GCS Actor Management On Failing Tests

[rkooo567]

This tracks tests that are broken when GCS actor management is on.

@ffbin and I are collaborating to fix all test failures to turn GCS actor management on by the end of next week.

Merged

• [P0]test_actor_restart_on_node_failure fails locally on Mac (@ffbin fixed it by GCS server error handling for actor creation #8899)
• [P0]GCS actor management should use a different Redis Accessor (@shanly [GCS] fix actor accessor initialize inside redis gcs client #9284 (review))
• [P0]Fix gcs actor manager destroy actor crash bug (@ffbin fixed it by Fix gcs actor manager destroy actor crash bug #9329)
• [P0]BUG fix: It will failed when register two or more detached actors with empty name. It will failed when register two or more detached actors with empty name #9282 (@shanly, [GCS] Fix detached actor with empty name #9283)
• [P0]test_drivers_release_resources crashes GCS on Mac (@rkooo567 fixed it by [Core] GCS Actor Management Race Condition #9215)
• [P0] Race condition at CreatingOnWorker [GCS Actor Management] Race condition between CreatingOnWorker and CancelOnWorker #9287 (@rkooo567 fixed it by [Core] GCS Actor Management Race Condition #9215)
• [P0] Dashboard shows ALIVE state when actor handles go out of scope [GCS Actor Management] Dashboard still renders ALIVE state when actors are DEAD  #9203 (@rkooo567 fixed by [GCS Actor Management] Fix a dashboard doesn't show the correct actor state. #9275)
• [P0] test_calling_put_on_actor_handle crashes GCS on Mac (@rkooo567, fixed by [Core] GCS Actor Management Race Condition #9215)
• [P0] Fix miss destroy actor notify message bug [GCS Actor Management] Removing actor as soon as it creates them doesn't work as expected. #9267 (@ffbin, Fix the bug that PollOwnerForActorOutOfScope's reply may be missing #9330 )
• [P0] Detached actor is broken. Detached actor is broken when gcs actor management is on. #9383 (@rkooo567)
• [P1] Worker resource leak when GCS server restarts [Core]Worker resource leak when GCS server restarts #9286 (@ffbin, [GCS]Fix lease worker leak bug when gcs server restarts #9315)
• [P0] test_multi_node.py::test_multi_driver_logging hangs test_multi_node.py::test_multi_driver_logging hangs when GCS actor management is turned on.  #9505 (@ffbin )

In Review

In Progress

Need Assignee

[rkooo567]

It seems like CI didn't turn on GCS actor management. (The test failures detected here was not detected in the CI #8845) I am doing some experiment to make sure CI turns it on properly.

#9382

[rkooo567]

It is on by default! Time to fix bugs :)