[HBASE-5074] support checksums in HBase block cache

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.94.0, 0.95.0
• Component/s: regionserver
• Hadoop Flags: Reviewed

Release Note

Adds hbase.regionserver.checksum.verify. If hbase.regionserver.checksum.verify is set to true, then hbase will read data and then verify checksums. Checksum verification inside hdfs will be switched off. If the hbase-checksum verification fails, then it will switch back to using hdfs checksums for verifiying data that is being read from storage. Also adds hbase.hstore.bytes.per.checksum -- number of bytes in a newly created checksum chunk -- and hbase.hstore.checksum.algorithm, name of an algorithm that is used to compute checksums.

You will currently only see benefit if you have the local read short-circuit enabled -- see http://hbase.apache.org/book.html#perf.hdfs.configs -- while HDFS-3429 goes unfixed.

• Tags: 0.96notable
• Assignee: Dhruba Borthakur
• Reporter: Dhruba Borthakur
• Created: 20/Dec/11 06:12
• Updated: 15/Oct/13 04:46
• Resolved: 09/Mar/12 00:32

Description

The current implementation of HDFS stores the data in one block file and the metadata(checksum) in another block file. This means that every read into the HBase block cache actually consumes two disk iops, one to the datafile and one to the checksum file. This is a major problem for scaling HBase, because HBase is usually bottlenecked on the number of random disk iops that the storage-hardware offers.

Issue Links

• is related to: HDFS-2699 Store data and checksums together in block file
• relates to: HBASE-5720 HFileDataBlockEncoderImpl uses wrong header size when reading HFiles with no checksums
• relates to: HBASE-5864 Error while reading from hfile in 0.94
• relates to: HBASE-6868 Skip checksum is broke; are we double-checksumming by default?