Wei-Chiu Chuang created HDDS-17625:
--------------------------------------
Summary: Verify Recon startup migrations preserve supported
pre-finalization downgrade paths
Key: HDDS-17625
URL: https://issues.apache.org/jira/browse/HDDS-17625
Project: Apache Ozone
Issue Type: Task
Reporter: Wei-Chiu Chuang
Recon intentionally finalizes its own layout automatically on startup,
independently of OM/SCM finalization. PR #10443 (HDDS-15374) explicitly
preserves this behavior:
https://github.com/apache/ozone/pull/10443
In Ozone 2.2.1, ReconServer invokes
ReconLayoutVersionManager.finalizeLayoutFeatures() during startup:
https://github.com/apache/ozone/blob/ozone-2.2.1/hadoop-ozone/recon/src/main/java/org/apache/hadoop/ozone/recon/ReconServer.java#L169-L179
The administrator guide states that the cluster can be downgraded before
finalization, but does not explain how this applies to Recon's migrated
metadata. Documentation clarification is tracked in HDDS-17624.
This task is a compatibility investigation, not an assertion that automatic
finalization is a bug. No downgrade failure has been reproduced in this
investigation.
Scope:
* Identify the supported source/target release pairs and audit Recon SQL and
RocksDB metadata changes, including data written after startup migration.
* For supported pairs, start the older Recon with representative metadata,
upgrade Recon while OM/SCM remain pre-finalized, let synchronization/tasks run,
then restart the older Recon with the resulting metadata.
* Verify startup, health, metadata reads, synchronization, and background tasks
after downgrade; startup success alone is insufficient.
* Determine which state can be rebuilt and which must be preserved. Specify any
required backup/restore procedure and information lost during a rebuild.
* Add appropriate compatibility coverage and document the supported downgrade
contract.
* If a supported downgrade fails, provide a reproducer and track the required
code fix. Consider compatible migrations, deferring incompatible changes, or
supported recovery; do not assume waiting for OM/SCM finalization alone is
sufficient because newer Recon code may require the migrated schema.
Acceptance criteria: Supported Recon downgrade paths have explicit evidence and
appropriate regression coverage, or a reproduced failure and linked code-fix
issue. The outcome informs HDDS-17624.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
