[ https://issues.apache.org/jira/browse/HBASE-7247?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13627662#comment-13627662 ]
Nicolas Liochon commented on HBASE-7247: ---------------------------------------- bq. This sync happens every time? Ain't it really expensive? More expensive that what used to be there (a write?). Why we need it? We already had the sync previously. We're just saving the final write. The sync can be expensive, I don't know if it's always expensive or sometimes optimized by ZK. My test hides its cost, because it's a single ZK node and the master is on the same node. A real life system with 5 ZK would be more realistic. In any case, saving the write is good. bq. Does this clear any existing watch? No, it does not set a watcher but should not impact existing ones. bq. Does the content of retransitionNodeOpening deserve to be generalized and moved back into transitionNode but with a flag for whether to write the new state or not? I don't know. Do we have a lot of case like this one? Globally, the performance problem is that we check the state before writing, hence we sync and we read. The next optimization I'm thinking about is to know that we were the previous writer, or that we read the previous version already, so we can send the write without syncing/reading, relying only on ZooKeeper versions. But it's easy to add bugs when doing this. > Assignment performances decreased by 50% because of > regionserver.OpenRegionHandler#tickleOpening > ------------------------------------------------------------------------------------------------ > > Key: HBASE-7247 > URL: https://issues.apache.org/jira/browse/HBASE-7247 > Project: HBase > Issue Type: Improvement > Components: master, Region Assignment, regionserver > Affects Versions: 0.95.2 > Reporter: Nicolas Liochon > Assignee: Nicolas Liochon > Fix For: 0.98.0, 0.95.1 > > Attachments: 7247.v1.patch, 7247.v2.patch > > > The regionserver.OpenRegionHandler#tickleOpening updates the region znode as > "Do this so master doesn't timeout this region-in-transition.". > However, on the usual test, this makes the assignment time of 1500 regions > goes from 70s to 100s, that is, we're 50% slower because of this. > More generally, ZooKeper commits to disk all the data update, and this takes > time. Using it to provide a keep alive seems overkill. At the very list, it > could be made asynchronous. > I'm not sure how necessary these updates are required (I need to go deeper in > the internal, feedback welcome), but it seems very important to optimize > this... The trival fix would be to make this optional. -- 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