[FLINK-4354] Implement heartbeat logic between TaskManager and ResourceManager

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Component/s: Runtime / Coordination
• Assignee: Zhijiang
• Reporter: Zhijiang
• Created: 10/Aug/16 09:11
• Updated: 13/Jul/18 06:41
• Resolved: 23/Mar/17 14:02

Description

It is part of work for Flip-6.

The HeartbeatManager is mainly used for monitoring heartbeat target and reporting payloads.

For ResourceManager side, it would trigger monitoring the HeartbeatTarget when receive registration from TaskManager, and schedule a task to requestHeartbeat at interval time. If not receive heartbeat response within duration time, the HeartbeatListener will notify heartbeat timeout, then the ResourceManager should remove the internal registered TaskManager.

For TaskManger side, it would trigger monitoring the HeartbeatTarget when receive registration acknowledgement from ResourceManager. An it will also be notified heartbeat timeout if not receive heartbeat request from ResourceManager within duration time.

The current implementation will not interact payloads via heartbeat, and it can be added if needed future.

Issue Links

• contains: FLINK-7472 Release task managers gracefully
• is blocked by: FLINK-4355 Implement TaskManager side of registration at ResourceManager
• is blocked by: FLINK-4478 Implement heartbeat logic
• is part of: FLINK-4478 Implement heartbeat logic
• links to: GitHub Pull Request #2770
• links to: GitHub Pull Request #3591