[ https://issues.apache.org/jira/browse/HBASE-5898?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13264470#comment-13264470 ]
Zhihong Yu commented on HBASE-5898: ----------------------------------- Consider the case where off heap cache is enabled. >From DoubleBlockCache: Suppose getBlock() is executed without the lock (first pass in the new loop of readBlock) and doesn't find cacheKey from onHeapCache but finds it in offHeapCache - it will call onHeapCache.cacheBlock(): {code} public Cacheable getBlock(BlockCacheKey cacheKey, boolean caching) { Cacheable cachedBlock; if ((cachedBlock = onHeapCache.getBlock(cacheKey, caching)) != null) { stats.hit(caching); return cachedBlock; } else if ((cachedBlock = offHeapCache.getBlock(cacheKey, caching)) != null) { if (caching) { onHeapCache.cacheBlock(cacheKey, cachedBlock); } {code} Another thread calls cacheBlock() around the same time and executes onHeapCache.cacheBlock() for the same cacheKey: {code} public void cacheBlock(BlockCacheKey cacheKey, Cacheable buf) { onHeapCache.cacheBlock(cacheKey, buf); offHeapCache.cacheBlock(cacheKey, buf); } {code} I think there is a race condition which didn't exist before the proposed change: the entries for the same cacheKey in onHeapCache and offHeapCache would diverge. If off heap cache is disabled, I don't see problem with proposed optimization. > Consider double-checked locking for block cache lock > ---------------------------------------------------- > > Key: HBASE-5898 > URL: https://issues.apache.org/jira/browse/HBASE-5898 > Project: HBase > Issue Type: Improvement > Components: performance > Affects Versions: 0.94.1 > Reporter: Todd Lipcon > Attachments: hbase-5898.txt > > > Running a workload with a high query rate against a dataset that fits in > cache, I saw a lot of CPU being used in IdLock.getLockEntry, being called by > HFileReaderV2.readBlock. Even though it was all cache hits, it was wasting a > lot of CPU doing lock management here. I wrote a quick patch to switch to a > double-checked locking and it improved throughput substantially for this > workload. -- 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