Skip to content

Related issues with pin nodes #133

Closed
Closed
Related issues with pin nodes#133

Description

@RobertKrawitz
RobertKrawitz
opened on Jul 27, 2021

I'm opening this as a single issue; all of this is tightly bound up together.

1. "Pin node" is not a standard Kubernetes/OpenShift term. "Node selector" would be a better term, as that's what it is.
2. It should not be necessary to assign the benchmark operator pod to a node selector; let the scheduler take care of that.
3. It shouldn't really be necessary to assign the workload to a node selector. For cases where it matters, it should be optional.
4. For workloads that should run on different nodes, pod anti-affinity would be a better construct and could be applied by the operator itself (I suspect it already does).

Activity

  1. ebattat commented on Jul 28, 2021

    @ebattat
    ebattat
    on Jul 28, 2021
    Member
    1. You right, but this is the name convention that use in Benchmark-Operator tool in all the workloads, we need to discuss it with them.
    2. This is already done. see in the readme PIN_NODE_BENCHMARK_OPERATOR environment variable
    3. We decided that node selector will turn on by default but we can control it through configuration file but true/false parameter, i.e.
      uperf template
    4. Lets discuss together and see what is the best way to use node selector feature. one option is to pass environment variable that called selector for enable/disable this feautre.

  2. RobertKrawitz commented on Jul 28, 2021

    @RobertKrawitz
    RobertKrawitz
    on Jul 28, 2021
    MemberAuthor
    1. If it's inherited from upstream, I withdraw this.
    2. I don't think this (or any of the other PIN_NODE) options should be mandatory at all.
    3. I don't understand?
    4. Sure. In general, I think that all of the PIN_NODE variables should be optional. If there are situations where they shouldn't be, I'd like to analyze them.
    5. I think that these should all be command line arguments, not environment variables, though. These are all options for a specific instance of an application (benchmark-runner), not common settings that should apply across everything, such as search path, display, etc. KUBEADMIN_PASSWORD and KUBECONFIG may qualify as the latter, but IMO the others don't.
  3. ebattat
    pinned this issue on Jul 28, 2021

  4. ebattat commented on Aug 1, 2021

    @ebattat
    ebattat
    on Aug 1, 2021
    Member

    I have moved all the nodes to be optional.

  5. ebattat commented on Aug 4, 2021

    @ebattat
    ebattat
    on Aug 4, 2021
    Member

    Please let me know if there is more issues related nodes.

  6. ebattat
    closed this as completedon Aug 4, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions