[MESOS-8886] Task in UNREACHABLE forever if a new agent register with the same IP when the old agent is unreachable state.

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: master
• Labels: mesosphere partition
• Assignee: Unassigned
• Reporter: Gilbert Song
• Created: 07/May/18 06:54
• Updated: 08/May/18 17:19

Description

For partition aware framework, if a new agent register with the master using the old agent's IP address (the old agent is in the UNREACHABLE state), the task (in UNREACHABLE state) should be transitioned to a terminal state when the old agent is removed.

Some backgrounds:
When a new agent register with the master using the old agent's IP address:
1. if the old agent is in DISCONNECTED state, the master will remove the old agent and mark all old tasks as LOST;
2. if the old agent is in UNREACHABLE state, the master does nothing regarding the agents or its tasks (they would've already been marked UNREACHABLE when the agent went unreachable).