feat: make transaction retry configurable

Author: olavloite

google-cloud-spanner/src/main/java/com/google/cloud/spanner/SessionImpl.java

@@ -301,7 +301,7 @@ public CommitResponse writeAtLeastOnceWithOptions(
     ISpan span = tracer.spanBuilder(SpannerImpl.COMMIT);
 
     try (IScope s = tracer.withSpan(span)) {
-      return SpannerRetryHelper.runTxWithRetriesOnAborted(
+      return spanner.getTransactionRetryHelper().runTxWithRetriesOnAborted(
           () -> new CommitResponse(spanner.getRpc().commit(request, getOptions())));
     } catch (RuntimeException e) {
       span.setStatus(e);

google-cloud-spanner/src/main/java/com/google/cloud/spanner/SpannerImpl.java

@@ -116,6 +116,7 @@ private static String nextDatabaseClientId(DatabaseId databaseId) {
 
   private final DatabaseAdminClient dbAdminClient;
   private final InstanceAdminClient instanceClient;
+  private final TransactionRetryHelper transactionRetryHelper;
 
   /**
    * Exception class used to track the stack trace at the point when a Spanner instance is closed.
@@ -145,6 +146,7 @@ static final class ClosedException extends RuntimeException {
     this.dbAdminClient = new DatabaseAdminClientImpl(options.getProjectId(), gapicRpc);
     this.instanceClient =
         new InstanceAdminClientImpl(options.getProjectId(), gapicRpc, dbAdminClient);
+    this.transactionRetryHelper = new TransactionRetryHelper(options.getDefaultTransactionRetrySettings());
     logSpannerOptions(options);
   }
 
@@ -200,6 +202,10 @@ SpannerRpc getRpc() {
     return gapicRpc;
   }
 
+  TransactionRetryHelper getTransactionRetryHelper() {
+    return transactionRetryHelper;
+  }
+
   /** Returns the default setting for prefetchChunks of this {@link SpannerImpl} instance. */
   int getDefaultPrefetchChunks() {
     return getOptions().getPrefetchChunks();

google-cloud-spanner/src/main/java/com/google/cloud/spanner/SpannerOptions.java

@@ -58,6 +58,7 @@
 import com.google.cloud.spanner.spi.v1.GapicSpannerRpc;
 import com.google.cloud.spanner.spi.v1.SpannerRpc;
 import com.google.cloud.spanner.v1.SpannerSettings;
+import com.google.cloud.spanner.v1.stub.SpannerStub;
 import com.google.cloud.spanner.v1.stub.SpannerStubSettings;
 import com.google.common.annotations.VisibleForTesting;
 import com.google.common.base.MoreObjects;
@@ -70,6 +71,7 @@
 import com.google.spanner.v1.ExecuteSqlRequest;
 import com.google.spanner.v1.ExecuteSqlRequest.QueryOptions;
 import com.google.spanner.v1.RequestOptions;
+import com.google.spanner.v1.RollbackRequest;
 import com.google.spanner.v1.SpannerGrpc;
 import com.google.spanner.v1.TransactionOptions;
 import com.google.spanner.v1.TransactionOptions.IsolationLevel;
@@ -204,6 +206,25 @@ public static GcpChannelPoolOptions createDefaultDynamicChannelPoolOptions() {
         .build();
   }
 
+  /**
+   * Use the same {@link RetrySettings} for retrying an aborted transaction as for retrying a {@link
+   * RollbackRequest}. The {@link RollbackRequest} automatically uses the default retry settings
+   * defined for the {@link SpannerStub}. By referencing these settings, the retry settings for
+   * retrying aborted transactions will also automatically be updated if the default retry settings
+   * are updated.
+   *
+   * <p>A read/write transaction should not time out while retrying. The total timeout of the retry
+   * settings is therefore set to 24 hours and there is no max attempts value.
+   *
+   * <p>These default {@link RetrySettings} are only used if no retry information is returned by the
+   * {@link AbortedException}.
+   */
+  public static final RetrySettings DEFAULT_TRANSACTION_RETRY_SETTINGS =
+      SpannerStubSettings.newBuilder().rollbackSettings().getRetrySettings().toBuilder()
+          .setTotalTimeoutDuration(Duration.ofHours(24L))
+          .setMaxAttempts(0)
+          .build();
+
   private final TransportChannelProvider channelProvider;
   private final ChannelEndpointCacheFactory channelEndpointCacheFactory;
 
