Fix flyte error translation (#1085)

* fix(FlyteEventTranslator): clean error codes

Clean error codes in the flyte event translator to replace the RetriesExhaused| prefix that comes with all dynamic node failures and retriable nodes

* test(FlyteEventTranslator): add cleaning tests

Add tests to ensure that error codes with the retries exhaused prefix are matching the expected error code

* Fix typo

---------

Co-authored-by: Brandon Segal <brandonsegal.k@gmail.com>

styx-scheduler-service/src/main/java/com/spotify/styx/flyte/FlyteEventTranslator.java

@@ -38,7 +38,7 @@
 
 public class FlyteEventTranslator {
   private static final Logger LOG = LoggerFactory.getLogger(FlyteEventTranslator.class);
-
+  private static final String RETRIES_EXHAUSTED = "RetriesExhausted|";
   private static final String PERSISTED_CODE = "USER:Persisted";
   private static final String NOT_RETRYABLE_CODE = "USER:NotRetryable";
   private static final String NOT_READY_CODE = "USER:NotReady";
@@ -96,7 +96,7 @@ static private List<Event> handleExited(ExecutionOuterClass.Execution execution,
       case FAILED:
       case ABORTED:
       case TIMED_OUT:
-        final String flyteCode = execution.getClosure().getError().getCode();
+        final String flyteCode = execution.getClosure().getError().getCode().replace(RETRIES_EXHAUSTED,"");
         final int styxCode = KNOWN_ERROR_CODES.getOrDefault(flyteCode, UNKNOWN_ERROR_EXIT_CODE);
         LOG.info("Issue 'terminate' event for: " + runState.workflowInstance());
         generatedEvents.add(Event.terminate(runState.workflowInstance(), Optional.of(styxCode)));

styx-scheduler-service/src/test/java/com/spotify/styx/flyte/FlyteEventTranslatorTest.java

@@ -138,6 +138,19 @@ private Object[] parametersForTestTransitionPollingToTerminateExitCodes() {
         new Object[] { Execution.WorkflowExecution.Phase.FAILED, "USER:AnythingElse", UNKNOWN_ERROR_EXIT_CODE },
         new Object[] { Execution.WorkflowExecution.Phase.ABORTED, "USER:AnythingElse", UNKNOWN_ERROR_EXIT_CODE},
         new Object[] { Execution.WorkflowExecution.Phase.TIMED_OUT, "USER:AnythingElse", UNKNOWN_ERROR_EXIT_CODE},
+        new Object[] { Execution.WorkflowExecution.Phase.FAILED, "RetriesExhausted|USER:NotReady", MISSING_DEPS_EXIT_CODE },
+        new Object[] { Execution.WorkflowExecution.Phase.FAILED, "RetriesExhausted|USER:NotReady", MISSING_DEPS_EXIT_CODE },
+        new Object[] { Execution.WorkflowExecution.Phase.ABORTED, "RetriesExhausted|USER:NotReady", MISSING_DEPS_EXIT_CODE },
+        new Object[] { Execution.WorkflowExecution.Phase.TIMED_OUT, "RetriesExhausted|USER:NotReady", MISSING_DEPS_EXIT_CODE },
+        new Object[] { Execution.WorkflowExecution.Phase.FAILED, "RetriesExhausted|USER:NotRetryable", UNRECOVERABLE_FAILURE_EXIT_CODE },
+        new Object[] { Execution.WorkflowExecution.Phase.ABORTED, "RetriesExhausted|USER:NotRetryable", UNRECOVERABLE_FAILURE_EXIT_CODE},
+        new Object[] { Execution.WorkflowExecution.Phase.TIMED_OUT, "RetriesExhausted|USER:NotRetryable", UNRECOVERABLE_FAILURE_EXIT_CODE},
+        new Object[] { Execution.WorkflowExecution.Phase.FAILED, "RetriesExhausted|USER:Persisted", SUCCESS_EXIT_CODE },
+        new Object[] { Execution.WorkflowExecution.Phase.ABORTED, "RetriesExhausted|USER:Persisted", SUCCESS_EXIT_CODE},
+        new Object[] { Execution.WorkflowExecution.Phase.TIMED_OUT, "RetriesExhausted|USER:Persisted", SUCCESS_EXIT_CODE},
+        new Object[] { Execution.WorkflowExecution.Phase.FAILED, "RetriesExhausted|USER:AnythingElse", UNKNOWN_ERROR_EXIT_CODE },
+        new Object[] { Execution.WorkflowExecution.Phase.ABORTED, "RetriesExhausted|USER:AnythingElse", UNKNOWN_ERROR_EXIT_CODE},
+        new Object[] { Execution.WorkflowExecution.Phase.TIMED_OUT, "RetriesExhausted|USER:AnythingElse", UNKNOWN_ERROR_EXIT_CODE},
         };
   }