Implementation:Sgl project Sglang Engine Shutdown

Knowledge Sources	SGLang
Domains	LLM_Serving, Resource_Management
Last Updated	2026-02-10 00:00 GMT

Overview

Concrete tool for shutting down the SGLang Engine and terminating all associated subprocesses.

Description

The Engine.shutdown() method calls kill_process_tree on the current process ID (with include_parent=False) to terminate all child subprocesses — the Scheduler and DetokenizerManager. This is also registered as an atexit handler during initialization and called automatically when using the context manager (with statement).

Usage

Call Engine.shutdown() when you are done with inference to release GPU memory and system resources. Prefer the context manager pattern for automatic cleanup.

Code Reference

Source Location

• Repository: sglang
• File: python/sglang/srt/entrypoints/engine.py
• Lines: L443-445

Signature

def shutdown(self):
    """Shutdown the engine"""
    kill_process_tree(os.getpid(), include_parent=False)

Import

# Engine.shutdown() is a method on the Engine instance
engine.shutdown()

I/O Contract

Inputs

Name	Type	Required	Description
self	Engine	Yes	Active Engine instance to shut down

Outputs

Name	Type	Description
(none)	None	All subprocesses terminated, resources freed

Usage Examples

Explicit Shutdown

import sglang as sgl

engine = sgl.Engine(model_path="meta-llama/Llama-3.1-8B-Instruct")
output = engine.generate("Hello!", {"max_new_tokens": 32})
print(output["text"])

# Explicit shutdown
engine.shutdown()

Context Manager (Recommended)

import sglang as sgl

with sgl.Engine(model_path="meta-llama/Llama-3.1-8B-Instruct") as engine:
    output = engine.generate("Hello!", {"max_new_tokens": 32})
    print(output["text"])
# Automatically calls engine.shutdown() on exit

Related Pages

Implements Principle

• Principle:Sgl_project_Sglang_Engine_Lifecycle_Management