@@ -264,6 +285,7 @@ public static GcpChannelPoolOptions createDefaultDynamicChannelPoolOptions() {
   private final boolean enableEndToEndTracing;
   private final String monitoringHost;
   private final TransactionOptions defaultTransactionOptions;
+  private final RetrySettings defaultTransactionRetrySettings;
   private final RequestOptions.ClientContext clientContext;
 
   enum TracingFramework {
@@ -934,6 +956,7 @@ protected SpannerOptions(Builder builder) {
     enableEndToEndTracing = builder.enableEndToEndTracing;
     monitoringHost = builder.monitoringHost;
     defaultTransactionOptions = builder.defaultTransactionOptions;
+    defaultTransactionRetrySettings = builder.defaultTransactionRetrySettings;
     clientContext = builder.clientContext;
   }
 
@@ -1196,6 +1219,7 @@ public static class Builder
     private String experimentalHost = null;
     private boolean usePlainText = false;
     private TransactionOptions defaultTransactionOptions = TransactionOptions.getDefaultInstance();
+    private RetrySettings defaultTransactionRetrySettings = TransactionRetryHelper.DEFAULT_TRANSACTION_RETRY_SETTINGS;
     private RequestOptions.ClientContext clientContext;
 
     private static String createCustomClientLibToken(String token) {
@@ -1302,6 +1326,7 @@ protected Builder() {
       this.enableEndToEndTracing = options.enableEndToEndTracing;
       this.monitoringHost = options.monitoringHost;
       this.defaultTransactionOptions = options.defaultTransactionOptions;
+      this.defaultTransactionRetrySettings = options.defaultTransactionRetrySettings;
       this.clientContext = options.clientContext;
     }
 
@@ -2055,6 +2080,13 @@ public Builder setDefaultTransactionOptions(
       return this;
     }
 
+    /** Sets the default {@link RetrySettings} for all read/write transactions that are executed using this client. These settings are used when the client automatically retries an aborted read/write transaction. The default is to retry for up to 24 hours without a limit for the maximum number of attempts. */
+    public Builder setDefaultTransactionRetrySettings(RetrySettings retrySettings) {
+      Preconditions.checkNotNull(retrySettings, "RetrySettings cannot be null");
+      this.defaultTransactionRetrySettings = retrySettings;
+      return this;
+    }
+
     /** Sets the default {@link RequestOptions.ClientContext} for all requests. */
     public Builder setDefaultClientContext(RequestOptions.ClientContext clientContext) {
       this.clientContext = clientContext;
@@ -2481,6 +2513,10 @@ public TransactionOptions getDefaultTransactionOptions() {
     return defaultTransactionOptions;
   }
 
+  public RetrySettings getDefaultTransactionRetrySettings() {
+    return this.defaultTransactionRetrySettings;
+  }
+
   @BetaApi
   public boolean isUseVirtualThreads() {
     return useVirtualThreads;

…le/cloud/spanner/SpannerRetryHelper.java …loud/spanner/TransactionRetryHelper.javagoogle-cloud-spanner/src/main/java/com/google/cloud/spanner/SpannerRetryHelper.java renamed to google-cloud-spanner/src/main/java/com/google/cloud/spanner/TransactionRetryHelper.java google-cloud-spanner/src/main/java/com/google/cloud/spanner/SpannerRetryHelper.java renamed to google-cloud-spanner/src/main/java/com/google/cloud/spanner/TransactionRetryHelper.java

@@ -39,50 +39,35 @@
  * that uses specific settings to only retry on aborted transactions, without a timeout and without
  * a cap on the number of retries.
  */
-class SpannerRetryHelper {
+class TransactionRetryHelper {
+  private final RetrySettings retrySettings;
 
