We replicate sharded, versioned, csv-files on-demand from S3 to Cloudflare edge for code run by workers.dev. Cache invalidation is tricky. Avoid it by ttl-expi

We replicate sharded, versioned, csv-files on-demand from S3 to Cloudflare edge for code run by workers.dev. Cache invalidation is tricky. Avoid it by ttl-expiring caches as frequently as budget allows. We, instead, employ a (probably over-engineered) hand-rolled set-reconciliation algorithm to match our read/write patterns.