[
https://issues.apache.org/jira/browse/HBASE-5001?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13172546#comment-13172546
]
stack commented on HBASE-5001:
------------------------------
It does caveat (saving Todd having to intercede), 'Profilers give you distorted
view')
Here's a few:
+ util.IdLock releaseLockEntry and getLockEntry take a long and then autobox to
check Map. The autoboxing shows as 8% (could keep the Long around down inside
IdLock and save '4%'):
3.9% - 3,255 kB - 2,322 alloc.
org.apache.hadoop.hbase.util.IdLock.releaseLockEntry
3.7% - 3,035 kB - 2,208 alloc. org.apache.hadoop.hbase.util.IdLock.getLockEntry
+ The put and remove from HBaseServer call queue looks expensive in profiler as
does keeping up our MVCC read/write points. Would be sweet to redo these w/
some kinda lockless datastructure (lmax disruptor?). The call queue shows as
20% if you add put and take.
HBaseServer this.callQueue = new LinkedBlockingQueue<Call>(maxQueueSize);
There are some allocation savings we could do too... a bit of sizing of results
could save some resizings. E.g.:
2.2% - 13,042 ms
org.apache.hadoop.hbase.util.ByteBufferOutputStream.checkSizeAndGrow
If you want access to jprofiler, key just say. jprofiler donated us a license
for committers.
> Improve the performance of block cache keys
> -------------------------------------------
>
> Key: HBASE-5001
> URL: https://issues.apache.org/jira/browse/HBASE-5001
> Project: HBase
> Issue Type: Improvement
> Affects Versions: 0.90.4
> Reporter: Jean-Daniel Cryans
> Assignee: Lars Hofhansl
> Priority: Minor
> Fix For: 0.92.0
>
> Attachments: 5001-0.92.txt, 5001-v1.txt, 5001-v2.txt
>
>
> Doing a pure random read test on data that's 100% block cache, I see that we
> are spending quite some time in getBlockCacheKey:
> {quote}
> "IPC Server handler 19 on 62023" daemon prio=10 tid=0x00007fe0501ff800
> nid=0x6c87 runnable [0x00007fe0577f6000]
> java.lang.Thread.State: RUNNABLE
> at java.util.Arrays.copyOf(Arrays.java:2882)
> at
> java.lang.AbstractStringBuilder.expandCapacity(AbstractStringBuilder.java:100)
> at
> java.lang.AbstractStringBuilder.append(AbstractStringBuilder.java:390)
> at java.lang.StringBuilder.append(StringBuilder.java:119)
> at
> org.apache.hadoop.hbase.io.hfile.HFile.getBlockCacheKey(HFile.java:457)
> at
> org.apache.hadoop.hbase.io.hfile.HFileReaderV2.readBlock(HFileReaderV2.java:249)
> at
> org.apache.hadoop.hbase.io.hfile.HFileBlockIndex$BlockIndexReader.seekToDataBlock(HFileBlockIndex.java:209)
> at
> org.apache.hadoop.hbase.io.hfile.HFileReaderV2$ScannerV2.seekTo(HFileReaderV2.java:521)
> at
> org.apache.hadoop.hbase.io.hfile.HFileReaderV2$ScannerV2.seekTo(HFileReaderV2.java:536)
> at
> org.apache.hadoop.hbase.regionserver.StoreFileScanner.seekAtOrAfter(StoreFileScanner.java:178)
> at
> org.apache.hadoop.hbase.regionserver.StoreFileScanner.seek(StoreFileScanner.java:111)
> at
> org.apache.hadoop.hbase.regionserver.StoreFileScanner.seekExactly(StoreFileScanner.java:219)
> at
> org.apache.hadoop.hbase.regionserver.StoreScanner.<init>(StoreScanner.java:80)
> at
> org.apache.hadoop.hbase.regionserver.Store.getScanner(Store.java:1689)
> at
> org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.<init>(HRegion.java:2857)
> {quote}
> Since the HFile name size is known and the offset is a long, it should be
> possible to allocate exactly what we need. Maybe use byte[] as the key and
> drop the separator too.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
