Beyond Static Limits: Adaptive Concurrency with TCP-Vegas in Go
Traditional concurrency limits (like bulkheads) are static. You pick a number—say, 10 concurrent requests— and hope for the best. But in the dynamic world of cloud infrastructure, "10" might be too...
Source: DEV Community
Traditional concurrency limits (like bulkheads) are static. You pick a number—say, 10 concurrent requests— and hope for the best. But in the dynamic world of cloud infrastructure, "10" might be too conservative when the network is fast, or dangerously high when a downstream service starts to queue. Static limits require manual tuning, which is often done after an outage has already happened. To build truly resilient systems, we need Adaptive Concurrency Control. Here is how to implement dynamic concurrency limits in Go using Resile, inspired by the TCP-Vegas congestion control algorithm. The Problem: The "Fixed-Limit" Trap Imagine your service talks to a database. You've set a bulkhead limit of 50 concurrent connections. Scenario A (Normal): Database latency is 10ms. 50 concurrent requests mean you're handling 5,000 RPS. Everything is fine. Scenario B (Degraded): Database latency spikes to 500ms due to a background maintenance task. Your 50 "slots" are now filled with slow requests. Yo
