alpass163gmail opened a new pull request, #18425:
URL: https://github.com/apache/iotdb/pull/18425
Description
What
Makes authority operations (GRANT, REVOKE, CREATE/DROP USER/ROLE, ALTER
USER, etc.) highly available by integrating their cache invalidation into
the existing ClusterCachePropagator infrastructure that was already used by
DDL operations.
Why
Previously, the AuthOperationProcedure used a bespoke retry loop with
SyncDataNodeClientPool to invalidate permission caches on DataNodes
one-at-a-time. This had several issues:
- Blocking sequential invalidation: each DataNode was contacted
synchronously in series, making the procedure slow and fragile when a node
is unreachable.
- No fencing awareness: the old code had no integration with
MetadataLeaseManager self-fencing. It would retry indefinitely against
nodes that would never respond.
- Inconsistent with the rest of the codebase: every other
metadata-broadcast procedure (CREATE TABLE, DROP TABLE, etc.) already used
ClusterCachePropagator with the PROCEED/WAIT/FAIL verdict model. Authority
was the odd one out.
How
AuthOperationProcedure — replaced the iterative dataNodesToInvalid list +
SyncDataNodeClientPool retry loop with a single call to
ClusterCachePropagator.propagate(), which broadcasts to all DataNodes in
parallel. The propagator safely skips any DataNode that has been silent for
longer than the fence timeout, relying on the self-fencing guarantee to
prevent stale reads.
ClusterAuthorityFetcher — replaced the checkCacheAvailable() guard (which
tracked a cacheOutDate flag and heartBeatTimeStamp) with
failIfMetadataLeaseFenced() at every metadata-access entry point. This is
the same lease-protection pattern used by the schema, table, and pipe
subsystems. Unused methods (refreshToken, invalidateAllCache) and their
i18n messages were removed.
BasicAuthorityCache — removed post-invalidation validator checks
(getIfPresent after invalidate) that could produce false negatives and
block cache operations on a completely healthy node.
Scope
- iotdb-core/confignode: AuthOperationProcedure, async request type +
manager registration
- iotdb-core/datanode: ClusterAuthorityFetcher, BasicAuthorityCache,
IAuthorityFetcher
- integration-test: new IoTDBTableAuthHAIT covering user/role/permission
management under both DataNode-down and DataNode-ReadOnly failure modes
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
