Devesh Kumar Singh created HDDS-15035:
-----------------------------------------
Summary: Harden DN container delete / import flows against stale
on-disk state and improve failure diagnosability
Key: HDDS-15035
URL: https://issues.apache.org/jira/browse/HDDS-15035
Project: Apache Ozone
Issue Type: Epic
Reporter: Devesh Kumar Singh
Assignee: Devesh Kumar Singh
*Description:*
Recent investigation of DN-side container inconsistencies identified a class
of issues where on-disk container state, in-memory state, and
replication/import behavior can diverge after partial delete or failed import
scenarios. The review discussion narrowed the actionable work to three areas:
- make DN replication/import more resilient when a stale destination
directory exists on one volume
- preserve the original filesystem exception in delete failure logs so field
diagnosis is possible
- harden failed-import cleanup ordering to avoid leaving the most problematic
partial on-disk shape
This epic tracks the short-term implementation work needed to improve
correctness, operational resilience, and debuggability in these paths without
redesigning the full delete-state model.
Scope:
- DN replication/import path handling of destination directory collisions
- delete-path logging improvements
- failed-import cleanup ordering
Out of scope:
- introduction of a new delete-pending state
- background reconciliation worker for failed deletes
- full redesign of delete semantics
- broad DN-wide duplicate-container scans unless later justified
Success criteria:
- DN import/replication can recover from per-volume stale directory
collisions when another eligible volume exists
- delete failures preserve the original root I/O cause in logs
- failed import cleanup no longer prefers the more dangerous chunks-only
residual state
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
