Don't Cancel Main Branch CI Runs: A Tradeoff Between Traceability and Cost

Change summary

The MCP Servers project updated its CI policy to not cancel in-progress CI runs on main. Per-ref cancellation remains for superseded branch pushes, but merges to main now each complete their own CI pipeline, preserving a one-to-one mapping between commit and CI result.

Why this is significant

This adjustment emphasizes forensics and auditability: when failures occur in production-related commits, having an immutable CI record tied to the merge commit simplifies root-cause analysis and blame attribution. It's a governance-forward stance that values traceability and post-merge accountability, useful for regulated environments or teams that need clear audit trails.

Business and engineering impacts

The tradeoff is increased CI consumption and potentially longer queue times. Organizations should expect higher compute costs and plan for capacity accordingly. Conversely, teams benefit from clearer diagnostics and historical records that make incident retrospectives more precise and less ambiguous.

Recommendations for leaders

- Decide where you sit on the traceability vs. cost spectrum: adopt per-commit retention for main if audit and incident response are priorities.
- Mitigate cost by optimizing pipelines (incremental builds, caching, test splitting) and introducing run retention policies (e.g., expire old runs after X days).
- Combine with flakiness-reduction investments (test reliability, gating) so CI runs for main remain meaningful and actionable.

This simple policy change reveals how CI policies encode organizational values - transparency and accountability versus efficiency - and leaders should align CI behavior with risk and compliance priorities.