fix(host): h2 keepalive on site gRPC listener (arch-review 2026-08-19 P4-3)

2026-08-19 14:49:37 -04:00
parent e63ffd90ee
commit b68f553f1e
3 changed files with 125 additions and 14 deletions
@@ -0,0 +1,56 @@
+using Microsoft.AspNetCore.Server.Kestrel.Core;
+using ZB.MOM.WW.ScadaBridge.Host;
+
+namespace ZB.MOM.WW.ScadaBridge.Host.Tests;
+
+/// <summary>
+/// The site node's Kestrel configuration, asserted against the code that actually ships (the other
+/// Host startup tests re-declare their own <c>ConfigureKestrel</c> lambda, which can never catch a
+/// regression in <c>Program.cs</c>'s).
+/// </summary>
+public class SiteKestrelSetupTests
+{
+    [Fact]
+    public void SiteListener_EnablesHttp2Keepalive()
+    {
+        // Arch-review 2026-08-19 P4-3. Kestrel's default KeepAlivePingDelay is TimeSpan.MaxValue —
+        // no server-initiated pings — so a half-open HTTP/2 connection is never detected on a
+        // stream that is legitimately silent. The LocalDb sync stream is exactly that: with the
+        // dialling node crashed or partitioned, this node's handler stays parked on a dead stream
+        // holding the single-stream guard, refusing the reconnect that would heal the pair.
+        var options = new KestrelServerOptions();
+
+        SiteKestrelSetup.Configure(options, grpcPort: 8083, metricsPort: 8084);
+
+        Assert.NotEqual(TimeSpan.MaxValue, options.Limits.Http2.KeepAlivePingDelay);
+        Assert.Equal(SiteKestrelSetup.Http2KeepAlivePingDelay, options.Limits.Http2.KeepAlivePingDelay);
+        Assert.Equal(SiteKestrelSetup.Http2KeepAlivePingTimeout, options.Limits.Http2.KeepAlivePingTimeout);
+        // Detection has to be quick enough to matter, and a ping must not be its own timeout.
+        Assert.True(SiteKestrelSetup.Http2KeepAlivePingTimeout < SiteKestrelSetup.Http2KeepAlivePingDelay);
+        Assert.True(
+            SiteKestrelSetup.Http2KeepAlivePingDelay + SiteKestrelSetup.Http2KeepAlivePingTimeout
+                <= TimeSpan.FromMinutes(2));
+    }
+
+    [Fact]
+    public void SiteListener_KeepsGrpcHttp2Only_AndMetricsHttp1Capable()
+    {
+        // The keepalive change must not disturb the protocol split: h2c-only on the gRPC port so
+        // Grpc.AspNetCore can serve it without TLS ALPN, HTTP/1.1-capable on the metrics port so a
+        // standard Prometheus scraper can reach /metrics.
+        var options = new KestrelServerOptions();
+
+        SiteKestrelSetup.Configure(options, grpcPort: 8083, metricsPort: 8084);
+
+        var endpoints = options.GetType()
+            .GetProperty("CodeBackedListenOptions", System.Reflection.BindingFlags.Instance
+                | System.Reflection.BindingFlags.NonPublic | System.Reflection.BindingFlags.Public)!
+            .GetValue(options) as System.Collections.IEnumerable;
+        Assert.NotNull(endpoints);
+
+        var protocols = endpoints!.Cast<ListenOptions>().Select(o => o.Protocols).ToList();
+        Assert.Equal(2, protocols.Count);
+        Assert.Equal(HttpProtocols.Http2, protocols[0]);
+        Assert.Equal(HttpProtocols.Http1AndHttp2, protocols[1]);
+    }
+}