[HBASE-12531] bug in cachedataonwrite

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 0.98.6.1
• Component/s: regionserver
• Assignee: Unassigned
• Reporter: dennislee
• Created: 19/Nov/14 04:38
• Updated: 17/Jun/22 18:37
• Resolved: 28/Dec/14 04:40

Description

when configuring hbase.rs.cacheblocksonwrite as true on a region server ,or setting cacheDataOnWrite as true on a column family in a table ,we flush the header bytes,ondisk bytebuffer and checksum bytes to disk,but only store the header and uncompressedBytesWithoutHeader to the block cache .

so if we read a block from block cache which cached on write ,the method getBufferWithoutHeaderof org.apache.hadoop.hbase.io.hfile.HFileBlock will cut off the head and checksum bytes,even if the checksum was never written ,and then we get a IllegalArgumentException thrown by ByteBuffer cause there is not enough bytes to read or skip at the end of the ByteBuffer.

I fixed this problem but I don't know how to commit a patch ,so I paste my code here :

org.apache.hadoop.hbase.io.hfile.HFileBlock.java

public ByteBuffer getBufferWithoutHeader() {
		int length,lengthWithoutHeader=buf.limit()-headerSize(),lengthWithoutHeaderAndCheckSum=lengthWithoutHeader - totalChecksumBytes();

        if(lengthWithoutHeader==uncompressedSizeWithoutHeader){ // NO check sum tail
            length=lengthWithoutHeader;
        } else if(lengthWithoutHeaderAndCheckSum==uncompressedSizeWithoutHeader){ // has check sum tail
            length=lengthWithoutHeaderAndCheckSum;
        } else {
            throw new IllegalArgumentException(this.toString()+",this block may be crashed");
        }
        ByteBuffer buffer =ByteBuffer.wrap(buf.array(),
                    buf.arrayOffset() + headerSize(),length)//length
                    .slice();
           
        return buffer;
}

Issue Links

• relates to: HBASE-12270 A bug in the bucket cache, with cache blocks on write enabled