Trino worker memory does not release

I’m running trino on k8s。After the job finished, the worker’s memory is still holding。

We have configured autoscaling recarding moemory usage。If the workers always hold meomry, the autoscaling will not work。
Are there any configuration to solve this issure ? Any suggestions?


This is normal and how Trino and any Java application works. Once memory is allocated to the JVM it is managed by the JVM. Internally garbage collection takes care of everything but for the operating system the memory seems used. Since you should use dedicated nodes for each worker and the coordinator anyway, this is what you want. All the available hardware is used to the max.