[HDDS-6549] Do not mark container replica as unhealthy on write failure

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Assignee: Unassigned
• Reporter: Hanisha Koneru
• Created: 04/Apr/22 22:47
• Updated: 06/Jun/23 20:09

Description

If an operation fails on an open/ closing container on a DN, it is marked unhealthy so that subsequent write transactions also fail. The container itself is marked for closing by sending a request to SCM.

Even if a write failed, other data in the container could be intact. Marking containers unhealthy should be a conservative operation. If let's say write fails on all replicas of a container, then all the replicas would be marked unhealthy. HDDS-6447 will ensure that replicas are not deleted if all the replicas are in unhealthy state. Irrespective of that, a container should not be marked as unhealthy when there is readable uncorrupted data in it.

Currently, we cannot mark the container with any other state as all container state changes are initiated from the SCM. If we quasi close a container replica, instead of marking it unhealthy, when there is a write failure, then all the state transition logic in SCM would also have to be changed/updated.

Issue Links

• is related to: HDDS-7096 Clarify locations of on-demand scanning