Subscribe
0%
Illustration for: Firenook swaps Firebase's Java emulators for a Rust engine and claims 40x faster writes. What holds up

Firenook swaps Firebase's Java emulators for a Rust engine and claims 40x faster writes. What holds up

Firenook is an Apache-2.0, drop-in replacement for the Firebase Local Emulator Suite that keeps your firebase.json, SDKs and ports but drops Java. Its headline numbers come from one workload, but an independent re-run reproduced sub-millisecond writes and a sub-second restart.

Listen to this article0:00 / 4:38
firenook-dev/firenook

A faster drop-in replacement for the Firebase Local Emulator Suite, with no Java dependency and data that persists across restarts.

License
Apache-2.0
Try it
npm install --save-dev firenook && npx firenook init
View on GitHub

Firenook is a new open-source replacement for Google's Firebase Local Emulator Suite, the set of local servers developers run to test Firestore, Auth, Storage and Cloud Functions without touching production. Its engine is written in Rust, so it needs no Java. The project's README calls it a "drop-in" alternative: it reads your existing firebase.json, works with the normal Firebase SDKs, listens on the official ports, and imports and exports the official data format. Version 0.2.1 shipped on November 4, and its Show HN post was among the top entries on Hacker News' Show page when we checked.

Why it matters: anyone who has used the official emulators on a large dataset knows the pain. Hacker News commenters complained about slow restarts, heavy memory use and an ageing UI. The author, who posts as sanjevirau, wrote that even beefier machines still ran out of memory with the official suite. A faster local loop changes how often you can reset test data and rerun integration tests.

What it supports, and what it doesn't

According to the README, Firenook serves these emulators on the official suite's ports:

Supported Not emulated (yet)
Firestore, Authentication, Storage Realtime Database
Cloud Functions, Extensions Hosting, App Hosting
Pub/Sub, Cloud Tasks, Eventarc Data Connect

If your firebase.json configures one of the unsupported services, Firenook still starts but prints a warning. On Hacker News, the author said Realtime Database support is planned. Requirements: Node 22, 24 or 26, on macOS, Windows x64, or Linux with glibc 2.28 or newer. Alpine Linux is not supported, which matters if your CI images are Alpine-based.

The README calls 0.2.0 the first stable release. The 0.2.1 notes list small fixes, including popup sign-in when you reach the emulator over a LAN address or container name on plain HTTP, and a new console-style sign-in page with test accounts. The release notes say there were no protocol, port or data-format changes, and that 0.2.0 and 0.2.1 can open each other's saved data. Pre-release notes for 0.2.0 also say the project moved to the Apache License 2.0 alone.

The benchmark claims

Firenook's own numbers, from its 0.2.0 release notes, come from replaying two hours of one application's traffic against both emulators:

Metric (Firenook's figures) Official suite Firenook
Peak emulator memory 22.57 GiB 2.11 GiB
Median write 208 ms 5.1 ms
Restart with data kept 30.8 s 0.59 s
Query shapes faster — 25 of 26 (one tie)

The release notes also claim listener updates about 19x faster with 61% less CPU. These are vendor figures from a single workload, so treat the "40x" and "10x" headlines as one data point, not a guarantee for your app.

An independent re-run

Developer Frank Chu re-ran part of the test on November 6 and published the results on DEV Community. On a 4-vCPU, 8 GB VM with Firenook 0.2.1, using plain REST calls against Firestore, Auth and Storage, he measured:

  • Median write: 0.85 ms (p99 1.77 ms)
  • Median read: 0.34 ms
  • Restart with 2,000 documents kept: 0.59 s
  • Memory after loading 2,000 documents: 114.7 MiB

The caveats are important, and he lists them himself. He did not run the official emulator, because his VM had no Java and too little RAM, so the comparison side still rests on Firenook's published data. His workload was also lighter than Firenook's, with no listeners or triggered functions, so his 0.85 ms isn't directly comparable to the 5.1 ms figure. He found one area where Firenook was slightly slower: storage downloads, which Firenook's own benchmark document also admits. He also found that --keep-data needs all three emulators running, not just Firestore. His view is that the absolute numbers matter more than the ratio. He credits the gap mostly to a native binary that stores data on disk, versus a Java emulator that holds everything in memory and only persists it on export.

Should you try it?

If your team's Firebase tests are slowed by emulator start-up, memory or export times, Firenook is cheap to try. It installs as a dev dependency and uses the same config, so rolling back means removing one package. Be careful if you depend on Realtime Database or Hosting emulation, or if you run CI on Alpine. The author says it took about two months to build, plus a month of internal use, so pin the version and watch the changelog. On Hacker News the author said local development comes first, then self-hosting, before any cloud product.

npm install --save-dev firenook
npx firenook init

Sources

  1. GitHub — firenook-dev/firenook github.com
  2. GitHub — firenook-dev/firenook releases (0.2.0 and 0.2.1 notes) github.com
  3. Hacker News — Show HN: Firenook news.ycombinator.com
  4. DEV Community — 0.85ms writes: I re-ran Firenook's 40x claim on a 4-core VM dev.to

#Firebase#Firestore#Rust#open source#developer tools#emulators#Show HN