We're going to double check with one of the core devs tomorrow, but most likely we need both labels on the pod. I think what we want to do is validate that the executor has given us valid run ids and reject their update if they haven't.

We're going to double check with one of the core devs tomorrow, but most likely we need both labels on the pod. I think what we want to do is validate that the executor has given us valid run ids and reject their update if they haven't.

I think adding validation on the server side is definitely ideal. I'd be happy to add that as well just let me know!

@Sovietaced - by all means!

@dave-gantenbein should be good for review. I think the go mod CI job flaked so should be unrelated.

Got a unit test flake that seems unrelated

=== RUN   TestGetJobsByState/anyOf
    getjobs_test.go:816: 
        	Error Trace:	/home/runner/work/armada/armada/internal/lookoutv2/repository/getjobs_test.go:816
        	Error:      	Not equal: 
        	            	expected: &model.Job{Annotations:map[string]string{}, Cancelled:<nil>, Cpu:15000, Duplicate:false, EphemeralStorage:107374182400, Gpu:8, JobId:"01hz0qfqhw4ag8rvtvw8rvde8b", JobSet:"job-set-1", LastActiveRunId:(*string)(0xc000726870), LastTransitionTime:time.Date(2022, time.March, 1, 15, 4, 5, 0, time.UTC), Memory:51539607552, Owner:"user-1", Namespace:(*string)(0xc0008184c0), Priority:12, PriorityClass:(*string)(0xc0007267f0), Queue:"queue-1", Runs:[]*model.Run{(*model.Run)(0xc000053e60)}, State:"RUNNING", Submitted:time.Date(2022, time.March, 1, 15, 4, 5, 0, time.UTC), CancelReason:(*string)(nil), Node:(*string)(0xc000726880), Cluster:"cluster-1", ExitCode:(*int32)(nil), RuntimeSeconds:70790764}
        	            	actual  : &model.Job{Annotations:map[string]string{}, Cancelled:<nil>, Cpu:15000, Duplicate:false, EphemeralStorage:107374182400, Gpu:8, JobId:"01hz0qfqhw4ag8rvtvw8rvde8b", JobSet:"job-set-1", LastActiveRunId:(*string)(0xc000796ff0), LastTransitionTime:time.Date(2022, time.March, 1, 15, 4, 5, 0, time.UTC), Memory:51539607552, Owner:"user-1", Namespace:(*string)(0xc000797008), Priority:12, PriorityClass:(*string)(0xc000797020), Queue:"queue-1", Runs:[]*model.Run{(*model.Run)(0xc0007de420)}, State:"RUNNING", Submitted:time.Date(2022, time.March, 1, 15, 4, 5, 0, time.UTC), CancelReason:(*string)(nil), Node:(*string)(0xc0007e0480), Cluster:"cluster-1", ExitCode:(*int32)(nil), RuntimeSeconds:70790765}
        	            	
        	            	Diff:
        	            	--- Expected
        	            	+++ Actual
        	            	@@ -63,3 +63,3 @@
        	            	  ExitCode: (*int32)(<nil>),
        	            	- RuntimeSeconds: (int32) 70790764
        	            	+ RuntimeSeconds: (int32) 70790765
        	            	 })
        	Test:       	TestGetJobsByState/anyOf
--- FAIL: TestGetJobsByState/anyOf (0.00s)
FAIL internal/lookoutv2/repository.TestGetJobsByState/anyOf (0.00s)

@Sovietaced Can you rebase this branch from master, then if the unit tests pass again at least once, this should be good to go.

@Sovietaced Can you rebase this branch from master, then if the unit tests pass again at least once, this should be good to go.

done

@Sovietaced - we discussed this and while it won't cause any negative impact, I'm wondering if you've updated things so the executors and server are on the same version, which would be the recommended pattern. If there's something blocking you from doing that let me know here or in slack and we can merge this, but again this isn't how we'd recommend you go about setting things up long term, as other issues could potentially creep in and we won't be testing this scenario...

@Sovietaced - we discussed this and while it won't cause any negative impact, I'm wondering if you've updated things so the executors and server are on the same version, which would be the recommended pattern. If there's something blocking you from doing that let me know here or in slack and we can merge this, but again this isn't how we'd recommend you go about setting things up long term, as other issues could potentially creep in and we won't be testing this scenario...

We haven't updated things yet and have been running a fork but I have already mentioned I realize this is not a supported deployment. I don't think the PR makes the codebase any worse considering there is no validation being done in the scheduler API but you do you.