>
> You said "nodes", not "jobs" :)
>
>
Oops sorry, it was late :)



> You can use ClusterNode.metrics() to get additional information like number
> of currently executed jobs. Will this work for you?
>
> I also recommend to take a look at existing implementations, like
> AdaptiveLoadBalancingSpi. In majority of use cases there is no need to
> implement your own SPIs.
>
>

I gave a look to it but unfortunately does not fit my use case. I'm using
Ignite to span posix processes that execute external tools / scripts.

Such jobs declare the amount of resources they need i.e. the number of
cpus, memory, etc. This if, for example a node has 16 cpus and it is
already executing a job using 8 cpus. The load balancer should be able to
route to that node only 1 job using 8 cpus or 4 using 2, etc.


The ClusterNode metrics instead reports the number of running jobs which in
my case is not useful.


Cheers,
Paolo

Reply via email to