Cloudflare has altered the way Durable Objects are located when a lookup follows a forwarding record to another data center. The change removes an intermediate proxy step, which matters to engineers because it can lower high‑percentile connection latency and reduce load on the first data center.
What Changed in Durable Object Routing
Previously, when a request arrived at a data center that did not have the object's location cached, that center would first open a connection to the destination data center, proxy the traffic back to the caller, and then the caller would open its own direct connection. The new behavior skips the initial proxy: the first data center simply returns the destination address, and the caller establishes the direct connection itself.
Latency and Resource Implications
By avoiding the extra connection, the round‑trip count drops by one. The source notes that this can reduce p99 connection latency for relocation‑related requests, though the exact benefit depends on workload and traffic patterns. Fewer proxy connections also mean the originating data center spends less time maintaining temporary connections, potentially freeing resources for other traffic.
Operational Considerations
No code changes or configuration updates are required to take advantage of the new routing path. Teams should monitor connection latency metrics, especially the p99 percentile, to quantify any improvement in their specific environment. Because the impact varies with traffic patterns, it is prudent to compare before‑and‑after measurements and adjust alert thresholds if latency baselines shift.
Related CloudNinjas coverage: hands-on guides.
What This Means For Practitioners
Review your Durable Object latency dashboards for any reduction in high‑percentile values and verify that existing monitoring continues to capture connection events correctly. No deployment changes are needed, but updating performance expectations and alerting rules may be worthwhile.
