fix: Set PooledConnectionLifetime for DNS re-resolution on long-lived connections

[phillipleblanc]

Problem

GrpcChannel.ForAddress creates long-lived HTTP/2 connections via SocketsHttpHandler. By default, PooledConnectionLifetime is infinite — connections are never recycled and DNS is never re-resolved.

For clients connecting to load-balanced endpoints (e.g. AWS ALBs), backend IPs can change while the HTTP/2 connection remains healthy, leaving the client stuck on a stale IP.

Fix

Set PooledConnectionLifetime = TimeSpan.FromMinutes(5) on both SocketsHttpHandler instances (unauthenticated and authenticated TLS paths). This forces periodic connection recycling, which triggers fresh DNS resolution.

This is the recommended .NET approach for handling DNS changes with long-lived HttpClient / gRPC connections.

Related

• fix: Use gRPC DnsNameResolver for periodic DNS re-resolution spice-java#39 — equivalent fix for the Java SDK (uses gRPC DnsNameResolver for periodic re-resolution)
• fix: Configure HTTP/2 keep-alive for faster detection of stale connections spice-rs#71 — equivalent fix for the Rust SDK (HTTP/2 keep-alive for stale connection detection)

[Copilot]

Pull request overview

This PR addresses stale DNS resolution for long-lived gRPC HTTP/2 connections by configuring connection recycling via SocketsHttpHandler.PooledConnectionLifetime.

Changes:

• Set PooledConnectionLifetime = TimeSpan.FromMinutes(5) on the unauthenticated TLS SocketsHttpHandler.
• Set PooledConnectionLifetime = TimeSpan.FromMinutes(5) on the authenticated TLS SocketsHttpHandler.

Comments suppressed due to low confidence (1)

Spice/src/Flight/SpiceFlightClient.cs:94

• PooledConnectionLifetime is only set when useTls is true (the SocketsHttpHandler path). When useTls is false the code falls back to HttpClientHandler, so the channel will still use the default (infinite) connection lifetime and won’t re-resolve DNS over time. If the DNS-staleness issue can also affect non-TLS/h2c usage, consider using a SocketsHttpHandler in the non-TLS branch as well so the lifetime can be configured consistently.

        if (useTls)
        {
            messageHandler = new SocketsHttpHandler
            {
                EnableMultipleHttp2Connections = true,
                // Force periodic connection recycling to trigger DNS re-resolution.
                // Without this, HTTP/2 connections are kept alive indefinitely and
                // the client can get stuck on stale IPs when backend targets change
                // (e.g. AWS ALB target rotation).
                PooledConnectionLifetime = TimeSpan.FromMinutes(5),
            };
        }
        else
#endif
        {
            messageHandler = new HttpClientHandler();
        }

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.