Within the Scala REPL everything is working fine. Even you application session is down but you run the same code again you will see there is a new job getting created. But zeppelin has such a problem. you can do a test. run a notebook and get the job finished at the end of the notebook. Then re-run the same notebook again then it will get stucked. Running spark code on Scala REPL and Zeppelin are different From: Paul Brenner Date: 2017-02-17 12:37 To: users Subject: Re: Re: Zeppelin unable to respond after some time I don’t believe that this explains my issue. Running the Scala REPL also keeps a session alive for as long as the REPL is running. I’ve had REPLs open for days (shhhhh don’t tell anyone) that have correspondingly kept sessions alive for the same period of time with no problem. I only see this issue in zeppelin.
We run zeppelin on a server and allow multiple users to connect, each with their own interpreters. We also find that zeppelin memory usage on the server will steadily creep up over time. Executing sys.exit in a spark paragraph, restarting the interpreter, and using yarn application -kill often will cause zeppelin to end the related interpreter process but not always. So over time we find that many zombie processes pile up and eat up resources. The only way to keep on top of this is to regularly login to the zeppelin server and kill zombie jobs. Here is a command that I’ve found helpful. When you know that a specific user has no active zeppelin interpreters running then execute the following: ps aux | grep zeppelin | grep "2BSGYY7S8" | grep java | awk -F " " '{print$2}' | xargs sudo -u yarn kill -9 where “2BSGYY7S8" is the interpreter id (found in interpreter.json) and “yarn” is actually the name of the user that originally started zeppelin with: zeppelin-daemon.sh start To kill every interpreter except for a specific users just flip it around with: ps aux | grep zeppelin | grep -v "2BSGYY7S8" |grep -v "zeppelin.server.ZeppelinServer" | grep java | awk -F " " '{print$2}' | xargs sudo -u yarn kill -9 If I do this every few days zeppelin keeps humming along pretty smoothly most of the time. Paul Brenner DATA SCIENTIST (217) 390-3033 On Fri, Feb 17, 2017 at 3:23 PM "xyun...@simuwell.com" <">"xyun...@simuwell.com" > wrote: The problem could be not only the resource, but the session. If you run a chunk of spark code and you should see the a running application in the spark UI, but in your code if you shut it down after the job is finished, then on the spark UI you will see the hob is finished. Within zeppelin, each job will start the spark session only once(different interpreter mode could be set if you want notebooks to share the session or not), if you closed it ,it will never restart it again. The only way to get the same code work again is to restat the interpreter or restart zeppelin. I`m not sure if I explain clearly, but hope it could help From: Paul Brenner Date: 2017-02-17 12:14 To: users Subject: Re: Re: Zeppelin unable to respond after some time I’ve definitely had this problem with jobs that don’t take all the resources on the cluster. Also, my experience matches what others have reported: just restarting zeppelin and re-runing the stuck paragraph solves the issue. I’ve also experienced this problem with for loops. Some for loops which write to disk but absolutely don’t have any variables that are increasing in size will hang in Zeppelin. If I run the exact same code in the scala REPL it goes through without problem. Paul Brenner DATA SCIENTIST (217) 390-3033 On Fri, Feb 17, 2017 at 2:12 PM "xyun...@simuwell.com" <">"xyun...@simuwell.com" > wrote: I have solved the similar issue before. You should check on spark UI and probably you will see your single job is taking all the resources. Therefore further job that submitting to the same cluster will just hang on there. When you restart zeppelin then the old job is killed and all the resource it took will be released xyun...@simuwell.com From: RUSHIKESH RAUT Date: 2017-02-17 02:29 To: users Subject: Re: Zeppelin unable to respond after some time Yes happens with r and spark codes frequently On Feb 17, 2017 3:25 PM, "小野圭二" <onoke...@gmail.com> wrote: yes, almost every time. There are not any special operations. Just run the tutorial demos. From my feeling, it happens in R demo frequently. 2017-02-17 18:50 GMT+09:00 Jeff Zhang <zjf...@gmail.com>: Is it easy to reproduce it ? 小野圭二 <onoke...@gmail.com>于2017年2月17日周五 下午5:47写道: I am facing on the same issue now. 2017-02-17 18:25 GMT+09:00 RUSHIKESH RAUT <rushikeshraut...@gmail.com>: Hi all, I am facing a issue while using Zeppelin. I am trying to load some data(not that big data) into Zeppelin and then build some visualization on it. The problem is that when I try to run the code first time it's working but after some time the same code doesn't work. It remains in running state on gui, but no logs are generated in Zeppelin logs. Also all further tasks are hanging in pending state. As soon as I restart Zeppelin it works. So I am guessing it's some memory issue. I have read that Zeppelin stores the data in memory so it is possible that it runs out of memory after some time. How do I debug this issue? How much is the default memory that Zeppelin takes at start? Also is there any way that I can run Zeppelin with specified memory so that I can start the process with more memory. Because it doesn't make sense to restart Zeppelin after every half hour Thanks, Rushikesh Raut