[HBASE-12213] HFileBlock backed by Array of ByteBuffers

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Component/s: regionserver, Scanners
• Hadoop Flags: Reviewed

Release Note

The read path from the Hfileblock is now backed by a new data structure represented by an abstract class ByteBuff. The name is similar to nio ByteBuffer.
The new ByteBuff class has various APIs to read and write primitives to the underlying implementation similar to how nio ByteBuffer provides.
We have 2 implementations of the ByteBuff now - a single ByteBuffer backed implementation (SingleByteBuff) and a multiple ByteBuffer backed implementation (MultiByteBuff).
The single byte buffer (SingleByteBuff) backed implementation will be used for the HFileBlocks that come from HDFS and from LRU cachce.
The blocks that gets created out of BucketCache will be of type MultiByteBuff and the read operations are performed over this data structure which intelligently moves across different Byte buffers that it is formed of.

• Assignee: ramkrishna.s.vasudevan
• Reporter: Anoop Sam John
• Created: 09/Oct/14 05:53
• Updated: 17/Jun/22 05:13
• Resolved: 17/Jul/15 08:00

Description

In L2 cache (offheap) an HFile block might have been cached into multiple chunks of buffers. If HFileBlock need single BB, we will end up in recreation of bigger BB and copying. Instead we can make HFileBlock to serve data from an array of BBs.

Issue Links

• is blocked by: HBASE-13916 Create MultiByteBuffer an aggregation of ByteBuffers
• relates to: HBASE-14116 Change ByteBuff.getXXXStrictlyForward to relative position based reads
• relates to: HBASE-14117 Check DBEs where fields are being read from Bytebuffers but unused.
• links to: RB