Multiple Login Nodes Definition in One HPC Cluster

General Discussion
, , ,
1

Hello everyone,

I am trying to configure right now in one HPC ondemand cluster with multiple login nodes.
if i configure the login nodes with shell access, then i have to define under cluster.d folder, but it makes also these login nodes in Active Jobs-> Cluster visible as Cluster,
I want just for one cluster name to view this section and jobs related not for four login nodes

so how could you make it without viewing all login nodes in Active Jobs → Cluster ?

Best Regards

2

We do not list the login nodes in clusters.d at our site. Since our login nodes are identical to each other, we us a DNS round robin name to distribute ssh connections across all of them. We list that name in /etc/ood/config/apps/shell/env as DEFAULT_SSHHOST=

In the past, where each cluster had a login node of its own, I recall using a section in, e.g., the cluster1.yml file

  login:
    host: "login.cluster1.xxx.xxx"

but we haven’t used that in several years…

For the current version of OOD, there’s details here that may be relevant:

Cheers,

Ric

3

We do something similar to Ric’s response, but instead of a round-robin DNS we use HAProxy to distribute SSH sessions to the different login nodes. If one is down, it detects and skips that one.

The DNS name to the HAProxy server is used in our cluster config file.

# /etc/ood/config/clusters.d/elsa.yml

---

v2:

  metadata:

    title: "ELSA Cluster"

  login:

    host: "elsa.hpc.xxx.xxx"

    default: true