Is there an existing issue for this?

• I have searched the existing issues.

Current Behavior

Protocols client pools are global instances retaining various clients within local maps with hashed configuration as key. They are not correctly cleaned up after an instance terminates execution.

Expected Behavior

• Use ExecutionId within hashing to correctly cleanup corresponding clients
• Retain global maps but introduce LRU policy where unused clients are atomically cleaned up (and atomically created if an engine will require it again in the future)