Hi List,
apologies if this has been asked before (or is obvious) - I did do some
reading & searching but can't quite figure the best way to achieve this.
Background - we have two productions clusters, both running SLURM. They
are not currently a multi-cluster setup; they are not running the same
version of SLURM. The older of the two clusters is to be retired within
the next couple of months. The setup for it's replacement will be
different enough that we decided to simply start from scratch (i.e.
build a new cluster); there'll also then be a multi-cluster setup.
However, we'd like to keep the accounting records from the old cluster
(i.e. the 'historical' cluster usage data). We don't want to import it
as 'live' data into the new cluster's database; accounting for the new
cluster should start from scratch. But we'd like to be able to
retrieve/query historical usage information still.
Is it feasible to import data from a 'dead' cluster into a slurmdbd?
What would the implication be of doing so, and how to do it - would we,
basically, have to add it as another cluster (i.e. pretend I have three
clusters instead of two)? Would this work even if there is no slurmctld
etc for it? (And how would I make it not visible to users - they really
shouldn't try to submit jobs to it, or query it or anything; from a user
point of view the old cluster should be gone from the system?)
The other option I can think of is have a separate slurmdbd server for
the historical data - nearly the same questions, would I also need a
slurmctld or something, or does it only need a slurm.conf and a database
server to be able to run 'sacct' against it? (So what's the minimum setup?)
Hope I'm making sense!
Tina
