gianm commented on code in PR #17824:
URL: https://github.com/apache/druid/pull/17824#discussion_r2017588026
##########
server/src/main/java/org/apache/druid/metadata/segment/cache/HeapMemorySegmentMetadataCache.java:
##########
@@ -71,10 +71,20 @@
/**
* In-memory implementation of {@link SegmentMetadataCache}.
* <p>
- * Reads can be done from the cache only after it has synced with the metadata
store.
- * Writes can happen even when the sync is in progress. This is because the
- * {@link #syncWithMetadataStore()} is able to reconcile differences based on
- * the update time of a transaction.
+ * Only used segments (excluding num_rows and schema_fingerpring) and
+ * pending segments are cached. Unused segments are not cached.
+ * <p>
+ * Non-leader Overlords also keep polling the metadata store to keep the cache
+ * up-to-date in case leadership changes.
+ * <p>
+ * Cache usage modes: {@link UsageMode}:
+ * <p>
+ * The map {@link #datasourceToSegmentCache} contains the cache for each
datasource.
Review Comment:
That sort of reference counting approach would work too. Yes, it's OK to do
this in a future PR, especially since this area of the code is still evolving
and isn't switched on by default yet.
--
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]
