fix: lightweight HTTP/2 parsing to reduce CPU on busy nodes

[mayankpande88]

Summary

• HTTP/2 frame parsing was consuming 70% of CPU (profiled via pprof on a pod with 47 containers). Most HTTP/2 traffic (gRPC, K8s API, service mesh) doesn't need payload capture — only status codes for L7 metrics.
• Adds lightweight mode to Http2Parser: skips DATA payload accumulation and RequestHeaders map for non-LLM traffic, auto-upgrades to full mode when :authority matches a known LLM provider
• Eliminates per-Parse() map allocations (statuses/grpcStatuses) by reusing maps with clear() — ~15% CPU savings
• Caps activeRequests at 100 per connection and parsers at 50 per container
• Fixes parser GC leak for connectionless parsers and removes duplicate GetActiveStreamsForLLM() call

What's preserved

• L7 metrics (status codes) — HEADERS frames always fully decoded
• LLM detection — auto-upgrades to full mode when LLM host detected
• gRPC status tracking — always decoded from HEADERS/Trailers
• Trace correlation — Http2Request still emits method/path/status

Test plan

• Deploy to dev cluster and verify L7 metrics still captured for HTTP/2 services
• Verify LLM tracking still works (Gemini/OpenAI calls captured)
• Compare CPU usage via pprof: expect 60-70% reduction on HTTP/2 hot path
• Check cAdvisor CPU: 4 pods at 1 core should drop to ~0.2-0.3 cores
• Verify no memory leak in googleHTTP2Parsers over time

[gemini-code-assist]

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request focuses on enhancing the efficiency of HTTP/2 parsing, particularly in environments with high HTTP/2 traffic. By introducing a lightweight parsing mode and optimizing memory management, it aims to drastically reduce CPU utilization and prevent resource exhaustion, while preserving essential L7 metrics and LLM detection capabilities.

Highlights

• Lightweight HTTP/2 Parsing: Introduced a lightweight HTTP/2 parsing mode to significantly reduce CPU consumption (up to 70%) on busy nodes by skipping payload capture for non-LLM traffic.
• Memory Allocation Optimization: Optimized memory allocation by reusing statuses and grpcStatuses maps, leading to approximately 15% CPU savings.
• Resource Limits: Implemented limits on active HTTP/2 requests per connection (100) and parsers per container (50) to prevent unbounded memory growth.
• Garbage Collection Fixes: Corrected a garbage collection leak for connectionless HTTP/2 parsers and removed a redundant call to GetActiveStreamsForLLM().

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• containers/container.go
  • Defined a constant maxHTTP2ParsersPerContainer to limit the number of HTTP/2 parsers.
  • Configured newly created Http2Parser instances to operate in lightweight mode and assigned an LLM host checker.
  • Adjusted LLM stream tracking to only activate for non-lightweight parsers.
  • Refined the garbage collection mechanism for HTTP/2 parsers to correctly handle connectionless instances.
• containers/llm.go
  • Added a helper function isLLMRelevantHost to identify LLM provider hosts.
• ebpftracer/l7/http2.go
  • Introduced maxActiveRequests to cap concurrent HTTP/2 streams per connection.
  • Modified Http2Parser to include Lightweight mode and an LLMHostChecker callback.
  • Implemented map reuse for statuses and grpcStatuses to reduce allocations.
  • Applied the maxActiveRequests limit when creating new HTTP/2 requests.
  • Added logic to dynamically switch from lightweight to full parsing mode upon detecting an LLM-relevant :authority header.
  • Conditionalized DATA frame payload accumulation based on the Lightweight flag.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces significant performance optimizations for HTTP/2 parsing, primarily by adding a lightweight parsing mode that avoids unnecessary payload processing for non-LLM traffic. This should substantially reduce CPU usage on busy nodes as intended. The changes also include other good optimizations like reusing maps to reduce allocations and capping resource usage to prevent unbounded memory growth. However, I've found a high-severity memory leak in the updated garbage collection logic for HTTP/2 parsers. While fixing a leak for connectionless parsers, it seems to have introduced a new one for parsers of closed connections in long-running processes. I've also left a medium-severity comment about a minor code duplication that could be refactored for better maintainability.