[PR #217] [MERGED] fix(overlaybd): preallocate cache blocks to avoid SIGBUS when disk is…

bsakaguchi commented

2026-10-29 14:21:34 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/AgentENV/pull/217
Author: @huang-jl
Created: 8/25/2026
Status: ✅ Merged
Merged: 8/26/2026
Merged by: @yingdi-shan

Base: main ← Head: hjl/fix-cache-sigbus

📝 Commits (1)

80e6470 fix(overlaybd): preallocate cache blocks to avoid SIGBUS when disk is full

📊 Changes

3 files changed (+254 additions, -7 deletions)

View changed files

📝 storage/overlaybd/src/backend/cache/full_file_cache/cache_entry.rs (+135 -0)
📝 storage/overlaybd/src/backend/cache/full_file_cache/cache_store.rs (+22 -7)
📝 storage/overlaybd/src/backend/cache/tests.rs (+97 -0)

📄 Description

What

Reserve the block's byte range with fallocate(FALLOC_FL_KEEP_SIZE) before handing out mmap write access (write_block and mmap_mut_for_block). Disk exhaustion now surfaces as a plain ENOSPC refill error, and reads fall back to the source exactly like the existing cache-capacity-full path. Filesystems without fallocate support (EOPNOTSUPP/ENOSYS) keep the previous behavior.

Why

The full-file cache writes refilled blocks through a shared mmap over a sparse data file. Writing into an unallocated hole makes the kernel allocate filesystem blocks at page-fault time; when the local disk is full that allocation fails and the process is killed with SIGBUS, taking down the whole ublk-daemon.

Design and behavior changes

Before write into mmap region of cache files, an extra fallocate syscall will be invoked.

Validation

make fmt
make clippy
make test-unit
Relevant Rust integration tests
make -C services test (required when services/ changes)
Generated clients/server regenerated with the documented make target
Documentation updated
Benchmarks or performance comparison completed

Checklist

The PR contains one coherent change and no unrelated formatting or refactoring.
New behavior is covered by tests, or I explained why testing is impractical.
Logs and examples contain no credentials, tokens, or private registry information.
I did not manually edit generated code without updating its source and regenerating it.

_{🔄 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/AgentENV/pull/217 **Author:** [@huang-jl](https://github.com/huang-jl) **Created:** 8/25/2026 **Status:** ✅ Merged **Merged:** 8/26/2026 **Merged by:** [@yingdi-shan](https://github.com/yingdi-shan) **Base:** `main` ← **Head:** `hjl/fix-cache-sigbus` --- ### 📝 Commits (1) - [`80e6470`](https://github.com/kvcache-ai/AgentENV/commit/80e647095fd3cb2d19583d824e5aaa4172cd305b) fix(overlaybd): preallocate cache blocks to avoid SIGBUS when disk is full ### 📊 Changes **3 files changed** (+254 additions, -7 deletions) <details> <summary>View changed files</summary> 📝 `storage/overlaybd/src/backend/cache/full_file_cache/cache_entry.rs` (+135 -0) 📝 `storage/overlaybd/src/backend/cache/full_file_cache/cache_store.rs` (+22 -7) 📝 `storage/overlaybd/src/backend/cache/tests.rs` (+97 -0) </details> ### 📄 Description ## What Reserve the block's byte range with fallocate(FALLOC_FL_KEEP_SIZE) before handing out mmap write access (write_block and mmap_mut_for_block). Disk exhaustion now surfaces as a plain ENOSPC refill error, and reads fall back to the source exactly like the existing cache-capacity-full path. Filesystems without fallocate support (EOPNOTSUPP/ENOSYS) keep the previous behavior. ## Why The full-file cache writes refilled blocks through a shared mmap over a sparse data file. Writing into an unallocated hole makes the kernel allocate filesystem blocks at page-fault time; when the local disk is full that allocation fails and the process is killed with SIGBUS, taking down the whole ublk-daemon. ## Design and behavior changes Before write into mmap region of cache files, an extra `fallocate` syscall will be invoked. ## Validation  - [x] `make fmt` - [x] `make clippy` - [x] `make test-unit` - [ ] Relevant Rust integration tests - [ ] `make -C services test` (required when `services/` changes) - [ ] Generated clients/server regenerated with the documented `make` target - [ ] Documentation updated - [ ] Benchmarks or performance comparison completed ## Checklist - [ ] The PR contains one coherent change and no unrelated formatting or refactoring. - [ ] New behavior is covered by tests, or I explained why testing is impractical. - [ ] Logs and examples contain no credentials, tokens, or private registry information. - [ ] I did not manually edit generated code without updating its source and regenerating it. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 14:21:34 +00:00