- Sources: Cloudflare blog post, HN discussion
- Summary: Cloudflare published an account of migrating cdnjs, which serves 9 billion requests a day, off 26 alphabet-sharded Cloud Functions and a 1.1 TB git repository onto its own developer platform. The post names the failure mode of the old design as partial success with no correlation id shared between the two logging systems, so a version could write to the edge, silently fail to reach the source of truth, and serve correctly for weeks with no alert.
- Why it matters: The missing correlation id was the diagnostic gap rather than the remedy, and the remedy the post describes is a re-platform onto Workflows, R2, D1, Queues and Containers with R2 as the single source of truth, which removes the two divergent writes instead of instrumenting them.
send feedback on this story