Eatmydata.ai

Disk writes that wait for confirmation are a correctness feature — one that costs real time when you are running a test suite or provisioning a container that will be destroyed in sixty seconds. eatmydata is an LD_PRELOAD utility: you prefix your command with it, and it intercepts fsync and related syscalls at the library level, returning success immediately instead of waiting for the kernel to flush to disk. No kernel patches, no application changes, no configuration files. The workflow is a single command prefix.

The differentiating feature is scope control through LD_PRELOAD. Because it operates at the process level, you apply it precisely — one apt-get, one test runner, one database seed script — without affecting the rest of the system. Debian packages it directly, so installation in a CI image is a single apt install line with no build step required.

eatmydata fits CI pipelines, ephemeral chroot setups, and any test environment where the disk state is thrown away at the end of the run. It does not fit anywhere else. Use it on a production database and a power loss or OOM event will corrupt data silently — the syscalls it suppresses are the ones that would have saved you. Teams that need this kind of acceleration in production, or need it across a whole system rather than a single process, are solving a different problem with different tools.