[HBASE-12295] Prevent block eviction under us if reads are in progress from the BBs

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.0.0
• Component/s: regionserver, Scanners
• Hadoop Flags: Reviewed

Release Note

We try to delay the eviction of the block till the cellblocks are formed at the Rpc layer. A simple reference counting mechanism is introduced when ever a block is accessed from the Bucket cache. Once a scanner completes using a block the reference count is decremented. The eviction of the block happens only when the reference count of that block is 0.
We also introduce a concept of ShareableMemory based on the type of blocks we create from the Block cache. The blocks from the ByteBufferIOEngine directly refer to the buckets in offheap and such blocks are marked SHARED memory type. The blocks from LRU, HDFS and file mode of Bucket cache are all marked EXCLUSIVE because these blocks have their own exclusive memory.
For the CP case, any cell coming out of SHARED memory block is copied before returning the results, because CPs can use the results as its state so that eviction cannot corrupt the results.

• Assignee: ramkrishna.s.vasudevan
• Reporter: ramkrishna.s.vasudevan
• Created: 20/Oct/14 03:16
• Updated: 17/Jun/22 05:13
• Resolved: 23/Jul/15 09:02

Description

While we try to serve the reads from the BBs directly from the block cache, we need to ensure that the blocks does not get evicted under us while reading. This JIRA is to discuss and implement a strategy for the same.

Issue Links

• is blocked by: HBASE-13926 Close the scanner only after Call#setResponse
• is related to: HBASE-21957 Unify refCount of BucketEntry and refCount of hbase.nio.ByteBuff into one