Crash-Consistent Checkpointing for AI Training on macOS/APFS

TL;DR

This paper investigates checkpointing protocols for AI training on macOS/APFS, proposing integrity validation methods that detect corruption with high accuracy, and analyzing the performance trade-offs involved.

Contribution

It introduces a format-agnostic integrity guard and compares three checkpoint installation modes, providing insights into reliability and performance trade-offs for AI training.

Findings

Integrity guard detects 99.8-100% corruptions with no false positives.

Performance overhead ranges from 56.5% to 570.6% depending on mode.

Atomic_dirsync mode offers higher durability at increased performance cost.

Abstract

Deep learning training relies on periodic checkpoints to recover from failures, but unsafe checkpoint installation can leave corrupted files on disk. This paper presents an experimental study of checkpoint installation protocols and integrity validation for AI training on macOS/APFS. We implement three write modes with increasing durability guarantees: unsafe (baseline, no fsync), atomic_nodirsync (file-level durability via fsync()), and atomic_dirsync (file + directory durability). We design a format-agnostic integrity guard using SHA-256 checksums with automatic rollback. Through controlled experiments including crash injection (430 unsafe-mode trials) and corruption injection (1,600 atomic-mode trials), we demonstrate that the integrity guard detects 99.8-100% of corruptions with zero false positives. Performance overhead is 56.5-108.4% for atomic_nodirsync and 84.2-570.6% for atomic_dirsync relative to the unsafe baseline. Our findings quantify the reliability-performance trade-offs and provide deployment guidance for production AI infrastructure.