-  /**
-   * Use the same {@link RetrySettings} for retrying an aborted transaction as for retrying a {@link
-   * RollbackRequest}. The {@link RollbackRequest} automatically uses the default retry settings
-   * defined for the {@link SpannerStub}. By referencing these settings, the retry settings for
-   * retrying aborted transactions will also automatically be updated if the default retry settings
-   * are updated.
-   *
-   * <p>A read/write transaction should not timeout while retrying. The total timeout of the retry
-   * settings is therefore set to 24 hours and there is no max attempts value.
-   *
-   * <p>These default {@link RetrySettings} are only used if no retry information is returned by the
-   * {@link AbortedException}.
-   */
-  @VisibleForTesting
-  static final RetrySettings txRetrySettings =
-      SpannerStubSettings.newBuilder().rollbackSettings().getRetrySettings().toBuilder()
-          .setTotalTimeoutDuration(Duration.ofHours(24L))
-          .setMaxAttempts(0)
-          .build();
+  TransactionRetryHelper(RetrySettings retrySettings) {
+    this.retrySettings = retrySettings;
+  }
 
   /** Executes the {@link Callable} and retries if it fails with an {@link AbortedException}. */
-  static <T> T runTxWithRetriesOnAborted(Callable<T> callable) {
+  <T> T runTxWithRetriesOnAborted(Callable<T> callable) {
     return runTxWithRetriesOnAborted(callable, DefaultErrorHandler.INSTANCE);
   }
 
-  static <T> T runTxWithRetriesOnAborted(Callable<T> callable, ErrorHandler errorHandler) {
+  <T> T runTxWithRetriesOnAborted(Callable<T> callable, ErrorHandler errorHandler) {
     return runTxWithRetriesOnAborted(
-        callable, errorHandler, txRetrySettings, NanoClock.getDefaultClock());
+        callable, errorHandler, this.retrySettings, NanoClock.getDefaultClock());
   }
 
   /**
    * Executes the {@link Callable} and retries if it fails with an {@link AbortedException} using
    * the specific {@link RetrySettings}.
    */
   @VisibleForTesting
-  static <T> T runTxWithRetriesOnAborted(
+  <T> T runTxWithRetriesOnAborted(
       Callable<T> callable, RetrySettings retrySettings, ApiClock clock) {
     return runTxWithRetriesOnAborted(callable, DefaultErrorHandler.INSTANCE, retrySettings, clock);
   }
 
   @VisibleForTesting
-  static <T> T runTxWithRetriesOnAborted(
+  <T> T runTxWithRetriesOnAborted(
       Callable<T> callable,
       ErrorHandler errorHandler,
       RetrySettings retrySettings,

google-cloud-spanner/src/main/java/com/google/cloud/spanner/TransactionRunnerImpl.java

@@ -1357,7 +1357,7 @@ private <T> T runInternal(final TransactionCallable<T> txCallable) {
             throw e;
           }
         };
-    return SpannerRetryHelper.runTxWithRetriesOnAborted(retryCallable, session.getErrorHandler());
+    return session.getSpanner().getTransactionRetryHelper().runTxWithRetriesOnAborted(retryCallable, session.getErrorHandler());
   }
 
   @Override

…loud/spanner/SpannerRetryHelperTest.java …/spanner/TransactionRetryHelperTest.javagoogle-cloud-spanner/src/test/java/com/google/cloud/spanner/SpannerRetryHelperTest.java renamed to google-cloud-spanner/src/test/java/com/google/cloud/spanner/TransactionRetryHelperTest.java google-cloud-spanner/src/test/java/com/google/cloud/spanner/SpannerRetryHelperTest.java renamed to google-cloud-spanner/src/test/java/com/google/cloud/spanner/TransactionRetryHelperTest.java

@@ -43,7 +43,7 @@
 import org.junit.runners.JUnit4;
 
 @RunWith(JUnit4.class)
-public class SpannerRetryHelperTest {
+public class TransactionRetryHelperTest {
   private static class FakeClock implements ApiClock {
     private long currentTime;
 
