I am now trying to run it on cloudera on AWS, but having a problem: H2O starts to launch but times out in 120 seconds
What I did:
- I created a single m1.xlarge instance on AWS, then installed the latest CDH with the free license.
- I was sure that the CDH is running smoothly.
- I increased the maximun java heap size to 1G for hdfs, mapreduce and yarn.
- I downloaded and unzipped the latest h2o tool, currently it is 2.8.3.2. I also tried 2.8.2.8.
- I ran hadoop jar h2odriver_hdp2.1.jar water.hadoop.h2odriver -libjars ../h2o.jar -mapperXmx 256m -nodes 1 -output hdfsOutputDirName
- It starts to launch but times out in 120 seconds.
- I tried different jar files, hdp2.0, cdh4, cdh4_yarn, but they are not the solution.
I found the possible reason and solution here: http://ift.tt/1HAi4na. They are
- Heap Size Restriction.
- Communication between nodes.
But it seems not working with my case because
- I think I've already taken care about the max heap size restriction.
- I am only testing it on a single machine.
I wish I could know the reason behind the failures.
Aucun commentaire:
Enregistrer un commentaire