When a client becomes ineligible for any reason (like a crash),Nomad moves allocations of running jobs on that client to another healthy client. (based on various constrains, spread etc.)
but when the client comes back to life and rejoins the cluster or alternatively, a new client is connected to the cluster, Nomad doesn’t reallocate those moved jobs for better and more even spread (not until it faces resource problems ).
i was wandering how can i ensure that when a failed client recovers (or a new client connects to cluster) nomad will return all the previously moved allocations back to it ?
Hi @maxim-design,
When a client {re-}joins the cluster, a number of evaluations a generated to asses required allocation placements including blocked workload and jobs of type system. Nomad unfortunately does not track previously migrated workload or perform periodic rebalancing which would result in the behaviour you desire.
The Nomad backlog does have #10039 which details a rebalancing feature, however, I am not aware of this being roadmapped presently. Adding a +1 will always be useful for internal prioritisation along with any comments which have not been previously mentioned.
Thanks,
jrasell and the Nomad team
1 Like
Thanks for the information. will definitely add my +1 to the mentioned thread, and will be awaiting eagerly.
1 Like