Handle EMR Exceptions in FlintCancelJob Operation

Signed-off-by: Vamsi Manohar <reddyvam@amazon.com>
opensearch-project · Mar 22, 2024 · 1c49359 · 1c49359
1 parent 4dc83b7
commit 1c49359
Show file tree

Hide file tree

Showing 13 changed files with 1,151 additions and 890 deletions.
diff --git a/spark/src/main/java/org/opensearch/sql/spark/client/EMRServerlessClient.java b/spark/src/main/java/org/opensearch/sql/spark/client/EMRServerlessClient.java
@@ -41,5 +41,6 @@ public interface EMRServerlessClient {
    * @param jobId jobId.
    * @return {@link CancelJobRunResult}
    */
-  CancelJobRunResult cancelJobRun(String applicationId, String jobId);
+  CancelJobRunResult cancelJobRun(
+      String applicationId, String jobId, boolean allowExceptionPropagation);
 }
diff --git a/spark/src/main/java/org/opensearch/sql/spark/client/EmrServerlessClientImpl.java b/spark/src/main/java/org/opensearch/sql/spark/client/EmrServerlessClientImpl.java
@@ -32,7 +32,7 @@ public class EmrServerlessClientImpl implements EMRServerlessClient {
 
   private static final int MAX_JOB_NAME_LENGTH = 255;
 
-  private static final String GENERIC_INTERNAL_SERVER_ERROR_MESSAGE = "Internal Server Error.";
+  public static final String GENERIC_INTERNAL_SERVER_ERROR_MESSAGE = "Internal Server Error.";
 
   public EmrServerlessClientImpl(AWSEMRServerless emrServerless) {
     this.emrServerless = emrServerless;
@@ -98,7 +98,8 @@ public GetJobRunResult getJobRunResult(String applicationId, String jobId) {
   }
 
   @Override
-  public CancelJobRunResult cancelJobRun(String applicationId, String jobId) {
+  public CancelJobRunResult cancelJobRun(
+      String applicationId, String jobId, boolean allowExceptionPropagation) {
     CancelJobRunRequest cancelJobRunRequest =
         new CancelJobRunRequest().withJobRunId(jobId).withApplicationId(applicationId);
     CancelJobRunResult cancelJobRunResult =
@@ -108,10 +109,14 @@ public CancelJobRunResult cancelJobRun(String applicationId, String jobId) {
                   try {
                     return emrServerless.cancelJobRun(cancelJobRunRequest);
                   } catch (Throwable t) {
-                    logger.error("Error while making cancel job request to emr:", t);
-                    MetricUtils.incrementNumericalMetric(
-                        MetricName.EMR_CANCEL_JOB_REQUEST_FAILURE_COUNT);
-                    throw new RuntimeException(GENERIC_INTERNAL_SERVER_ERROR_MESSAGE);
+                    if (allowExceptionPropagation) {
+                      throw t;
+                    } else {
+                      logger.error("Error while making cancel job request to emr:", t);
+                      MetricUtils.incrementNumericalMetric(
+                          MetricName.EMR_CANCEL_JOB_REQUEST_FAILURE_COUNT);
+                      throw new RuntimeException(GENERIC_INTERNAL_SERVER_ERROR_MESSAGE);
+                    }
                   }
                 });
     logger.info(String.format("Job : %s cancelled", cancelJobRunResult.getJobRunId()));

diff --git a/spark/src/main/java/org/opensearch/sql/spark/dispatcher/BatchQueryHandler.java b/spark/src/main/java/org/opensearch/sql/spark/dispatcher/BatchQueryHandler.java
@@ -58,7 +58,7 @@ protected JSONObject getResponseFromExecutor(AsyncQueryJobMetadata asyncQueryJob
   @Override
   public String cancelJob(AsyncQueryJobMetadata asyncQueryJobMetadata) {
     emrServerlessClient.cancelJobRun(
-        asyncQueryJobMetadata.getApplicationId(), asyncQueryJobMetadata.getJobId());
+        asyncQueryJobMetadata.getApplicationId(), asyncQueryJobMetadata.getJobId(), false);
     return asyncQueryJobMetadata.getQueryId().getId();
   }
 

diff --git a/spark/src/main/java/org/opensearch/sql/spark/execution/session/InteractiveSession.java b/spark/src/main/java/org/opensearch/sql/spark/execution/session/InteractiveSession.java
@@ -80,7 +80,8 @@ public void close() {
     if (model.isEmpty()) {
       throw new IllegalStateException("session does not exist. " + sessionModel.getSessionId());
     } else {
-      serverlessClient.cancelJobRun(sessionModel.getApplicationId(), sessionModel.getJobId());
+      serverlessClient.cancelJobRun(
+          sessionModel.getApplicationId(), sessionModel.getJobId(), false);
     }
   }
 

diff --git a/spark/src/main/java/org/opensearch/sql/spark/flint/operation/FlintIndexOp.java b/spark/src/main/java/org/opensearch/sql/spark/flint/operation/FlintIndexOp.java
@@ -5,10 +5,13 @@
 
 package org.opensearch.sql.spark.flint.operation;
 
+import static org.opensearch.sql.spark.client.EmrServerlessClientImpl.GENERIC_INTERNAL_SERVER_ERROR_MESSAGE;
 import static org.opensearch.sql.spark.execution.statestore.StateStore.deleteFlintIndexState;
 import static org.opensearch.sql.spark.execution.statestore.StateStore.getFlintIndexState;
 import static org.opensearch.sql.spark.execution.statestore.StateStore.updateFlintIndexState;
 
+import com.amazonaws.services.emrserverless.model.ResourceNotFoundException;
+import com.amazonaws.services.emrserverless.model.ValidationException;
 import java.util.Locale;
 import java.util.Optional;
 import java.util.concurrent.TimeUnit;
@@ -145,11 +148,18 @@ public void cancelStreamingJob(
     String jobId = flintIndexStateModel.getJobId();
     try {
       emrServerlessClient.cancelJobRun(
-          flintIndexStateModel.getApplicationId(), flintIndexStateModel.getJobId());
-    } catch (IllegalArgumentException e) {
-      // handle job does not exist case.
+          flintIndexStateModel.getApplicationId(), flintIndexStateModel.getJobId(), true);
+    } catch (ResourceNotFoundException e) {
+      // JobId Not Found Exception.
       LOG.error(e);
       return;
+    } catch (ValidationException e) {
+      // Exception when the job is not in cancellable state and already in terminal state.
+      LOG.error(e);
+      return;
+    } catch (Exception e) {
+      LOG.error(e);
+      throw new RuntimeException(GENERIC_INTERNAL_SERVER_ERROR_MESSAGE);
     }
 
     // pull job state until timeout or cancelled.

diff --git a/spark/src/test/java/org/opensearch/sql/spark/asyncquery/AsyncQueryExecutorServiceSpec.java b/spark/src/test/java/org/opensearch/sql/spark/asyncquery/AsyncQueryExecutorServiceSpec.java
@@ -243,7 +243,8 @@ public GetJobRunResult getJobRunResult(String applicationId, String jobId) {
     }
 
     @Override
-    public CancelJobRunResult cancelJobRun(String applicationId, String jobId) {
+    public CancelJobRunResult cancelJobRun(
+        String applicationId, String jobId, boolean allowExceptionPropagation) {
       cancelJobRunCalled++;
       return new CancelJobRunResult().withJobRunId(jobId);
     }