Latency spikes under concurrent requests on DigitalOcean Load Balancer
0 reputation · 12 Nov 2024, 22:36 UTC
Goal
Determine whether latency that appears only under concurrent requests is introduced by the DigitalOcean Load Balancer or by backend Droplet resource saturation, given that DigitalOcean does not publish a concurrent-connection threshold for Load Balancer degradation.
Load Balancers distribute TCP and HTTP traffic across Droplets and support health checks, sticky sessions, and TLS termination. Managed Databases expose a configurable max_connections limit that can queue or refuse connections under concurrency. DigitalOcean monitoring reports aggregate utilization but does not isolate load-balancer-induced latency from backend latency, and behavior varies by Droplet size and region.
Because no specific threshold is documented, the decision point is whether to tune the Load Balancer idle timeout and connection draining, scale backend Droplets, or adjust Managed Database max_connections for a concurrent-request workload.
What observable behavior distinguishes Load Balancer-induced concurrency latency from backend Droplet saturation? Does DigitalOcean document a concurrent-connection limit or latency threshold for Load Balancers? Should idle timeout and connection draining be adjusted before scaling backend capacity for concurrency-dependent latency?
