[FLINK-11177] Node health manager

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Abandoned
• Component/s: Runtime / Coordination
• Assignee: ryantaocer
• Reporter: ryantaocer
• Created: 17/Dec/18 09:20
• Updated: 31/Jul/20 10:17
• Resolved: 31/Jul/20 10:17

Description

• A NodeHealthManager module aims to monitor the health states of nodes including machine, and task manager workload.
• It's beyond the general blacklist machenism since blacklist is an extreme case.
• It provides runtime metrics to the scheduler in JM who can then more such as load balance, bad TMs skipping, and even slots scoring.