slotPerWorker attribute and number of worker per node

[kpouget]

Hello, I am confused by the meaning of the slotPerWorker attribute:

type MPIJobSpec struct {

	// Specifies the number of slots per worker used in hostfile.
	// Defaults to 1.
	// +optional
	SlotsPerWorker *int32 `json:"slotsPerWorker,omitempty"`
        ...
}

Does worker refers to worker pods or worker nodes?

I would like to deploy 1 MPI job per worker node, but it seems to to be what the MPI operator does: (MPI worker pods ...-worker-0 and ...-worker-1 are both on the node worker05)

Is this a bug, or is there a way to deploy one pod per node?

[issue-label-bot[bot]]

Issue-Label Bot is automatically applying the labels:

Label	Probability
kind/question	0.76

Please mark this comment with :thumbsup: or :thumbsdown: to give our bot feedback! Links: app homepage, dashboard and code for this bot.

[carmark]

Hello @kpouget ,

The slotsPerWorker comes from MPI's slots, you may ref it at here.

The Worker means worker pod, you can setup it in yaml file. You may use pod non-affinity feature to schedule them.

[kpouget]

Hello @carmark,

The Worker means worker pod, you can setup it in yaml file.

I'm not sure to agree/observe what you're saying:

I request 4 worker replicas and 2 slots per worker, but I get 4 worker pods

edit: there is a mix of solver/mesher in the screenshot but both run with the same settings

[carmark]

@kpouget The replicas number is same as worker number. It will get 4 worker pods if you request 4 worker replicas.

The slots will not impact the the number of worker pods, it will only be setup in mpi hostfile(/etc/mpi/hostfile).

[tenzen-y]

It seems that this was resolved. If you have any questions, feel free to open new issues. /close

[google-oss-prow[bot]]

@tenzen-y: Closing this issue.

In response to this:

It seems that this was resolved. If you have any questions, feel free to open new issues. /close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.