The CLI's install fetches bubblewrap by a source list of its own from a mirror the runners resolve, not the image's mirror list (#743)

Merged
DO
6d ago
+85-17

User description

Fixes #743, "A Run's CLI install fails whole when the runner's Ubuntu mirror cannot be resolved".

Protected paths: .github/workflows/run-jobs.yml and .github/workflows/eval.yml. Pre-approved by the Maintainer 2026-11-08, ruled by the Factory Builder 2026-11-11. Neither is a model Lens input. The Factory cannot approve this PR, because its Runs fail on this very fault, so it waits for the Maintainer's hand approval. Not to be enqueued by the lane.

What was wrong

Every review Run since 13:56Z timed out in "Install the CLI at the Engine's pin" (run 43046213251). The image's apt reads /etc/apt/apt-mirrors.txt, which tries azure.archive.ubuntu.com first for each file, and each lookup stalled 20 s.

It is wider than the Azure mirror. Canary run 43047814971, with the block pointed at archive.ubuntu.com and security.ubuntu.com, failed too:

W: Failed to fetch http://archive.ubuntu.com/ubuntu/dists/noble/InRelease  Temporary failure resolving 'archive.ubuntu.com'
E: Unable to locate package bubblewrap

Canary run 43048275193 then resolved each host on the runner:

  • None of azure.archive.ubuntu.com, archive.ubuntu.com, security.ubuntu.com, ubuntu.com, launchpad.net or ppa.launchpadcontent.net resolved.
  • mirrors.edge.kernel.org, mirror.math.princeton.edu, packages.microsoft.com and github.com did.

The fix

CLI_INSTALL_RUN, which holds both workflows word for word (and the canary's install, harness/eval/canary.py), now does three things:

  1. It writes /tmp/ubuntu.list: https://mirrors.edge.kernel.org/ubuntu for noble, noble-updates and noble-security, main alone, each [signed-by=/usr/share/keyrings/ubuntu-archive-keyring.gpg].
  2. It runs apt-get update and apt-get install both with -o Dir::Etc::sourcelist=/tmp/ubuntu.list -o Dir::Etc::sourceparts=-. The image's mirror list and its ubuntu.sources are never read.
  3. It sets -o Acquire::Retries=2 -o Acquire::https::Timeout=10 on both apt lines, so a dropped connection fails inside the 3-minute step. A name that never resolves is not bounded by apt, so the step's own timeout still ends that case.

There is no branching in the step. Ubuntu's signature on the index still holds the package, as decision 0106 has it, and the transport holds nothing, as before. On canary run 43101071038 this exact block installed bubblewrap 0.9.0-1ubuntu0.3 over HTTPS in about 3 s.

The trade is a third-party Ubuntu mirror (kernel.org's) in place of the image's Azure one. The keyring holds integrity; availability is the mirror's.

A test, Workflows::test_the_cli_install_reads_only_signed_sources_over_https_by_its_own_list, holds the block to what it may say: exactly ten lines; three deb lines each fully matching a signed-by, https, Ubuntu-suite pattern; two apt lines each with exactly its options. Eleven mutants under "Issue 743", all killed. The two 0106 mutants are re-pointed at the new lines.

Owed:

  • The record amending decision 0106's rows "The fix", "The exception to 'nothing is installed'" (which says "from the runner's own Azure mirror") and the row naming the archive being down. Its number has been asked of the Factory Builder; it lands in a follow-on PR.

Verify

make checks passes. The named tests pass and 6 of 6 mutants are killed. Full verify is CI's. The first real Run after merge and a cut confirms the step.

Reviews

ReadHeadState
Claude + security603028caNothing blocking. Its caveat (unproven on a runner) answered by canary runs 43047814971 and 43048275193, and the host moved off ubuntu.com in 01a234d9. Mutant name and order fixed. The record and the structural check are owed, above
Carbon (gemini-4-carbon-high)603028caA clean read. The host change after it (01a234d9) is a one-word swap in the same lines
Second round (the bots' 10 threads: SonarCloud clear-text, Gemini timeouts, CodeAnt's one-line mutant)672d497aHTTPS, timeouts on both lines, and the allow-list test with a mutant for each line; each thread answered and resolved
Carbon, round 2d5f7a58eA clean read
Claude + security, round 2d5f7a58eNothing blocking. The test turned from banned words into an allow-list, the install's retries capped, the mutants named for the test that kills them: fixed in 672d497a
Muse (Wonder Full)01a234d9One blocking: a 0106 mutant still named the old update line, so the merge group's shard would fail on a stale mutant. Fixed in the head after it; a sweep finds no stale mutant
Astra (gpt-6-astra)01a234d9A clean read

🤖 Generated with Claude Code


CodeAnt-AI Description

Prevent CLI setup failures when runner Ubuntu mirrors are unreachable

What Changed

  • Run and evaluation jobs now install bubblewrap from a dedicated Ubuntu mirror instead of the runner’s configured mirrors.
  • Package downloads use HTTPS, Ubuntu’s signing key, and bounded retries and connection waits.

Impact

✅ Fewer CLI setup timeouts
✅ Run jobs can install bubblewrap when runner-configured mirrors are unreachable
✅ Evaluation jobs use the same reliable installation

💡 Usage Guide

Checking Your Pull Request

Every time you make a pull request, our system automatically looks through it. We check for security issues, mistakes in how you're setting up your infrastructure, and common code problems. We do this to make sure your changes are solid and won't cause any trouble later.

Talking to CodeAnt AI

Got a question or need a hand with something in your pull request? You can easily get in touch with CodeAnt AI right here. Just type the following in a comment on your pull request, and replace "Your question here" with whatever you want to ask:

@codeant-ai ask: Your question here

This lets you have a chat with CodeAnt AI about your pull request, making it easier to understand and improve your code.

Example

@codeant-ai ask: Can you suggest a safer alternative to storing this secret?

Preserve Org Learnings with CodeAnt

You can record team preferences so CodeAnt AI applies them in future reviews. Reply directly to the specific CodeAnt AI suggestion (in the same thread) and replace "Your feedback here" with your input:

@codeant-ai: Your feedback here

This helps CodeAnt AI learn and adapt to your team's coding style and standards.

Example

@codeant-ai: Do not flag unused imports.

Retrigger review

Ask CodeAnt AI to review the PR again, by typing:

@codeant-ai: review

Check Your Repository Health

To analyze the health of your code repository, visit our dashboard at https://app.codeant.ai. This tool helps you identify potential issues and areas for improvement in your codebase, ensuring your repository maintains high standards of code health.