apache / amoro

Apache Amoro (incubating) is a Lakehouse management system built on open data lake formats.
https://amoro.apache.org/
Apache License 2.0
747 stars 260 forks source link

[Improvement]: Eliminate AMS Full GC impactions deriving from local terminal clean spark context #2968

Open nicochen opened 1 week ago

nicochen commented 1 week ago

Search before asking

What would you like to be improved?

We experienced and suffered from AMS full GC for a long period. we dig into this case and finally found out local spark session terminal triggers java.lang.System.gc() every 30 minutes to clean contexts. As arthas stack traces as shown below: image

How should we improve?

For external kyuubi session terminal, there is no need to revises anything. The problem is only related to local session which is embedding into AMS itself. Simply add config 'spark.cleaner.referenceTracking' = 'false' to solve the problem.

Are you willing to submit PR?

Subtasks

No response

Code of Conduct