Apache9 commented on a change in pull request #3947:
URL: https://github.com/apache/hbase/pull/3947#discussion_r776588895
##########
File path:
hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ChunkCreator.java
##########
@@ -134,51 +134,36 @@ public static ChunkCreator getInstance() {
return instance;
}
- /**
- * Creates and inits a chunk. The default implementation for a specific
chunk size.
- * @return the chunk that was initialized
- */
- Chunk getChunk(ChunkType chunkType) {
- return getChunk(CompactingMemStore.IndexType.ARRAY_MAP, chunkType);
Review comment:
> MemStoreLABImpl.getNewExternalChunk(ChunkType) is called by
CellChunkImmutableSegment.allocIndexChunks to allocate
IndexChunk(CellChunkImmutableSegment itself is IndexType.CHUNK_MAP), because
the IndexType.ARRAY_MAP, it does not put the IndexChunk in
ChunkCreator.chunkIdMap when the IndexChunk could not be pooled , but it does
no matter because it is a IndexChunk not a DataChunk, and IndexChunk is put in
CellChunkMap.chunks, which prevents it is GC-ed. If we put the IndexChunk in
ChunkCreator.chunkIdMap when using the correct
> IndexType.CHUNK_MAP, there not cause memory leak because when the
MemStoreLAB is closed, ChunkCreator.removeChunk is invoked to remove the chunk
from this ChunkCreator.chunkIdMap
Here, you also explained that, in CellChunkImmutableSegment, when creating
IndexChunk, we used to use ARRAY_MAP, but after the change, we will use
CHUNK_MAP. And the performance is about reading, not writing... I'm not sure
why they used to choose ARRAY_MAP for index chunks...
--
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]
