Repository navigation
ci: retry Go module downloads after every module cache restore - #8349
Conversation
A cache miss, for example on a pull request that changes go.sum, fetches modules fresh, and one dropped Go proxy stream failed the whole job. The restore-go-modules action now runs the shared bounded download retry from its own pinned checkout, the same way setup-desktop-build does, and the retry script is bound to the audited action content. Part of #8053 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pin all restore-go-modules consumers to the commit that adds the bounded download, and restore modules in the workflow-contract job, whose tests resolved modules with no cache and no retry. Part of #8053 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
✅MegaLinter analysis: Success✅ Linters with no issuesactionlint, bash-exec, git_diff, hadolint, jscpd, jsonlint, lychee, markdown-table-formatter, markdownlint, prettier, prettier, shellcheck, shfmt, stylelint, syft, trivy-sbom, trufflehog, v8r, v8r, yamllint See detailed reports in MegaLinter artifacts
|
Code coverage overviewLanguages: Go Go / code-coverage/go
View coverage of the 3 most impacted files.
ℹ️ Calculations represent line coverage. Results may change as calculations are updated. |
@coderabbitai review |
Action not completed
|
|
No actionable comments were generated in the recent review.
|
@coderabbitai review |
Action performed
|

Why
When a pull request changes the Go dependency list, CI downloads every Go module fresh, and a single dropped connection to the Go module mirror fails the job. Each such failure costs a dependency rebase and another full CI cycle of about 70 minutes, even though nothing in the pull request was wrong.
What
Every CI job that restores the shared Go module cache now downloads any missing modules with the same short, bounded retry the release builds already use, so a brief mirror outage no longer turns a correct pull request red. A genuinely missing or tampered module still fails the job, and every attempt's error stays visible in the log. The workflow-contract check, which previously downloaded modules with no cache and no retry, now uses the shared cache too.
Part of #8053