Fail fast strategy should properly mark ppl as failed when one of the jobs fail (#507)

Author: skipi

zebra/lib/zebra/workers/task_finisher.ex

@@ -61,7 +61,7 @@ defmodule Zebra.Workers.TaskFinisher do
       log(task.id, "Processing task, results #{inspect(results)}, states #{inspect(states)}")
 
       if Enum.all?(states, fn s -> s == Job.state_finished() end) do
-        case calculate_task_result(results) do
+        case calculate_task_result(results, task) do
           {:ok, result} ->
             finish(task, jobs, result)
             publish(task)
@@ -92,8 +92,13 @@ defmodule Zebra.Workers.TaskFinisher do
     log(task.id, "marked as finished")
   end
 
-  def calculate_task_result(job_results) do
+  def calculate_task_result(job_results, task) do
     cond do
+      # If fail_fast:stop is active and there's a failure, show as failed
+      task.fail_fast_strategy == "stop" &&
+          Enum.any?(job_results, fn r -> r == Job.result_failed() end) ->
+        {:ok, "failed"}
+
       Enum.any?(job_results, fn r -> r == Job.result_stopped() end) ->
         {:ok, "stopped"}
 

zebra/test/zebra/workers/task_finisher_test.exs

@@ -60,20 +60,45 @@ defmodule Zebra.Workers.TaskFinisherTest do
   describe ".calculate_task_result" do
     test "when any of the jobs are stopped => result is stopped" do
       results = ["passed", "failed", "stopped"]
+      {:ok, task} = Support.Factories.Task.create()
 
-      assert {:ok, "stopped"} = W.calculate_task_result(results)
+      assert {:ok, "stopped"} = W.calculate_task_result(results, task)
     end
 
     test "when any of the jobs are failed => result is failed" do
       results = ["passed", "failed", "passed"]
+      {:ok, task} = Support.Factories.Task.create()
 
-      assert {:ok, "failed"} = W.calculate_task_result(results)
+      assert {:ok, "failed"} = W.calculate_task_result(results, task)
     end
 
     test "when all of the jobs are passed => result is passed" do
-      results = ["passed", "failed", "passed"]
+      results = ["passed", "passed", "passed"]
+      {:ok, task} = Support.Factories.Task.create()
+
+      assert {:ok, "passed"} = W.calculate_task_result(results, task)
+    end
+
+    test "when fail_fast:stop is active and jobs are stopped due to failure => result is failed" do
+      results = ["passed", "failed", "stopped", "stopped"]
+      {:ok, task} = Support.Factories.Task.create(%{fail_fast_strategy: "stop"})
+
+      assert {:ok, "failed"} = W.calculate_task_result(results, task)
+    end
+
+    test "when fail_fast:stop is active but no jobs failed => result is stopped" do
+      results = ["passed", "stopped", "stopped"]
+      {:ok, task} = Support.Factories.Task.create(%{fail_fast_strategy: "stop"})
+
+      assert {:ok, "stopped"} = W.calculate_task_result(results, task)
+    end
+
+    test "when fail_fast:cancel is active and jobs are stopped due to failure => result is stopped" do
+      results = ["passed", "failed", "stopped"]
+      {:ok, task} = Support.Factories.Task.create(%{fail_fast_strategy: "cancel"})
 
-      assert {:ok, "failed"} = W.calculate_task_result(results)
+      # With cancel strategy, we don't override the normal logic
+      assert {:ok, "stopped"} = W.calculate_task_result(results, task)
     end
   end