Error: Failed to initialize the interactive session

I am trying to validate a cluster profile. I was able to do all tests but the parallel pool test. I have attached my validation report below. Any help is appreciated.

VALIDATION REPORT

Profile: beoshock

Scheduler Type: Generic

Stage: Cluster connection test (parcluster)

Status: Passed

Start Time: Thu May 13 12:21:31 CDT 2021

Finish Time: Thu May 13 12:21:31 CDT 2021

Running Duration: 0 min 0 sec

Description:

Error Report:

Command Line Output:

Debug Log:

Stage: Job test (createJob)

Status: Passed

Start Time: Thu May 13 12:21:31 CDT 2021

Finish Time: Thu May 13 12:21:57 CDT 2021

Running Duration: 0 min 26 sec

Description:

Error Report:

Command Line Output:

Debug Log:

Stage: SPMD job test (createCommunicatingJob)

Status: Passed

Start Time: Thu May 13 12:21:59 CDT 2021

Finish Time: Thu May 13 12:22:37 CDT 2021

Running Duration: 0 min 38 sec

Description: Job ran with 2 workers.

Error Report:

Command Line Output:

Debug Log:

Stage: Pool job test (createCommunicatingJob)

Status: Passed

Start Time: Thu May 13 12:22:39 CDT 2021

Finish Time: Thu May 13 12:23:06 CDT 2021

Running Duration: 0 min 27 sec

Description: Job ran with 2 workers.

Error Report:

Command Line Output:

Debug Log:

Stage: Parallel pool test (parpool)

Status: Failed

Start Time: Thu May 13 12:23:08 CDT 2021

Finish Time: Thu May 13 12:24:41 CDT 2021

Running Duration: 1 min 33 sec

Description: Failed to initialize the interactive session.

Error Report: Failed to initialize the interactive session.

Caused by:

Error using parallel.internal.pool.AbstractInteractiveClient>iThrowIfBadParallelJobStatus (line 433)

The interactive communicating job errored with the following message: MatlabPoolPeerInstance{fLabIndex=1, fNumberOfLabs=2, fUuid=b10ec9e0-6fbc-43e5-8566-67ed5d06514d} was unable to find the host for MacBook-Pro:27370 due to a JVM UnknownHostException: null

[Kojiro Saito]

It seems that you're trying to do parpool with MATLAB Parallel Server using Generic cluster profile.

parpool requires communication between remote servers and client PC, but you got UnknownHostException error which means a worker on remote servers cannot resolve the hostname of client PC.

Is the remote server (MATALB Parallel Server) in the same network of your client PC?

Answer 1

[Raymond Norris]

Let me add to Kojiro's comment. Validation's last stage runs an interactive job with parpool. Look at the following graphic

Notice that the workers running on your cluster need to connect back to your desktop machine. The error you're seeing is that the workers can't "find" the client machine by the shortname. There are two options you can try

• If your desktop is on the network (e.g. VPN), then before you call parpool, set the client hostname in MATLAB, as such:

ip  = java.net.InetAddress.getLocalHost.getHostAddress().string
pctconfig('hostname',ip);

One thing to note: setting the hostname in pctconfig needs to be called before any calls to PCT (e.g. parfor, parpool, etc.)

• Instead of a parallel pool, use a cluster pool with parforOptions. This will only work if you're calling parfor. For example:

c = parcluster;
opts = parforOptions(c);
parfor (idx = 1:100,opts)
   A(idx,1) = rand;
end

[Michael Loibl]

The suggestion with the pctconfig resolved it for me.

Is it possible to set the hostname permanently to the ip?

Or would it make more sense to change something on the cluster resp. the workers?

[Raymond Norris]

@Michael Loibl you have two options

• Add the pctconfig code in your startup.m file
• In R2023a, we changed the startup of the server socket to be on the cluster. It's possible this might work, without change, just out of the box. If you can, run R2023a to see if it works (without needing to call pctconfig).