Wellington Chevreuil created HBASE-29623:
--------------------------------------------
Summary: Blocks for CFs with BlockCache disabled may still get
cached on write or compaction
Key: HBASE-29623
URL: https://issues.apache.org/jira/browse/HBASE-29623
Project: HBase
Issue Type: Bug
Reporter: Wellington Chevreuil
Assignee: Wellington Chevreuil
When disabling block cache on specific CFs, blocks for the given CF still get
cached during flushes or compaction if "hbase.rs.cacheblocksonwrite" and/or
"hbase.rs.cachecompactedblocksonwrite" are set to true at the global
configuration.
Even worse, this CF would always be cached uncompressed, even if
"hbase.block.data.cachecompressed" is turned on in the global configuration.
Having different SLAs for different portions of the dataset is a valid
scenario, and we should have a finer control over which data should benefit
from caching.
[The reference guide|https://hbase.apache.org/book.html#data.blocks.in.fscache]
section for the CF BLOCKCACHE configuration flag states that disabling
BLOCKCACHE at individual CFs should guarantee only META blocks are cached. We
should honour that behaviour.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
