You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
The CLI's install fetches bubblewrap from more than one mirror, so one unreachable mirror does not fail a Run #774
Why.#748 moved the install's source list to one mirror, mirrors.edge.kernel.org, over HTTPS (record 0195). On 2026-11-12 from 01:47Z, runners intermittently could not connect to it at all: Cannot initiate the connection to mirrors.edge.kernel.org:443 (2605:f480:58:1:0:1994:3:14). - connect (101: Network is unreachable) Could not connect to mirrors.edge.kernel.org:443 (142.0.200.124)
That was followed by E: Unable to locate package bubblewrap. It failed two of #771's review Runs (sf-run 43136883440 and the one before it) and the factory/v0.1.9 eval dispatch 43140773688. A single mirror is a single point of failure for every Run and every eval.
Done when
The source list in the CLI install block of .github/workflows/run-jobs.yml and .github/workflows/eval.yml names at least two mirrors from different operators, each over HTTPS and each signed-by Ubuntu's archive keyring. One is the runner image's own mirror (azure.archive.ubuntu.com; ubuntu.com resolves again since about 22:00Z on 2026-11-11), and one is kernel.org. apt reads both; if one is unreachable, update warns and install takes the package from the other.
tools/check_workflows.py's CLI_INSTALL_RUN pins the new block word for word, with the signed-by check on every deb line.
The record amends 0195, saying why two operators and not one, and what still fails only if both are down.
Activity