Copilot commented on code in PR #14657:
URL: https://github.com/apache/gravitino/pull/14657#discussion_r4478233128
##########
flink-connector/v2.2/flink/build.gradle.kts:
##########
@@ -142,6 +142,7 @@ dependencies {
testImplementation(libs.metrics.core)
testImplementation(libs.minikdc)
testImplementation(libs.mockito.core)
Review Comment:
`mockito-inline` is typically used as a replacement for `mockito-core` (it
provides the inline mock maker). Keeping both can lead to redundant/duplicated
artifacts on the test classpath and makes dependency resolution harder to
reason about. Prefer swapping `libs.mockito.core` for `libs.mockito.inline`
(and apply the same change to the other Flink version modules updated in this
PR) unless there’s a concrete reason to keep both.
##########
flink-connector/flink-common/src/main/java/org/apache/gravitino/flink/connector/catalog/BaseCatalog.java:
##########
@@ -1216,7 +1218,22 @@ protected static Representation[]
buildSqlRepresentation(String dialect, String
}
protected Catalog catalog() {
- return GravitinoCatalogManager.get().getGravitinoCatalogInfo(getName());
+ // The Gravitino client has no catalog cache and Flink calls catalog()
before nearly every
+ // schema or table request, so load the catalog once per instance. Load it
again when the
+ // shared manager has been replaced, since the handle keeps using the REST
client of the
+ // manager that loaded it.
+ GravitinoCatalogManager manager = GravitinoCatalogManager.get();
+ LoadedCatalog loaded = loadedCatalog;
+ if (loaded == null || loaded.manager != manager) {
+ synchronized (this) {
+ loaded = loadedCatalog;
+ if (loaded == null || loaded.manager != manager) {
+ loaded = new LoadedCatalog(manager,
manager.getGravitinoCatalogInfo(getName()));
+ loadedCatalog = loaded;
+ }
+ }
+ }
+ return loaded.catalog;
Review Comment:
There’s a race window where `GravitinoCatalogManager.get()` can start
returning a new manager after line 1225, but this call can still return a
catalog handle loaded from the old manager (or skip reload because it compares
against the stale `manager` local). If the old manager has been closed,
returning a handle bound to its REST client can fail. Consider re-fetching the
manager inside the synchronized block (and using that manager for the second
check/load), or performing an additional `GravitinoCatalogManager.get()` check
immediately before returning to ensure the returned handle matches the current
manager.
--
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]