[ https://issues.apache.org/jira/browse/HBASE-5898?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13529517#comment-13529517 ]
Lars Hofhansl commented on HBASE-5898: -------------------------------------- I tracked it down to locking on istream in HFileBlock.readAtOffset. Here's my test case: 20m rows, single column family, single column, blockcache disabled for the scan, all data fits into the OS buffer cache. Full scan over all rows. One client scanning: 15s (regionserver keeps one core busy ~120%) Two clients scanning along the same set of blocks: They both time out. (regionserver is a 5-6% CPU, clearly just waiting) Then I changed readAtOffset to always do preads. Now: One client scanning: 39s (regionserver at ~120%) Two clients scanning: 39s each (regionserver at ~210%) So not sure how to proceed. Generally switching to pread obviously does not work (scan time almost tripled). seek + read does not scale to multiple threads. I *bet* this is the issue folks have been seeing and this patch does not address that. > 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 > Assignee: Todd Lipcon > Priority: Critical > Fix For: 0.94.3, 0.96.0 > > Attachments: 5898-0.94.txt, 5898-TestBlocksRead.txt, 5898-v2.txt, > 5898-v3.txt, 5898-v4.txt, 5898-v4.txt, HBASE-5898-0.patch, > HBASE-5898-1.patch, HBASE-5898-1.patch, 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 For more information on JIRA, see: http://www.atlassian.com/software/jira