Yes, tried tunnels too. There is significant variability among individual requests, but when benchmarking at scale I found no meaningful difference in p50 and p

Yes, tried tunnels too. There is significant variability among individual requests, but when benchmarking at scale I found no meaningful difference in p50 and p90 between “Worker -> CF Tunnel -> EC2 -> backend app” and “Worker -> AWS Global Accelerator -> EC2 -> backend app”