Merge nucleic/lucid-river-toad-6efj into dev

spikes/WslcSpike/Program.cs

@@ -36,6 +36,10 @@ internal static class Program
         var iterations = int.TryParse(Arg(args, "--iterations"), out var n) ? n : 5;
         var broker = Arg(args, "--broker") ?? FindBroker();
         var skipRecovery = args.Contains("--no-recovery");
+        // narOS runs `naros-init` as PID 1 and stays up (docs/NAROS.md). A stock image usually
+        // does not: alpine's PID 1 is /bin/sh, which exits immediately with no tty, so the
+        // container is `Exited` before the first exec and every later step fails `not_running`.
+        var sleepInit = args.Contains("--sleep-init");
 
         if (broker is null || !File.Exists(broker))
         {
@@ -53,7 +57,7 @@ internal static class Program
 
         try
         {
-            await RunScenarioAsync(broker, sessionName, image, container, repo, iterations);
+            await RunScenarioAsync(broker, sessionName, image, container, repo, iterations, sleepInit);
             if (!skipRecovery) await RunRecoveryAsync(broker, sessionName);
             return 0;
         }
@@ -69,7 +73,7 @@ internal static class Program
 
     private static async Task RunScenarioAsync(
         string brokerPath, string sessionName, string image, string containerName,
-        string repo, int iterations)
+        string repo, int iterations, bool sleepInit)
     {
         await using var broker = BrokerClient.Spawn(brokerPath);
 
@@ -131,19 +135,29 @@ internal static class Program
         Console.WriteLine($"    pulled in {pullWatch.Elapsed.TotalSeconds:F1}s");
 
         Step($"container.create {containerName}");
-        await broker.CallAsync("container.create", new
+        var create = new Dictionary<string, object?>
         {
-            name = containerName,
-            image,
-            volumes = new[] { new { host = repo, guest = "/work", ro = false } },
-            networkingMode = "Bridged",
-            hostname = "nucleic-spike",
-            env = new Dictionary<string, string> { ["NUCLEIC_SPIKE"] = "1" },
-        });
+            ["name"] = containerName,
+            ["image"] = image,
+            ["volumes"] = new[] { new { host = repo, guest = "/work", ro = false } },
+            ["networkingMode"] = "Bridged",
+            ["hostname"] = "nucleic-spike",
+            ["env"] = new Dictionary<string, string> { ["NUCLEIC_SPIKE"] = "1" },
+        };
+        if (sleepInit) create["initArgv"] = new[] { "/bin/sh", "-c", "sleep 3600" };
+        await broker.CallAsync("container.create", create);
         await broker.CallAsync("container.start", new { name = containerName });
         var state = (await broker.CallAsync("container.state", new { name = containerName }))
             .GetProperty("state").GetString();
         Console.WriteLine($"    state: {state}");
+        // Check here rather than letting the first exec fail `not_running`: a container whose
+        // PID 1 exited looks identical to a container that failed to start, and the remedy
+        // (supply a long-running init) is nothing like the remedy for a real start failure.
+        if (state != "running")
+            throw new InvalidOperationException(
+                $"container is '{state}' immediately after start — its PID 1 exited. narOS runs "
+                + "naros-init and stays up; a stock image (alpine's PID 1 is /bin/sh) does not. "
+                + "Re-run with --sleep-init to give it a long-running init process.");
 
         Step("exec: stdio round-trip");
         var (code, output) = await ExecAsync(broker, containerName, ["/bin/sh", "-c", "echo hello from \"$(uname -n)\"; echo to-stderr 1>&2"]);