Hi Ever, As Haibo noted, that’s a known regression.
If you fall back to the older approach of having multiple TMs per slave, each with one slot, then Flink (as of 1.7/1.8) does a better job of distributing work. — Ken > On Jul 1, 2019, at 9:23 PM, Haibo Sun <[email protected]> wrote: > > Hi, Ever > > This is a regression wrt the pre Flip-6 code, and the following JIRA > dedicated to this issue. > > https://issues.apache.org/jira/browse/FLINK-12122 > <https://issues.apache.org/jira/browse/FLINK-12122> > > Best, > Haibo > > > > At 2019-07-02 11:42:27, "Ever" <[email protected]> wrote: > Hi, there're 3 taskManager nodes within our testing flink cluster, whose > version is 1.8. And each one have 10 taskslots. > > Now I have a job with parallelism 3. > I expected the 3 tasks will be located at 3 different taskManagers, just as > Example 2 below: > <[email protected]> > > But it came out that all 3 tasks are all located at the same taskmanager. > <[email protected]> > > <[email protected]> > > Why? -------------------------- Ken Krugler +1 530-210-6378 http://www.scaleunlimited.com Custom big data solutions & training Flink, Solr, Hadoop, Cascading & Cassandra
