dpol1 opened a new pull request, #3062:
URL: https://github.com/apache/hugegraph/pull/3062
## Purpose of the PR
- close #3036
`CachedSchemaTransactionV2` registers a JVM-global watch on the meta store
so that a
schema change on one node clears the schema cache on the other nodes. That
watch goes
through `EtcdMetaDriver.listen` / `listenPrefix`, which handed jetcd the
bare
`Consumer<WatchResponse>` overload. That overload drops `onError` and
`onCompleted`, so
when jetcd ends a watch (for example after a transport reconnect tears
down the gRPC
stream) the listener stopped receiving events with no log line and no
exception. The node
kept serving stale schema and nothing reported it. `PdMetaDriver` does not
have this
problem because its `KvClient` already re-subscribes on error.
## Main Changes
- `EtcdMetaDriver.listen` / `listenPrefix` now register a `Watch.Listener`
instead of the
bare `Consumer` overload. On `onError` or `onCompleted` the driver
re-subscribes after a
1s backoff on a daemon thread, which is the same recovery `PdMetaDriver`
gets through
`KvClient`. A `WARN` is logged before each re-subscribe, so the failure
is now visible.
- Removed `CachedSchemaTransactionV2.resetMetaListenerForReconnect()`. It
was a manual
stopgap added in #3011 with no callers, and it never shipped in a tagged
release. Now
that the driver recovers on its own, the JVM-global register-once flag
is correct to stay
set, so the manual reset has nothing left to do. The lifecycle comment on
`metaEventListenerRegistered` is updated to describe the self-heal
behavior.
- The `MetaDriver` interface is unchanged, so neither implementor needs
edits beyond
`EtcdMetaDriver`.
Known limitation: the re-subscribe opens a fresh watch without a stored
revision, so any
cache-clear events emitted during the short reconnect window are not
replayed. This is the
same behavior as `PdMetaDriver` / `KvClient`. The change removes the
permanent silent
failure. It does not add gap replay.
## Verifying these changes
- [x] Need tests and can be verified as follows:
- New `EtcdMetaDriverTest` (JUnit + Mockito) mocks the jetcd
`Client`/`Watch`, captures
the registered `Watch.Listener`, and asserts that both `onError` and
`onCompleted`
trigger a re-subscribe (a second `watch(...)` call), and that
`onNext` still reaches
the consumer.
- Local run: `EtcdMetaDriverTest` 4/4, `CachedSchemaTransactionTest`
18/18, and
`MetaManagerSchemaCacheClearEventTest` 6/6 after the stopgap removal.
- Red-green: commenting out the re-subscribe makes the three recovery
tests fail;
restoring it makes them pass.
## Does this PR potentially affect the following parts?
- [ ] Dependencies
- [ ] Modify configurations
- [ ] The public API
- [ ] Other affects (typed here)
- [x] Nope
## Documentation Status
- [ ] `Doc - TODO`
- [ ] `Doc - Done`
- [x] `Doc - No Need`
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
