Do not fail service containers without the healthcheck

This reverts commit fec24e8341.

src/Runner.Worker/Container/ContainerInfo.cs

@@ -92,6 +92,8 @@ namespace GitHub.Runner.Worker.Container
         public bool IsJobContainer { get; set; }
         public bool IsAlpine { get; set; }
 
+        public bool FailedInitialization { get; set; }
+
         public IDictionary<string, string> ContainerEnvironmentVariables
         {
             get

src/Runner.Worker/ContainerOperationProvider.cs

@@ -98,12 +98,41 @@ namespace GitHub.Runner.Worker
                 await StartContainerAsync(executionContext, container);
             }
 
+            await RunContainersHealthcheck(executionContext, containers);
+        }
+
+        public async Task RunContainersHealthcheck(IExecutionContext executionContext, List<ContainerInfo> containers)
+        {
             executionContext.Output("##[group]Waiting for all services to be ready");
+
+            var unhealthyContainers = new List<ContainerInfo>();
             foreach (var container in containers.Where(c => !c.IsJobContainer))
             {
-                await ContainerHealthcheck(executionContext, container);
+                var healthcheck = await ContainerHealthcheck(executionContext, container);
+
+                if (!(string.Equals(healthcheck, "healthy", StringComparison.OrdinalIgnoreCase) || string.IsNullOrEmpty(healthcheck)))
+                {
+                    unhealthyContainers.Add(container);
+                }
+                else
+                {
+                    executionContext.Output($"{container.ContainerNetworkAlias} service is healthy.");
+                }
             }
             executionContext.Output("##[endgroup]");
+
+            if (unhealthyContainers.Count > 0)
+            {
+                foreach (var container in unhealthyContainers)
+                {
+                    executionContext.Output($"##[group]Service container {container.ContainerNetworkAlias} failed.");
+                    await _dockerManager.DockerLogs(context: executionContext, containerId: container.ContainerId);
+                    executionContext.Error($"Failed to initialize container {container.ContainerImage}");
+                    container.FailedInitialization = true;
+                    executionContext.Output("##[endgroup]");
+                }
+                throw new InvalidOperationException("One or more containers failed to start.");
+            }
         }
 
         public async Task StopContainersAsync(IExecutionContext executionContext, object data)
@@ -299,9 +328,8 @@ namespace GitHub.Runner.Worker
 
             if (!string.IsNullOrEmpty(container.ContainerId))
             {
-                if (!container.IsJobContainer)
+                if (!container.IsJobContainer && !container.FailedInitialization)
                 {
-                    // Print logs for service container jobs (not the "action" job itself b/c that's already logged).
                     executionContext.Output($"Print service container logs: {container.ContainerDisplayName}");
 
                     int logsExitCode = await _dockerManager.DockerLogs(executionContext, container.ContainerId);
@@ -395,14 +423,14 @@ namespace GitHub.Runner.Worker
             }
         }
 
-        private async Task ContainerHealthcheck(IExecutionContext executionContext, ContainerInfo container)
+        private async Task<string> ContainerHealthcheck(IExecutionContext executionContext, ContainerInfo container)
         {
             string healthCheck = "--format=\"{{if .Config.Healthcheck}}{{print .State.Health.Status}}{{end}}\"";
             string serviceHealth = (await _dockerManager.DockerInspect(context: executionContext, dockerObject: container.ContainerId, options: healthCheck)).FirstOrDefault();
             if (string.IsNullOrEmpty(serviceHealth))
             {
                 // Container has no HEALTHCHECK
-                return;
+                return String.Empty;
             }
             var retryCount = 0;
             while (string.Equals(serviceHealth, "starting", StringComparison.OrdinalIgnoreCase))
@@ -413,14 +441,7 @@ namespace GitHub.Runner.Worker
                 serviceHealth = (await _dockerManager.DockerInspect(context: executionContext, dockerObject: container.ContainerId, options: healthCheck)).FirstOrDefault();
                 retryCount++;
             }
-            if (string.Equals(serviceHealth, "healthy", StringComparison.OrdinalIgnoreCase))
+            return serviceHealth;
-            {
-                executionContext.Output($"{container.ContainerNetworkAlias} service is healthy.");
-            }
-            else
-            {
-                throw new InvalidOperationException($"Failed to initialize, {container.ContainerNetworkAlias} service is {serviceHealth}.");
-            }
         }
 
         private async Task<string> ContainerRegistryLogin(IExecutionContext executionContext, ContainerInfo container)