@@ -58,6 +58,8 @@ public long millisTime() {
     }
   }
 
+  private final TransactionRetryHelper retryHelper = new TransactionRetryHelper(SpannerOptions.DEFAULT_TRANSACTION_RETRY_SETTINGS);
+
   @Test
   public void testRetryDoesNotTimeoutAfterTenMinutes() {
     final FakeClock clock = new FakeClock();
@@ -72,8 +74,8 @@ public void testRetryDoesNotTimeoutAfterTenMinutes() {
         };
     assertEquals(
         2,
-        SpannerRetryHelper.runTxWithRetriesOnAborted(
-                callable, SpannerRetryHelper.txRetrySettings, clock)
+        retryHelper.runTxWithRetriesOnAborted(
+                callable, SpannerOptions.DEFAULT_TRANSACTION_RETRY_SETTINGS, clock)
             .intValue());
   }
 
@@ -93,8 +95,8 @@ public void testRetryDoesFailAfterMoreThanOneDay() {
         assertThrows(
             SpannerException.class,
             () ->
-                SpannerRetryHelper.runTxWithRetriesOnAborted(
-                    callable, SpannerRetryHelper.txRetrySettings, clock));
+                retryHelper.runTxWithRetriesOnAborted(
+                    callable, SpannerOptions.DEFAULT_TRANSACTION_RETRY_SETTINGS, clock));
     assertEquals(ErrorCode.ABORTED, e.getErrorCode());
     assertEquals(1, attempts.get());
   }
@@ -119,7 +121,7 @@ public void testCancelledContext() {
         assertThrows(
             SpannerException.class,
             () ->
-                withCancellation.run(() -> SpannerRetryHelper.runTxWithRetriesOnAborted(callable)));
+                withCancellation.run(() -> retryHelper.runTxWithRetriesOnAborted(callable)));
     assertEquals(ErrorCode.CANCELLED, e.getErrorCode());
   }
 
@@ -135,14 +137,14 @@ public void testTimedOutContext() {
     SpannerException e =
         assertThrows(
             SpannerException.class,
-            () -> withDeadline.run(() -> SpannerRetryHelper.runTxWithRetriesOnAborted(callable)));
+            () -> withDeadline.run(() -> retryHelper.runTxWithRetriesOnAborted(callable)));
     assertEquals(ErrorCode.DEADLINE_EXCEEDED, e.getErrorCode());
   }
 
   @Test
   public void noException() {
     Callable<Integer> callable = () -> 1 + 1;
-    assertThat(SpannerRetryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
+    assertThat(retryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
   }
 
   @Test(expected = IllegalStateException.class)
@@ -151,7 +153,7 @@ public void propagateUncheckedException() {
         () -> {
           throw new IllegalStateException("test");
         };
-    SpannerRetryHelper.runTxWithRetriesOnAborted(callable);
+    retryHelper.runTxWithRetriesOnAborted(callable);
   }
 
   @Test
@@ -164,7 +166,7 @@ public void retryOnAborted() {
           }
           return 1 + 1;
         };
-    assertThat(SpannerRetryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
+    assertThat(retryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
   }
 
   @Test
@@ -177,7 +179,7 @@ public void retryMultipleTimesOnAborted() {
           }
           return 1 + 1;
         };
-    assertThat(SpannerRetryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
+    assertThat(retryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
   }
 
   @Test(expected = IllegalStateException.class)
@@ -190,7 +192,7 @@ public void retryOnAbortedAndThenPropagateUnchecked() {
           }
           throw new IllegalStateException("test");
         };
-    SpannerRetryHelper.runTxWithRetriesOnAborted(callable);
+    retryHelper.runTxWithRetriesOnAborted(callable);
   }
 
   @Test
@@ -238,7 +240,7 @@ public void testExceptionWithRetryInfo() {
     // The following call should take at least 100ms, as that is the retry delay specified in the
     // retry info of the exception.
     Stopwatch watch = Stopwatch.createStarted();
-    assertThat(SpannerRetryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
+    assertThat(retryHelper.runTxWithRetriesOnAborted(callable)).isEqualTo(2);
     long elapsed = watch.elapsed(TimeUnit.MILLISECONDS);
     // Allow 1ms difference as that should be the accuracy of the sleep method.
     assertThat(elapsed).isAtLeast(RETRY_DELAY_MILLIS - 1);