ECS: return failed status when exit code is empty (#1724)

treasure-data · Feb 9, 2023 · 10843b3 · 10843b3
1 parent a270498
commit 10843b3
Show file tree

Hide file tree

Showing 2 changed files with 38 additions and 2 deletions.
diff --git a/digdag-standards/src/main/java/io/digdag/standards/command/EcsCommandExecutor.java b/digdag-standards/src/main/java/io/digdag/standards/command/EcsCommandExecutor.java
@@ -397,8 +397,13 @@ CommandStatus createNextCommandStatus(
             // To fetch log until all logs is written in CloudWatch,
             // finish this poll once and wait finish marker in head of this method in next poll, considering risk of crushing in this poll.
             nextStatus.put("task_finished_at", Instant.now().getEpochSecond());
-            // Set exit code of container finished to nextStatus
-            nextStatus.put("status_code", task.getContainers().get(0).getExitCode());
+            // Set exit code of container finished to nextStatus.
+            // If exit code doesn't exist, something's wrong with execution, so set status code to 1 to make workflow fail
+            Integer exitCode = task.getContainers().get(0).getExitCode();
+            if (exitCode == null) {
+                logger.debug("Container has no exit code. The status code will be set as error (1)");
+            }
+            nextStatus.put("status_code", exitCode != null ? exitCode : 1);
         }
 
         // always return false to check if all logs are fetched. (return in head of this method after checking finish marker.)

diff --git a/digdag-standards/src/test/java/io/digdag/standards/command/EcsCommandExecutorTest.java b/digdag-standards/src/test/java/io/digdag/standards/command/EcsCommandExecutorTest.java
@@ -186,4 +186,35 @@ public void testGetErrorMessageFromTask()
             }
         }
     }
+
+    @Test
+    public void testErrorOnEmptyExitCodeFromContainer()
+            throws Exception
+    {
+        final EcsClient ecsClient = mock(EcsClient.class);
+        final CommandContext commandContext = mock(CommandContext.class);
+        final EcsCommandExecutor executor = spy(new EcsCommandExecutor(
+                systemConfig, ecsClientFactory, dockerCommandExecutor,
+                storageManager, projectArchiveLoader, commandLogger));
+        final Task task = mock(Task.class);
+        final Container container = mock(Container.class);
+
+        final ObjectNode previousStatusJson = om.createObjectNode();
+        previousStatusJson.put("cluster_name", "my_cluster");
+        previousStatusJson.put("task_arn", "my_task_arn");
+        previousStatusJson.put("executor_state", om.createObjectNode());
+        previousStatusJson.put("awslogs", om.createObjectNode().nullNode());
+
+        doReturn(mock(EcsClientConfig.class)).when(executor).createEcsClientConfig(any(Optional.class), any(Config.class), any(Config.class));
+        doReturn(ecsClient).when(ecsClientFactory).createClient(any(EcsClientConfig.class));
+        doReturn(mock(TaskRequest.class)).when(commandContext).getTaskRequest();
+        doReturn("stopped").when(task).getLastStatus();
+        doReturn(null).when(container).getExitCode();
+        doReturn(Arrays.asList(container)).when(task).getContainers();
+        doReturn(task).when(ecsClient).getTask(previousStatusJson.get("cluster_name").asText(), previousStatusJson.get("task_arn").asText());
+
+        CommandStatus commandStatus = executor.poll(commandContext, previousStatusJson);
+
+        assertThat(commandStatus.getStatusCode(), is(1));
+    }
 }