search

hpc / hpc-collab

This project provides provisioned HPC cluster models using underlying virtualization mechanisms.
Other
10 stars 7 forks source link

new cluster recipe: multiple slurmdbd instances, for linked multicluster w/QOS overlap #221

Closed ssenator closed 3 years ago

ssenator commented 3 years ago

slurm multicluster implementations are constructed with a single slurmdbd logical instance (may be composed of a primary and alternate) with a single data base logical instance (again, possibly in an db cluster, but a single logical unit).

The slurm_acct_db schema includes a global QOS table. This prevents differing enforcement policies on differing cluster instances if both pick a common QOS name, such as "normal" or "standard". With multiple slurmdbd instances, and other clusters registered as "external" using the slurm 20.02 AccountingStorageExternalHost feature, there may be QOS name overlap with differing values for each cluster. This enables multicluster job submission, with each cluster's schema management independently.

ssenator commented 3 years ago

changes merged into primary recipe, vx conjoined cluster