You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@zeppelin.apache.org by "Mori[A]rty (JIRA)" <ji...@apache.org> on 2017/04/17 12:04:41 UTC
[jira] [Created] (ZEPPELIN-2414) Memory leak under scoped mode of
SparkInterpreter caused by inapproprately setting Thread.contextClassLoader
Mori[A]rty created ZEPPELIN-2414:
------------------------------------
Summary: Memory leak under scoped mode of SparkInterpreter caused by inapproprately setting Thread.contextClassLoader
Key: ZEPPELIN-2414
URL: https://issues.apache.org/jira/browse/ZEPPELIN-2414
Project: Zeppelin
Issue Type: Bug
Components: Interpreters
Affects Versions: 0.6.2
Environment: {quote}
jdk version: jdk1.7.0_67
spark interpreter env:
export MASTER=local\[4\]
export SPARK_SUBMIT_OPTIONS="--driver-memory 2G"
{quote}
Reporter: Mori[A]rty
Fix For: 0.6.3
When using scoped mode of SparkInterpreter, after several repetitions of these three steps (create-notebook -> run a paragraph -> remove-notebook), the heap of RemoteInterpreterServer process will grow to 100% rapidly.
For example, in my local environment, RemoteInpreterServer's max heap size is 2 GB. After I repeatedly run a simple paragrah
{quote}%spark
sc{quote}
15 times (each time in a new notebook and delete the notebook after running), RemoteInpreterServer's heap has no more free space and the 15th execution of the paragrah was never finished.
heap occupation:
{quote}
S0 S1 E O P YGC YGCT FGC FGCT GCT
0.00 0.00 100.00 99.98 49.38 19 2.304 1093 798.293 800.597
{quote}
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)