compute nodes with fixed ip or shutdown instead of terminating compute nodes

[cezfer]

I have an HPC cluster that uses licensed software and one of the licensing parameters is the IP of the compute nodes. I read the documentation and I didn't find a way to specify the IP that the compute node should be created on or instead of terminating the node just turn it off. Is there any way to configure this?

[hgreebe]

Hello,
There is no way to specify the ip address of a compute node, but you can use pre/post install scripts to assign secondary ip to the compute node.

Currently, our configuration is to terminate compute nodes if they are down, so they can not just be "turned off".

[cezfer]

Thanks. But I need the primary IP to not change. Can I manually change the launch template to “power off”? Obter o Outlook para iOS<https://aka.ms/o0ukef>

…

________________________________ De: hgreebe ***@***.***> Enviado: Thursday, January 2, 2025 12:16:18 PM Para: aws/aws-parallelcluster ***@***.***> Cc: cezfer ***@***.***>; Author ***@***.***> Assunto: Re: [aws/aws-parallelcluster] compute nodes with fixed ip or shutdown instead of terminating compute nodes (Issue #6620) Hello, There is no way to specify the ip address of a compute node, but you can use pre/post install scripts to assign secondary ip to the compute node<https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/MultipleIP.html>. Currently, our configuration is to terminate compute nodes if they are down, so they can not just be "turned off". — Reply to this email directly, view it on GitHub<#6620 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AQPTOK65QCXOYH7ODPXIQIL2IVJ4FAVCNFSM6AAAAABTOMALQ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNRXHEZTAMJYHE>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[hgreebe]

If you manually change the launch template, and subsequent update may change it back. I believe slurm allows other ways to license: https://slurm.schedmd.com/licenses.html

[cezfer]

the licensing issue is with an IBM solver. Em ter., 7 de jan. de 2025 às 17:32, hgreebe ***@***.***> escreveu:

…

If you manually change the launch template, and subsequent update may change it back. I believe slurm allows other ways to license: https://slurm.schedmd.com/licenses.html — Reply to this email directly, view it on GitHub <#6620 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQPTOK3RRVK4EADUMDKBKX32JQ2VHAVCNFSM6AAAAABTOMALQ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNZWGE3DQMZZHA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- Att, Cezar

[hanwen-cluster]

Even if you stop and start the instance, the IPs will change. The best way is to use secondary IP addresses or keep the compute nodes running all the time.

Would you like this to become a feature request? If yes, could you provide more context (e.g. cluster configuration file, why secondary IP addresses cannot solve the problem)?

[cezfer]

Yes, I want to open an issue. I need to install an agent on the compute nodes that reports the hardware configurations of the nodes to an external license server. This license server uses the server's primary IP as one of its parameters, Each new IP reported means that I have an additional server.

I see two possible solutions:

• An option to specify the server's private primary IP.
• Another option to just shut down the server and not destroy it when it finishes running.

[hanwen-cluster]

Thank you for the information. I've created an item in our backlog and our team will discuss it.

[cezfer]

Thank you Obter o Outlook para iOS<https://aka.ms/o0ukef>

…

________________________________ De: hanwen-cluster ***@***.***> Enviado: Wednesday, January 22, 2025 1:57:34 PM Para: aws/aws-parallelcluster ***@***.***> Cc: cezfer ***@***.***>; Author ***@***.***> Assunto: Re: [aws/aws-parallelcluster] compute nodes with fixed ip or shutdown instead of terminating compute nodes (Issue #6620) Thank you for the information. I've created an item in our backlog and our team will discuss it. — Reply to this email directly, view it on GitHub<#6620 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AQPTOK32PVMFIWESZK7NK2D2L7EX5AVCNFSM6AAAAABTOMALQ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMMBXG43TKNJYHE>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[hanwen-cluster]

For now, here is a possible workaround:
Each compute node has a fixed DNS name. For example, a compute node in "cluster-basic" under "queue-1", "compute-resource-1" has a DNS name queue-1-dy-compute-resource-1-1.cluster-basic.pcluster. The license server could rely on the DNS names.

[cezfer]

unfortunately the DNS name alone does not resolve the issue Obter o Outlook para iOS<https://aka.ms/o0ukef>

…

________________________________ De: hanwen-cluster ***@***.***> Enviado: Wednesday, January 22, 2025 2:08:11 PM Para: aws/aws-parallelcluster ***@***.***> Cc: cezfer ***@***.***>; Author ***@***.***> Assunto: Re: [aws/aws-parallelcluster] compute nodes with fixed ip or shutdown instead of terminating compute nodes (Issue #6620) For now, here is a possible workaround: Each compute node has a fixed DNS name. For example, a compute node in "cluster-basic" under "queue-1", "compute-resource-1" has a DNS name queue-1-dy-compute-resource-1-1.cluster-basic.pcluster. The license server could rely on the DNS names. — Reply to this email directly, view it on GitHub<#6620 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AQPTOK3ILRQLBOE6YQ23J5L2L7F7XAVCNFSM6AAAAABTOMALQ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMMBXG44TSNJYGM>. You are receiving this because you authored the thread.Message ID: ***@***.***>