XExenoscopic·Dec 7, 2022Submission
Show HN: Improve Docker Desktop Performance with Synchronized Filesystem Caches
Hey HN, I wanted to share a Docker Desktop extension that uses Mutagen (the open-source[0] file sync tool for developers) to improve bind mount performance in Docker Desktop. It allows you to create synchronized filesystem caches inside the Docker Desktop VM that can automatically replace bind mounts. This gives you ext4 filesystem performance inside the Docker Desktop VM, with low-latency synchronization to-and-from the host filesystem.
Docker themselves actually shipped this functionality in Docker Desktop back in 2020, but they decided to pivot back to virtual filesystems with gRPC FUSE and Virtiofs. While these work fairly well, there are still substantial gains to be had for workflows that are readdir(), stat(), read(), and write()-heavy. This includes things like package installs (e.g. npm and Composer), dynamic language runtimes (e
Read the rest
.g. PHP or Node.js), and compiling code.
This new implementation is significantly more performant[1], offers a more detailed UI, and gives you the option of setting the user and group IDs for files sync'd into the VM. You can even create multiple caches of the same files with different UIDs/GIDs, allowing containers with different UIDs/GIDs to access the same files without permissions conflicts.
This extension is closed-source and requires a subscription for some functionality, but that money helps to support the corresponding open-source project.
I'd be keen to hear your feedback. There are a few minor limitations (mostly SDK limitations), but nothing too significant. The next step is probably going to be adding support for remote Docker engines, but I'd be interested to know if there are other pressing features that people would like to see.
Disclaimer: I am a Docker Captain, though this tool is not developed, sponsored, or endorsed by Docker, Inc.
---
[0]: https://github.com/mutagen-io/mutagen
[1]: The Docker Desktop EULA prevents me from publishing benchmarks, but the performance difference will be the same as the difference between gRPC FUSE or Virtiofs and a "native" ext4 volume inside the VM. This will differ between hardware and virtualization frameworks. The best option is testing your own workflow.