prototype support for step retries

src/Runner.Worker/ActionRunner.cs

@@ -64,6 +64,8 @@ namespace GitHub.Runner.Worker
 
         public Pipelines.ActionStep Action { get; set; }
 
+        public Int32 Retries => Action?.Retries ?? 0;
+
         public TemplateToken Timeout => Action?.TimeoutInMinutes;
 
         public async Task RunAsync()

src/Runner.Worker/JobExtensionRunner.cs

@@ -26,6 +26,7 @@ namespace GitHub.Runner.Worker
         public TemplateToken ContinueOnError => new BooleanToken(null, null, null, false);
         public string DisplayName { get; set; }
         public IExecutionContext ExecutionContext { get; set; }
+        public Int32 Retries => 0;
         public TemplateToken Timeout => new NumberToken(null, null, null, 0);
         public object Data => _data;
 

src/Runner.Worker/StepsRunner.cs

@@ -24,6 +24,7 @@ namespace GitHub.Runner.Worker
         TemplateToken ContinueOnError { get; }
         string DisplayName { get; set; }
         IExecutionContext ExecutionContext { get; set; }
+        Int32 Retries { get; }
         TemplateToken Timeout { get; }
         Task RunAsync();
     }
@@ -280,6 +281,10 @@ namespace GitHub.Runner.Worker
                 step.ExecutionContext.Error("An error occurred when attempting to determine the step timeout.");
                 step.ExecutionContext.Error(ex);
             }
+
+            int attempt = 1;
+            while (true)
+            {
                 if (timeoutMinutes > 0)
                 {
                     var timeout = TimeSpan.FromMinutes(timeoutMinutes);
@@ -348,6 +353,25 @@ namespace GitHub.Runner.Worker
                 }
                 Trace.Info($"Step result: {step.ExecutionContext.Result}");
 
+                if (step.ExecutionContext.Result == TaskResult.Failed && attempt <= step.Retries)
+                {
+                    attempt++;
+                    step.ExecutionContext.Result = null;
+                    step.ExecutionContext.ResultCode = null;
+                    // todo: replace the step cancellation token source
+                    // todo: reset the step.ExecutionContext.CommandResult
+                    // todo: create a new timeline record, e.g. "My display name (#2)"
+                    // todo: clear outputs? What will we do on a job? probably clear outputs since merging from separate timeline attempts would otherwise be complex
+                    // todo: consider intrastate - i guess it makes sense this doesn't get cleared
+                    // todo: reconcile all of the above wrt composite steps
+                    // todo: reconcile all of the above wrt pre/post
+                    // todo: distinguish retryable vs non-retryable failures? e.g. if an exception bubbles from the handler
+                    continue;
+                }
+
+                break;
+            }
+
             // Complete the step context.
             step.ExecutionContext.Debug($"Finishing: {step.DisplayName}");
         }

src/Sdk/DTPipelines/Pipelines/ActionStep.cs

@@ -22,6 +22,7 @@ namespace GitHub.DistributedTask.Pipelines
             this.Reference = actionToClone.Reference?.Clone();
 
             Environment = actionToClone.Environment?.Clone();
+            Retries = actionToClone.Retries;
             Inputs = actionToClone.Inputs?.Clone();
             ContextName = actionToClone?.ContextName;
             DisplayNameToken = actionToClone.DisplayNameToken?.Clone();
@@ -46,6 +47,9 @@ namespace GitHub.DistributedTask.Pipelines
         [DataMember(EmitDefaultValue = false)]
         public TemplateToken Environment { get; set; }
 
+        [DataMember(EmitDefaultValue = false)]
+        public Int32 Retries { get; set; }
+
         [DataMember(EmitDefaultValue = false)]
         public TemplateToken Inputs { get; set; }