> On Feb. 3, 2014, 9:10 p.m., Gunther Hagleitner wrote: > > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezSessionPoolManager.java, > > line 28 > > <https://reviews.apache.org/r/17471/diff/1/?file=453262#file453262line28> > > > > BlockingQueue should be able to tell you length, right?
Blocking queue has a size that reflects the number of elements in the queue not the size of the queue itself. The length can be computed using the size and the remaining capacity from the queue in a synchronized method which seems somewhat round-about to me. > On Feb. 3, 2014, 9:10 p.m., Gunther Hagleitner wrote: > > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezSessionPoolManager.java, > > line 65 > > <https://reviews.apache.org/r/17471/diff/1/?file=453262#file453262line65> > > > > don't think this is needed Addressed above. > On Feb. 3, 2014, 9:10 p.m., Gunther Hagleitner wrote: > > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezSessionPoolManager.java, > > line 73 > > <https://reviews.apache.org/r/17471/diff/1/?file=453262#file453262line73> > > > > is this the right name? shouldn't that be a yarn var? The tez queue name is the variable used by tez to allocate containers on a specific queue. This is again dependent on the yarn queues already. So according to the tez team we can just depend on this variable. Also, mapreduce has the behavior that its variable will return the queue 'default' if no queue is specified. We would like to choose a set of queues as default and not depend on yarn's default queue. > On Feb. 3, 2014, 9:10 p.m., Gunther Hagleitner wrote: > > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezSessionPoolManager.java, > > line 106 > > <https://reviews.apache.org/r/17471/diff/1/?file=453262#file453262line106> > > > > It doesn't look like you're keeping track of this sessionstate here. I > > think we should. The user should always get/return sessions and we handle > > the alloc/dealloc. (why can't return close the session for non default for > > instance?) We want to retain the efficiencies in having pre-launched containers. Closing the non-default sessions would let go of that. - Vikram ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/17471/#review33496 ----------------------------------------------------------- On Jan. 28, 2014, 10:34 p.m., Vikram Dixit Kumaraswamy wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/17471/ > ----------------------------------------------------------- > > (Updated Jan. 28, 2014, 10:34 p.m.) > > > Review request for hive. > > > Bugs: HIVE-6325 > https://issues.apache.org/jira/browse/HIVE-6325 > > > Repository: hive-git > > > Description > ------- > > Enable using multiple concurrent sessions in tez. > > > Diffs > ----- > > common/src/java/org/apache/hadoop/hive/conf/HiveConf.java 84ee78f > itests/util/src/main/java/org/apache/hadoop/hive/ql/QTestUtil.java 9ad5986 > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TestTezSessionState.java > PRE-CREATION > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezSessionPoolManager.java > PRE-CREATION > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezSessionState.java b8552a3 > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezSessionStateFactory.java > PRE-CREATION > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/TezTask.java c6f431c > ql/src/java/org/apache/hadoop/hive/ql/session/SessionState.java d7edda1 > ql/src/test/org/apache/hadoop/hive/ql/exec/tez/TestTezSessionPool.java > PRE-CREATION > service/src/java/org/apache/hive/service/server/HiveServer2.java fa13783 > > Diff: https://reviews.apache.org/r/17471/diff/ > > > Testing > ------- > > Added multi-threaded junit tests. > > > Thanks, > > Vikram Dixit Kumaraswamy > >
