] How use less ports

[Dan Bradley <dan@xxxxxxxxxxxx>]

On 6/11/13 3:11 AM, Romain wrote:
> Dan Bradley <dan@...> writes:
>
> > Inbound access must be allowed to the shared port.  Outbound access must
> > be allowed to all ephemeral ports or all ports in a range that you
> > define by configuring OUT_LOWPORT and OUT_HIGHPORT.  Some guidelines on
> > how many ports are required may be found here:
> http://research.cs.wisc.edu/htcondor/manual/v7.8/3_7Networking_includes.html#
> SECTION00471500000000000000
> > --Dan
> Ok, so I can't use 9618 and 9614 as single ports open (in and out) ? with:
> 9618 for collector and for shared_port
> 9614 for negotitator
>
> I've to open inbound access to 9618, 9614 and a shared_port (9616 for
> example) and open outbound access for all my range port (for example 9600 to
> 9800) for all nodes, is it that ?

That sounds correct.  I can't think of any good reason to have the 
negotiator use its own fixed port.  It should be easiest to just have it 
use the shared port, which should be the default behavior if you enable 
shared_port.  The collector can also be configured to use shared_port if 
you define COLLECTOR_HOST appropriately throughout the pool.

> The range port that will be defined by the number of max job can be run ?

Yes.  For outgoing connections from the submit machine, you need a port 
range (OUT_LOWPORT to OUT_HIGHPORT) that scales with the maximum number 
of running jobs.  Other nodes can have a smaller port range.  If the 
large outbound port range is a problem, you can configure the execute 
nodes to use CCB.  This causes all connections between the submit and 
execute nodes to be in the direction execute --> submit, which reduces 
the number of required outbound ports on the submit machine.

I'd recommend not complicated things unless you really have to.

> For resume what I can do:
>
> All nodes can join the pool, when I do condor_status all nodes appear
> When I submit jobs from the only execute and submit network (network 1), I
> can see these on the Master with condor_q -global
> When I submit jobs from the execute/submit and master network (network 2) I
> can't see jobs with condor_q -global from the master, i've this :
>
> -- Failed to fetch ads from: <X.X.X.228:9618?sock=15911_c3e8_3> :
> THE_SUBMIT_MACHINE (X.X.X.228)
> CEDAR:6001:Failed to connect to <X.X.X.228:9618?sock=15911_c3e8_3>
>
> Jobs submit from network 1, run on network 1 machines, jobs from network 2
> apprently never start to run...

It is not clear to me what configuration you are using when you observe 
the above behavior.

--Dan