[HBASE-23277] CompactedHFilesDischarger may lead to high cpu usage

• Type: Brainstorming
• Status: Resolved
• Priority: Minor
• Resolution: Incomplete
• Assignee: Unassigned
• Reporter: junfei liang
• Created: 11/Nov/19 03:32
• Updated: 11/Jun/22 23:48
• Resolved: 11/Jun/22 23:48

Description

in one of  online cluster(RS:16C72G, hbase version is 2.x) ,  cpu average usage is 30%,  but cpu usage may above 80%  occasionally.  by a  enchanced 'top -H' script, i find  'RS_COMPACTED_FILES_DISCHARGER' threads can use  50%+ cpu ( 10 * 80% ).  

 

 

by jstack , we found most of stacks is at.

 "org.apache.hadoop.hbase.io.hfile.LruBlockCache#evictBlocksByHfileName"

   public int evictBlocksByHfileName(String hfileName) {
      int numEvicted = 0;
      for (BlockCacheKey key : map.keySet()) {
         if (key.getHfileName().equals(hfileName))

{              if (evictBlock(key))              ++numEvicted;          }

      }
      if (victimHandler != null)

{           numEvicted += victimHandler.evictBlocksByHfileName(hfileName);       }

      return numEvicted;
 }

 

when close storefile, now we evict it's blocks from cache,  but we have to  traverse the whole blockcache to locate the block cached, and can lead to high cpu usage.

do we need evict blockcache when close file, even for a small file ( size less than 10MB)?  it's a lru block cache, the unused block will be automatically removed from cache when blockcache usage is high?