[PR #968] [MERGED] [Store] feat: store key-value data in buckets #1928

New issue

Closed

opened 2026-10-29 21:31:27 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 21:31:27 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/Mooncake/pull/968
Author: @zhuxinjie-nz
Created: 10/26/2025
Status: ✅ Merged
Merged: 10/31/2025
Merged by: @ykwd

Base: main ← Head: sequential-storage-dev

📝 Commits (10+)

c749715 [Store] feat: implement a SequentialStorageBackend to store key-value data in buckets for efficient data management and maintenance.
754fa85 fix(store): The test directory for StorageBackendTest does not exist.
80aca15 Store: Fix test data path issue in StorageBackendTest
5b4c4e1 Store: Fix a memory leak issue in StorageBackendTest
6653ffa Limit thread number in ci make to avoid ci failure out of not enough disk space
33d43c2 Limit thread number in ci make to avoid ci failure out of not enough disk space
f08ef11 Remove previous build output to avoid ci failure out of not enough disk space
20a79fc try to optimize ci disk space
f6d861c Store: Implement SharedMutex Thread Safety Analysis and apply it to StorageBackend
c25c9ad Store: Rename SequentialStorageBackend to BucketStorageBackend

📊 Changes

10 files changed (+1378 additions, -47 deletions)

View changed files

📝 .github/workflows/ci.yml (+9 -4)
📝 mooncake-store/include/mutex.h (+132 -3)
📝 mooncake-store/include/storage_backend.h (+190 -3)
📝 mooncake-store/include/types.h (+5 -0)
📝 mooncake-store/include/utils.h (+2 -0)
📝 mooncake-store/src/storage_backend.cpp (+543 -36)
📝 mooncake-store/src/utils.cpp (+6 -0)
📝 mooncake-store/tests/CMakeLists.txt (+2 -1)
➕ mooncake-store/tests/mutex_test.cpp (+176 -0)
➕ mooncake-store/tests/storage_backend_test.cpp (+313 -0)

📄 Description

Summary

Implement a SequentialStorageBackend to store key-value data in buckets for efficient data management and maintenance.

Motivation

Group KV data into buckets to avoid creating too many small files.
Add metadata support in file storage for KV data to enable LRU/LFU/creation-time-based eviction later.

Behavior & Compatibility

The SequentialStorageBackend currently supports only low-level bucket I/O and is not fully integrated into the pipeline. No impact on Mooncake in its current state.
The follow-up implementation approach is inconsistent with the current StorageBackend's design pattern.Key inconsistencies include:
1. On write: create buckets by synchronizing the offloading queue with the master via a heartbeat mechanism, enabling better scalability for independently deployed clients and supporting high-availability client scenarios.
2. On read: enable distributed read capability for bucket data via global metadata registration, coordinated through the Transfer Engine.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/kvcache-ai/Mooncake/pull/968 **Author:** [@zhuxinjie-nz](https://github.com/zhuxinjie-nz) **Created:** 10/26/2025 **Status:** ✅ Merged **Merged:** 10/31/2025 **Merged by:** [@ykwd](https://github.com/ykwd) **Base:** `main` ← **Head:** `sequential-storage-dev` --- ### 📝 Commits (10+) - [`c749715`](https://github.com/kvcache-ai/Mooncake/commit/c749715cf3d54a98a9161d8c36d60c739ab3eb5a) [Store] feat: implement a SequentialStorageBackend to store key-value data in buckets for efficient data management and maintenance. - [`754fa85`](https://github.com/kvcache-ai/Mooncake/commit/754fa85a8ddefb582e4933ce945ed7f629f4e9a9) fix(store): The test directory for StorageBackendTest does not exist. - [`80aca15`](https://github.com/kvcache-ai/Mooncake/commit/80aca15a53c8e70524df30605b8506eb1074508b) Store: Fix test data path issue in StorageBackendTest - [`5b4c4e1`](https://github.com/kvcache-ai/Mooncake/commit/5b4c4e1eb69276ac85910e7266683a8c9c266b7e) Store: Fix a memory leak issue in StorageBackendTest - [`6653ffa`](https://github.com/kvcache-ai/Mooncake/commit/6653ffaf7205dfd6ee75aa74d9d944d547014f6a) Limit thread number in ci make to avoid ci failure out of not enough disk space - [`33d43c2`](https://github.com/kvcache-ai/Mooncake/commit/33d43c26603472c60f921f61cd0196dcd35fa05f) Limit thread number in ci make to avoid ci failure out of not enough disk space - [`f08ef11`](https://github.com/kvcache-ai/Mooncake/commit/f08ef1105f2b2a85232f74d579e6b7c954a89e0d) Remove previous build output to avoid ci failure out of not enough disk space - [`20a79fc`](https://github.com/kvcache-ai/Mooncake/commit/20a79fcdecfc9766e764cd698fdf3ea643020353) try to optimize ci disk space - [`f6d861c`](https://github.com/kvcache-ai/Mooncake/commit/f6d861c42ad15101289ee324cf8a9c0adfe1b750) Store: Implement SharedMutex Thread Safety Analysis and apply it to StorageBackend - [`c25c9ad`](https://github.com/kvcache-ai/Mooncake/commit/c25c9adc3f2e43d9aa513f560a182e74c6eeff8b) Store: Rename SequentialStorageBackend to BucketStorageBackend ### 📊 Changes **10 files changed** (+1378 additions, -47 deletions) <details> <summary>View changed files</summary> 📝 `.github/workflows/ci.yml` (+9 -4) 📝 `mooncake-store/include/mutex.h` (+132 -3) 📝 `mooncake-store/include/storage_backend.h` (+190 -3) 📝 `mooncake-store/include/types.h` (+5 -0) 📝 `mooncake-store/include/utils.h` (+2 -0) 📝 `mooncake-store/src/storage_backend.cpp` (+543 -36) 📝 `mooncake-store/src/utils.cpp` (+6 -0) 📝 `mooncake-store/tests/CMakeLists.txt` (+2 -1) ➕ `mooncake-store/tests/mutex_test.cpp` (+176 -0) ➕ `mooncake-store/tests/storage_backend_test.cpp` (+313 -0) </details> ### 📄 Description ### Summary Implement a SequentialStorageBackend to store key-value data in buckets for efficient data management and maintenance. ### Motivation - Group KV data into buckets to avoid creating too many small files. - Add metadata support in file storage for KV data to enable LRU/LFU/creation-time-based eviction later. ### Behavior & Compatibility - The SequentialStorageBackend currently supports only low-level bucket I/O and is not fully integrated into the pipeline. No impact on Mooncake in its current state. - The follow-up implementation approach is inconsistent with the current StorageBackend's design pattern.Key inconsistencies include: 1. On write: create buckets by synchronizing the offloading queue with the master via a heartbeat mechanism, enabling better scalability for independently deployed clients and supporting high-availability client scenarios. 2. On read: enable distributed read capability for bucket data via global metadata registration, coordinated through the Transfer Engine. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 21:31:27 +00:00