src/Test/L0/Worker/ContainerOperationProviderL0.cs

@@ -0,0 +1,126 @@
+using GitHub.Runner.Worker;
+using GitHub.Runner.Worker.Container;
+using Xunit;
+using Moq;
+using GitHub.Runner.Worker.Container.ContainerHooks;
+using System.Threading.Tasks;
+using System.Collections.Generic;
+using System.Runtime.CompilerServices;
+using GitHub.DistributedTask.WebApi;
+using System;
+
+namespace GitHub.Runner.Common.Tests.Worker
+{
+
+    public sealed class ContainerOperationProviderL0
+    {
+
+        private TestHostContext _hc;
+        private Mock<IExecutionContext> _ec;
+        private Mock<IDockerCommandManager> _dockerManager;
+        private Mock<IContainerHookManager> _containerHookManager;
+        private ContainerOperationProvider containerOperationProvider;
+        private Mock<IJobServerQueue> serverQueue;
+        private Mock<IPagingLogger> pagingLogger;
+        private List<string> healthyDockerStatus = new List<string> { "healthy" };
+        private List<string> emptyDockerStatus = new List<string> { string.Empty };
+        private List<string> unhealthyDockerStatus = new List<string> { "unhealthy" };
+        private List<string> dockerLogs = new List<string> { "log1", "log2", "log3" };
+
+        List<ContainerInfo> containers = new List<ContainerInfo>();
+
+        [Fact]
+        [Trait("Level", "L0")]
+        [Trait("Category", "Worker")]
+        public async void RunServiceContainersHealthcheck_UnhealthyServiceContainer_AssertFailedTask()
+        {
+            //Arrange
+            Setup();
+            _dockerManager.Setup(x => x.DockerInspect(_ec.Object, It.IsAny<string>(), It.IsAny<string>())).Returns(Task.FromResult(unhealthyDockerStatus));
+
+            //Act
+            try
+            {
+                await containerOperationProvider.RunContainersHealthcheck(_ec.Object, containers);
+            }
+            catch (InvalidOperationException)
+            {
+
+                //Assert
+                Assert.Equal(TaskResult.Failed, _ec.Object.Result ?? TaskResult.Failed);
+            }
+        }
+
+        [Fact]
+        [Trait("Level", "L0")]
+        [Trait("Category", "Worker")]
+        public async void RunServiceContainersHealthcheck_UnhealthyServiceContainer_AssertExceptionThrown()
+        {
+            //Arrange
+            Setup();
+            _dockerManager.Setup(x => x.DockerInspect(_ec.Object, It.IsAny<string>(), It.IsAny<string>())).Returns(Task.FromResult(unhealthyDockerStatus));
+
+            //Act and Assert
+            await Assert.ThrowsAsync<InvalidOperationException>(() => containerOperationProvider.RunContainersHealthcheck(_ec.Object, containers));
+
+        }
+
+        [Fact]
+        [Trait("Level", "L0")]
+        [Trait("Category", "Worker")]
+        public async void RunServiceContainersHealthcheck_healthyServiceContainer_AssertSucceededTask()
+        {
+            //Arrange
+            Setup();
+            _dockerManager.Setup(x => x.DockerInspect(_ec.Object, It.IsAny<string>(), It.IsAny<string>())).Returns(Task.FromResult(healthyDockerStatus));
+
+            //Act
+            await containerOperationProvider.RunContainersHealthcheck(_ec.Object, containers);
+
+            //Assert
+            Assert.Equal(TaskResult.Succeeded, _ec.Object.Result ?? TaskResult.Succeeded);
+
+        }
+
+        [Fact]
+        [Trait("Level", "L0")]
+        [Trait("Category", "Worker")]
+        public async void RunServiceContainersHealthcheck_healthyServiceContainerWithoutHealthcheck_AssertSucceededTask()
+        {
+            //Arrange
+            Setup();
+            _dockerManager.Setup(x => x.DockerInspect(_ec.Object, It.IsAny<string>(), It.IsAny<string>())).Returns(Task.FromResult(emptyDockerStatus));
+
+            //Act
+            await containerOperationProvider.RunContainersHealthcheck(_ec.Object, containers);
+
+            //Assert
+            Assert.Equal(TaskResult.Succeeded, _ec.Object.Result ?? TaskResult.Succeeded);
+
+        }
+
+        private void Setup([CallerMemberName] string testName = "")
+        {
+            containers.Add(new ContainerInfo() { ContainerImage = "ubuntu:16.04" });
+            _hc = new TestHostContext(this, testName);
+            _ec = new Mock<IExecutionContext>();
+            serverQueue = new Mock<IJobServerQueue>();
+            pagingLogger = new Mock<IPagingLogger>();
+
+            _dockerManager = new Mock<IDockerCommandManager>();
+            _containerHookManager = new Mock<IContainerHookManager>();
+            containerOperationProvider = new ContainerOperationProvider();
+
+            _hc.SetSingleton<IDockerCommandManager>(_dockerManager.Object);
+            _hc.SetSingleton<IJobServerQueue>(serverQueue.Object);
+            _hc.SetSingleton<IPagingLogger>(pagingLogger.Object);
+
+            _hc.SetSingleton<IDockerCommandManager>(_dockerManager.Object);
+            _hc.SetSingleton<IContainerHookManager>(_containerHookManager.Object);
+
+            _ec.Setup(x => x.Global).Returns(new GlobalContext());
+
+            containerOperationProvider.Initialize(_hc);
+        }
+    }
+}