[HBASE-3320] Compaction parameter minCompactSize should be configurable

• Type: Improvement
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.90.0, 0.92.0
• Fix Version/s: 0.90.0
• Hadoop Flags: Reviewed
• Assignee: Jonathan Gray
• Reporter: Jonathan Gray
• Created: 08/Dec/10 20:27
• Updated: 20/Nov/15 12:42
• Resolved: 09/Dec/10 01:21

Description

In the new compaction algorithm, there is a minCompactSize parameter which is used to include files in selection if they are under this size.

Currently this is:

    this.minCompactSize = this.region.memstoreFlushSize * 3 / 2; // +50% pad

In a cluster, we had high setting of flush size but were doing frequent small flushes because of HLog eviction. This led to pathological cases in compaction file selection that this algorithm is trying to avoid in the first place.

Should make this configurable but the existing value should be fine for default in most applications.

Issue Links

• duplicates: HBASE-3290 Max Compaction Size