ramkrish86 commented on a change in pull request #257: HBASE-22463 Some paths
in HFileScannerImpl did not consider block#release which will exhaust the
ByteBuffAllocator
URL: https://github.com/apache/hbase/pull/257#discussion_r288941586
##########
File path:
hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileReaderImpl.java
##########
@@ -520,20 +520,18 @@ public HFileScannerImpl(final HFile.Reader reader, final
boolean cacheBlocks,
}
void updateCurrBlockRef(HFileBlock block) {
- if (block != null && this.curBlock != null &&
- block.getOffset() == this.curBlock.getOffset()) {
+ if (block != null && curBlock != null && block.getOffset() ==
curBlock.getOffset()) {
return;
}
- // We don't have to keep ref to EXCLUSIVE type of block
- if (this.curBlock != null && this.curBlock.usesSharedMemory()) {
Review comment:
So already as part of this JIRA you removed the MEMTYPE and so even the
bucket cache engines like FileIOEngine will remove the memtype and so all the
blocks will go through the release() way by adding to prevBlocks and then
getting released. The new PR #268 will remove the need for addition to
prevBlocks as it is a noop. So now both bucket cache and read from HDFS both
will create onheap/offheap blocks and the type of block will determine what it
is .
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
