Can you patch Ruby faster now?
- Name
- Dennis Paagman
- @djfpaagman
Three weeks ago I checked how fast you could actually patch Ruby. For 4.0.9, the official Docker image took almost 16 hours, and some hosting platforms more than four days.
Last night, Ruby 4.0.10 fixed seven vulnerabilities in Regexp and String case conversion (of course it’s regex). Ruby’s security team confirmed remote code execution was possible for some of them. Exactly the kind of release where every hour counts. Ruby 3.4.13 and 3.3.14 were released at the same time, but I only tracked 4.0.10 here.
The base image for Ruby 4.0.10 in Rails’ generated Dockerfile was available 1 hour and 14 minutes after the source tarball went up. For 4.0.9, it took nearly 16 hours. Ruby now triggers the Dockerfile update automatically, and a maintainer still merges the PR that kicks off publication.
What improved
- The official Docker image went from almost 16 hours to 1 h 14 m. On 28 October, Ruby release manager @k0kubun asked the Docker maintainers how Ruby could help, and then added a workflow to docker-library/ruby that Ruby’s release process triggers. The Dockerfiles were updated after 21 minutes, instead of seven hours. The official-images PR is still merged by hand, but k0kubun marked it
[security]and @tianon merged it within three minutes, without waiting for CI. That’s the image Rails, Kamal and Fly.io use by default, so this is a big win. - Heroku went from 2.5 hours to 6 minutes. @schneems changed Heroku’s check for new Rubies from every two hours to every five minutes. DigitalOcean, Northflank and Scalingo use the same binaries, so they got it just as fast.
- Some of it was people, not automation. ruby-install went from 5.5 days to 17 minutes, thanks to @havenwood. I’m not sure if this was triggered automatically, as 3.4.13 and 3.3.14 haven’t been added yet. mise’s precompiled Rubies (and so Railway) went from 8 hours to about one, because @jdx triggered the build by hand. It still checks for new versions on a schedule, twice a day.
I checked availability on 19 November at 07:45 UTC, about eight hours after the tarball went up.1
What didn’t
- A few got slower. asdf pins a ruby-build version, so it needed a PR to bump it, which took six hours. bitnami took twice as long as last time.
- Two builds were still pending. The rails/devcontainer images are built by a workflow someone has to start by hand, and that hadn’t happened yet. rv’s release job had been queued for over seven hours.
- CircleCI is still waiting. Last time
cimg/rubytook 1.5 hours, but this time the PR has been open for over five hours. - Windows is still waiting. No RubyInstaller release yet, so no winget or setup-ruby on Windows either.
- The managed runtimes haven’t moved. AWS Lambda, Elastic Beanstalk and Google Cloud are still on 4.0.9. Last time they took 3.5 to 4.5 days.
When Ruby 4.0.10 became available
I repeated the timing check for Ruby 4.0.10. The 4.0.10 clock starts when the source tarball became available, on Wednesday 18 November 2026, 23:34 UTC.2
Green rows got a lot faster. “Same” means the timings were within a couple of minutes. ⚠️ flags a long wait, ⏳ marks a release still pending at the time of the check, and ❌ means the version was unavailable through that source. I’ll update this post as the pending releases come in.
Installers and version managers
| Resource | 4.0.9 | 4.0.10 | Change |
|---|---|---|---|
| ruby-install | ⚠️ 5 d 10 h 46 m | 17 m3 | −5 d 10 h 29 m |
| ruby-build (rbenv, and mise when compiling) | 43 m | 42 m | same |
| ruby/ruby-builder binaries | 48 m | 10 m | −38 m |
| jdx/ruby binaries (mise’s precompiled Rubies) | 8 h 15 m | 1 h 07 m | −7 h 08 m |
| Homebrew ruby-build | 1 h 10 m | 1 h 22 m | +12 m |
| asdf | 2 h 43 m | 6 h 11 m | +3 h 28 m |
| RVM | ⚠️ 18 h 34 m | ⏳ PR open | – |
| rv | ⚠️ 21 h 40 m | ⏳ release job queued4 | – |
| RubyInstaller (Windows, also RVM 4 Windows) | ⚠️ 16 h 35 m | ⏳ | – |
| rbenv for Windows | ❌ | ❌ still on 4.0.9 | – |
Container images
| Resource | 4.0.9 | 4.0.10 | Change |
|---|---|---|---|
| rubylang/ruby (by the Ruby core team) | 8 m | 9 m | same |
| ruby (by the Docker community) | ⚠️ 15 h 57 m | 1 h 14 m | −14 h 43 m |
| sorah/ruby (by Ruby committer @sorah) | 2 h 29 m | 1 h 14 m | −1 h 15 m |
