[MESOS-9558] Track removed agents' operations in the Framework struct

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: master
• Labels: foundations mesosphere
• Epic Link: Operation Feedback Improvements
• Assignee: Unassigned
• Reporter: Greg Mann
• Created: 06/Feb/19 20:43
• Updated: 05/Apr/19 17:00

Description

The master currently relies on the agent ID in operation reconciliation requests in order to send OPERATION_UNREACHABLE and OPERATION_GONE_BY_OPERATOR updates. This also means that in the case of an operation that was terminal-but-unacknowledged at the time of agent removal (i.e. operation was FINISHED but the update not yet acknowledged), the framework will receive OPERATION_UNREACHABLE/OPERATION_GONE_BY_OPERATOR for reconciliation requests, when the master could report the actual terminal state if it were retained.

We should consider updating the master to track unreachable/gone operations explicitly in order to provide better reconciliation information in these cases.