github-actions[bot] commented on code in PR #69629:
URL: https://github.com/apache/doris/pull/69629#discussion_r4385226347
##########
fe/fe-core/src/main/java/org/apache/doris/datasource/metacache/MetaCache.java:
##########
@@ -662,6 +687,90 @@ public void invalidate(String localName, long id) {
}
}
+ /** Fences old object loads while retaining unrelated cached objects and
their ID mappings. */
+ public void invalidateLookup(String localName, boolean allObjects) {
+ metaObjLifecycleLock.writeLock().lock();
+ try {
+ synchronized (namesMutationLock) {
+ invalidateNames();
+ metaObjGeneration.incrementAndGet();
+ lookupObservationGeneration++;
+ if (allObjects) {
+ metaObjCache.invalidateAll();
+ idToName.clear();
+ } else {
+ metaObjCache.invalidate(localName);
+ idToName.entrySet().removeIf(entry ->
entry.getValue().equals(localName));
+ }
+ }
+ } finally {
+ metaObjLifecycleLock.writeLock().unlock();
+ }
+ }
+
+ /** Blocks missing-result repair during a metadata-only refresh without
evicting names or objects. */
+ public Runnable beginMetadataRefresh() {
+ metaObjLifecycleLock.writeLock().lock();
+ try {
+ metadataRefreshes++;
+ lookupObservationGeneration++;
+ } finally {
+ metaObjLifecycleLock.writeLock().unlock();
+ }
+ return () -> {
+ metaObjLifecycleLock.writeLock().lock();
+ try {
+ metadataRefreshes--;
+ lookupObservationGeneration++;
+ } finally {
+ metaObjLifecycleLock.writeLock().unlock();
+ }
+ };
+ }
+
+ /** Captures matching resident membership and object state; never loads
remote metadata. */
+ public Optional<Runnable> observeLookup(String localName, String
remoteName, T expected,
+ boolean allObjects, Runnable invalidateMetadata) {
+ metaObjLifecycleLock.readLock().lock();
+ try {
+ synchronized (namesMutationLock) {
+ NamesCacheValue names = namesCache.getIfPresent("");
+ Optional<T> value = metaObjCache.getIfPresent(localName);
+ Pair<String, String> pair = names == null ? null :
names.names.get(localName);
+ if (metadataRefreshes != 0 || names == null || !names.complete
|| pair == null
Review Comment:
[P2] Retire hot identities when refreshed names already omit the table
In a case-sensitive HMS catalog, warm `SHOW TABLES` and `DESC ctl.db.t`,
drop `t` remotely, and let the normal asynchronous names refresh finish so a
later `SHOW TABLES` omits it. `loadNames()` replaces only the names snapshot;
the old table object and `idToName` entry remain. `getTableNullable("t")` still
returns that object through `getMetaObjInternal()`'s resident fast path. A
subsequent `SHOW CREATE TABLE ctl.db.t` performs the guarded live HMS read and
confirms `NoSuchObjectException`, but this `pair == null` check prevents any
repair action. Repeated requests keep the old object hot, and numeric lookup
continues resolving the dropped table indefinitely. Permit repair of the
matching resident object/ID when the retained complete names snapshot already
omits the name (or retire omitted objects during names refresh), while
preserving replacement/epoch fences. Add a names-refresh-first regression
asserting object and numeric-route retirement. This is distinct from the prior
SHOW
CREATE entry-point thread: the read now enters the guard, but observation
rejects the refreshed membership state.
--
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]
