feat: disk cache v1.0

[blueswhen]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @blueswhen, 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 introduces a significant feature by implementing a disk cache layer for the KV cache system. The primary goal is to extend the effective memory capacity for KV cache pages by allowing less frequently used pages to be stored on disk and reloaded on demand. This change aims to improve performance and resource utilization, especially for scenarios with large models or many concurrent requests, by reducing the reliance on expensive GPU or CPU memory for all cached data. The integration includes new worker processes, updated cache management logic, and new metrics to monitor disk cache activity.

Highlights

• Disk Cache Integration: Introduced a new disk-based caching mechanism to offload and retrieve KV cache pages, enhancing overall cache capacity and memory management.
• New DiskCacheWorker: Added a dedicated DiskCacheWorker class to manage the persistence and retrieval of KV cache pages to and from disk using PyLocalCacheService.
• Enhanced CPU Cache Client: Modified the CpuKvCacheClient with new methods for encoding/decoding offload page indexes, improved page lifecycle management (e.g., mark_pages_recyclable, recycle_pages), and support for grouped page offloading.
• Metrics and Logging: Added disk_prompt_cache_len to the request object and integrated its tracking and logging, allowing for better visibility into disk cache hit rates.
• Optimized Cache Matching Logic: Updated the multi_level_kv_cache manager to incorporate disk cache lookups, allowing the system to first check CPU cache, then disk cache for missing pages, and load them back into CPU memory as needed.

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 a disk cache feature, which is a significant addition to the multi-level KV cache system. The changes span multiple files to integrate disk-level caching, including a new DiskCacheWorker for background offloading/loading, modifications to the CPU cache client to handle new page states, and updates to the cache manager to orchestrate lookups across CPU and disk caches. While the overall approach is sound, there are several areas for improvement regarding maintainability, performance, and code clarity. Specifically, I've identified some hardcoded values that should be configurable, debugging artifacts that should be removed, and performance concerns with busy-wait loops and an excessively large thread pool. Refactoring the complex cache matching logic would also improve readability and future maintenance.