Off-heap memory leak in Spark < 2.1: PoolThreadCaches aren't getting freed?
See original GitHub issueExpected behavior
PoolThreadCache should be freed when thread it is bound to terminates.
Actual behavior
We observed off-heap memory leak in our application that is run in Spark 1.6.2 on Amazon EMR 4.8.0. Analyzing driver’s heap dump I noticed that PoolThreadLocalCache’s caches counter doesn’t decrease which makes me thinking that PoolThreadCache is not getting freed: At the same time there are not that many threads:
Steps to reproduce
I don’t think it is specific to our application but rather to Spark/Netty itself.
Minimal yet complete reproducer code (or URL to code)
I can provide heap dumps if needed.
Netty version
4.0.29.Final
JVM version (e.g. java -version
)
java version “1.8.0_45” Java™ SE Runtime Environment (build 1.8.0_45-b14) Java HotSpot™ 64-Bit Server VM (build 25.45-b02, mixed mode)
OS version (e.g. uname -a
)
Amazon (emr-4.8.0), Linux 4.4.11-23.53.amzn1.x86_64 #1 SMP Wed Jun 1 22:22:50 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux
Issue Analytics
- State:
- Created 6 years ago
- Comments:9 (6 by maintainers)
Top GitHub Comments
Then maybe you should create an issue in spark ?
On 6. June 2017 at 20:28:25, Yauheni Audzeichyk (notifications@github.com) wrote:
@Scottmitch https://github.com/scottmitch unfortunately, I can not upgrade to a newer version of Netty as Spark on EMR comes bundled with it. Also that property is not available in 4.0.29.Final.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/netty/netty/issues/6819#issuecomment-306575848, or mute the thread https://github.com/notifications/unsubscribe-auth/AAa0QgUoNu8bbvpg1aiPGpRLwev-mwvIks5sBZpJgaJpZM4NxPV3 .
@yaud - great thanks for closing the loop!