Rob Earhart issues

Results 15 issues of


                                            Rob Earhart

Report task outcome when FEs fail to start

We need to send a task outcome when a function executor fails to start up. (The protocol allows tasks to be allocated to an FE before the FE completes startup.)

Remove the substring "fail" from test outputs except for failing tests

Using the string "fail" for anything other than a test outcome makes it harder than necessary to grep our logs for failing tests; given GitHub's test runner log output viewer's...

Implement limited-resource tests

We need to verify that with multiple graphs competing for limited resources, the server makes progress and finishes all of the invocations.

Implement Dynamic Scheduling

The overarching goal here is to smoothly run with a dynamic number of task executors, varying in physical location and hardware resources.

Merge heartbeats with executor state reporting

Run healthchecks after task failures

Implement churn tests

Add tests to demonstrate that we are not removing functions from allow-listed machines when the cluster doesn't have sufficient resources for running workflows.

Review executor protocol, optimize if possible

Implement model-based testing

The idea: let's use a model to validate the correctness of the Indexify state machine, to try to drive out any edge-case issues and increase our overall reliability.

List graph versions with active invocations

`GET /namespace//computegraphs/` should return a list of versions