[BUG] Latest task status may not always be reported

[hamersaw]

Describe the bug

Every time FlytePropeller checks a task status it constructs a pluginRequestedTransition which contains information regarding the state change of the plugin. This information is used to determine whether the current state has altered from the previously reported state which is determined using the Phase and PhaseVersion information that each plugin reports. If this has not changed (based on these two field) then FltyePropeller immediately returns and bypasses sending a TaskExecutionEvent reporting the task state to FlyteAdmin.

The problem is that plugins may report new status with the same Phase and PhaseVersion but with an updated Reason, the Reason field contains information regarding task execution. One such scenario is where a k8s Pod requests more resources than are available within the cluster. The initial TaskExecutionEvent will report a Reason "task submitted to K8s". In a subsequent evaluation the plugin updates this Reason to "Unschedulable:0/1 nodes are available: 1 Insufficient memory.", however this is never reported to FlyteAdmin because the Phase and PhaseVersion remain the same.

Expected behavior

The FlyteConsole UI should always display the latest status. This requires FlytePropeller to correctly identify an update and therefore to send an event to FlyteAdmin reporting the change.

Additional context to reproduce

Start a workflow with resource requests larger than what is available in the cluster. In the task status' pane in the UI it will always display the "RUNNING" phase with a reason of "task submitted to K8s".

@task(requests=Resources(mem="12G"))
def say_hello_resources(name: str) -> str:
    return f"hello {name}"

@workflow
def my_wf_resources(name: str) -> str:
    res = say_hello_resources(name=name)
    return res

Screenshots

Are you sure this issue hasn't been raised already?

• Yes

Have you read the Code of Conduct?

• Yes