Add config option to control maximum gateway worker thread count

[Hellebore]

Motivation

On very high (burst) load scenarios, we can get connection refused errors quite quickly, if there are no worker threads available.

We should make it configurable, to see if we can bypass some of them.

Changes

• Add config option GATEWAY_WORKER_COUNT to control maximum number of worker threads

TODO

What's left to do:

• Test real life impact of this for certain high connection count scenarios
• Discuss naming of the variable
• Discuss where to set the variable.. We might want to opt to set it in the GatewayServer (in http/hypercorn.py) instead, or even in the serve_gateway variable, to limit the variables use to the "main gateway".

[Hellebore]

This pull request was cloned from https://github.com/localstack/localstack/pull/10016

Note: the URL is not a link to avoid triggering a notification on the original pull request.

[Hellebore]

Sourcery Review:

Hey @Hellebore - I've reviewed your changes and they look great!

General suggestions:

• Consider validating the GATEWAY_WORKER_COUNT value to ensure it falls within a reasonable range, preventing misconfiguration that could lead to system instability.
• Document the new configuration option, including guidance on how to choose an appropriate value based on expected load and available system resources.
• Review the impact of this change on existing deployments to ensure that the default behavior remains consistent and that performance is not negatively impacted for users who do not adjust the new setting.

Here's what I looked at during the review

• 🟡 General issues: 3 issues found
• 🟢 Security: all looks good
• 🟢 Testing: all looks good
• 🟢 Complexity: all looks good
• 🟢 Docstrings: all looks good

---

LangSmith trace

_{Help me be more useful! Please click 👍 or 👎 on each comment to tell me if it was helpful.}

[Hellebore]

suggestion (performance): Consider providing a default value for threads that aligns with system capabilities.

Using None as a default for threads and later substituting it with a hardcoded value (1000) might not be optimal for all environments. Consider a default that scales with the number of available CPU cores.

Suggested change

	threads: int | None = None,
	threads: int = os.cpu_count() or 4,

[Hellebore]

suggestion (code_refinement): Use of or for default parameter handling is clear but consider centralizing default values.

While the use of or for providing a default value is straightforward, centralizing default values (e.g., as constants) can make future adjustments and readability better.

Suggested change

	self.executor = _ThreadPool(threads or 1000, thread_name_prefix="asgi_gw")
	DEFAULT_THREAD_COUNT = 1000 # Centralized default value for easier adjustments and readability
	self.executor = _ThreadPool(threads or DEFAULT_THREAD_COUNT, thread_name_prefix="asgi_gw")

[Hellebore]

suggestion (code_refinement): Ensure environment variable fallback values are consistent across the codebase.

The fallback of 1000 for GATEWAY_WORKER_COUNT is used directly in multiple places. Consider defining this as a constant to ensure consistency and ease of maintenance.

Suggested change

	GATEWAY_WORKER_COUNT = int(os.environ.get("GATEWAY_WORKER_COUNT") or 1000)
	# Define a constant for the default gateway worker count
	DEFAULT_GATEWAY_WORKER_COUNT = 1000
	GATEWAY_WORKER_COUNT = int(os.environ.get("GATEWAY_WORKER_COUNT", DEFAULT_GATEWAY_WORKER_COUNT))