| bkuhlmann/alpine-ruby | 8 h 50 m | 3 h 20 m | −5 h 30 m |
| bitnami/ruby | 1 h 59 m | 4 h 15 m | +2 h 16 m |
| rails/devcontainer | 9 h 34 m | ⏳ PR merged, image not built yet5 | – |
CI
| Resource | 4.0.9 | 4.0.10 | Change |
|---|---|---|---|
| GitHub Actions setup-ruby (Linux/macOS) | 1 h 12 m | 45 m | −27 m |
| CircleCI cimg/ruby | 1 h 36 m | ⏳ PR open | – |
| GitHub Actions setup-ruby (Windows) | ⚠️ 28 h 34 m | ⏳ waiting for RubyInstaller | – |
Hosting platforms
| Resource | 4.0.9 | 4.0.10 | Change |
|---|---|---|---|
| Render (native runtime, compiles versions it hasn’t prebuilt) | 0 m | 0 m6 | same |
| Heroku | 2 h 28 m | 6 m | −2 h 22 m |
| DigitalOcean App Platform (uses Heroku’s buildpack) | 2 h 28 m | 6 m | −2 h 22 m |
| Northflank (Heroku builder) | 2 h 28 m | 6 m | −2 h 22 m |
| Scalingo | 2 h 29 m | 6 m | −2 h 23 m |
| Railway (uses mise) | 8 h 15 m | 1 h 07 m | −7 h 08 m |
| Fly.io, Kamal and anything else using the official Docker community image | ⚠️ 15 h 57 m | 1 h 14 m | −14 h 43 m |
| Hatchbox (precompiled Rubies) | Unknown | ≈ 1 h 15 m (build time)7 | – |
| Google Cloud (Cloud Run, App Engine) | ⚠️ ≈ 3 d 14 h | ⏳ | – |
| AWS Elastic Beanstalk | ⚠️ 4 d 4 h 07 m | ⏳ | – |
| AWS Lambda | ⚠️ 4 d 16 h 29 m | ⏳8 | – |
System packages
| Resource | 4.0.9 | 4.0.10 | Change |
|---|---|---|---|
| Snap | 7 m | 11 m | +4 m |
| sorah-rbpkg (APT packages for Ubuntu and Debian) | 2 h 26 m | 1 h 11 m | −1 h 15 m |
| Homebrew | 2 h 41 m | 1 h 24 m | −1 h 17 m |
| nixpkgs | 3 h 47 m | ⏳ PR open | – |
| Fedora | ⚠️ 11 h 03 m | ⏳ | – |
| winget | ⚠️ 20 h 54 m | ⏳ waiting for RubyInstaller | – |
| Debian, Ubuntu, Alpine, Arch | ❌ no Ruby 4 at all | ❌ | – |
Wrapping up
The advice from last time still stands: check the whole path from your laptop to production, automate what you can, and pick a setup where you can update Ruby on your own schedule.
You also don’t have to wait for me anymore to see how fast things are: Speedshop built Ruby Watch. It tracks where new Ruby releases have landed, checking up to once a minute.
Again, most of the people who make this happen are volunteers. Thanks for all the effort that’s being put into automating this and making it faster. 🙏
-
I used the same sources as last time: Git commit dates, GitHub release dates, Docker Hub’s last pushed date, and for Heroku and Scalingo the
Last-Modifiedheader of the Ruby binary (Scalingo serves the same binaries as Heroku, with the same ETag). For Homebrew it’s the bottle commit. For bitnami it’s the commit, as bitnami/ruby no longer publishes version tags on Docker Hub. bkuhlmann/alpine-ruby doesn’t tag by Ruby version either, so I used the push oflatest, which matches the commit updating to 4.0.10. ↩ -
The v4.0.10 tag was created at 22:24 UTC, 1 h 10 m before the tarball appeared on cache.ruby-lang.org. The announcement was committed at 23:52 UTC, 18 minutes after the tarball. ↩
-
The commit in postmodern/ruby-versions is dated 23:04 UTC, half an hour before the tarball appeared, but its checksums match the published tarball. GitHub’s activity log shows it was pushed at 23:52 UTC, so that’s the time I used. ↩
-
The PR adding 4.0.10 was merged 31 minutes after the release, but the release workflow has been queued ever since. ↩
-
An automated PR was opened after an hour and merged after about six hours. The images are built by a workflow that has to be started by hand, and that hadn’t happened yet. ↩
-
Render’s native runtime falls back to compiling a Ruby version it hasn’t prebuilt, as @nateberkopec found, so it only needs the tarball. ↩
-
Hatchbox precompiles Rubies itself. The
rubybinary on my Hatchbox server has a modification time of 00:48:38 UTC, kept from when Hatchbox built it. It was unpacked on my server hours later, when I deployed. Hatchbox’s email saying 4.0.10, 3.4.13 and 3.3.14 were ready arrived at 03:02 UTC. I used asdf’s timing as an estimate in the 4.0.9 post and don’t have a comparable Hatchbox timestamp for that release. ↩ -
The
public.ecr.aws/lambda/ruby:4.0image was rebuilt about 1.5 hours after the release, but it still contains Ruby 4.0.9. ↩
I've got some availability for new clients! If you're looking for someone with 15 years of Rails experience and a deep feel for Hotwire, who's all in on building with AI agents, let's chat.
Get in touch