[FLINK-21380] Restarting/Canceling/Failing expose globally-terminal EG

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.13.0
• Component/s: Runtime / Coordination
• Labels: pull-request-available
• Assignee: Chesnay Schepler
• Reporter: Chesnay Schepler
• Created: 16/Feb/21 12:56
• Updated: 19/Feb/21 07:35
• Resolved: 19/Feb/21 07:06

Description

When the declarative scheduler is in the Restarting state it can expose a fully Canceled EG, because it only overrides getJobStatus, but the archived execution graph still contains the Canceled job state.
This breaks the contract that the EG only transitions into a terminal state once, and never transitions into another state afterwards.

Related to that the Canceling/Failing states can also expose such an EG because State#onGloballyTerminalState is called asynchronously. A globally-terminal EG should only be visible in the Finished state however.

Issue Links

• links to: GitHub Pull Request #14951