fix: update retry strategy for mutation calls to handle aborted transactions

aakashanandg · aakashanandg · commit e2b7413ec1db · 2025-01-02T18:58:29.000+05:30
diff --git a/.gitignore b/.gitignore
@@ -62,3 +62,7 @@ system_tests/local_test_setup
 # Make sure a generated file isn't accidentally committed.
 pylintrc
 pylintrc.test
+
+
+# Ignore coverage files
+.coverage*
diff --git a/google/cloud/spanner_dbapi/transaction_helper.py b/google/cloud/spanner_dbapi/transaction_helper.py
@@ -20,7 +20,7 @@
 
 from google.cloud.spanner_dbapi.batch_dml_executor import BatchMode
 from google.cloud.spanner_dbapi.exceptions import RetryAborted
-from google.cloud.spanner_v1.session import _get_retry_delay
+from google.cloud.spanner_v1._helpers import _get_retry_delay
 
 if TYPE_CHECKING:
     from google.cloud.spanner_dbapi import Connection, Cursor
diff --git a/google/cloud/spanner_v1/_helpers.py b/google/cloud/spanner_v1/_helpers.py
@@ -27,11 +27,15 @@
 from google.protobuf.internal.enum_type_wrapper import EnumTypeWrapper
 
 from google.api_core import datetime_helpers
+from google.api_core.exceptions import Aborted
 from google.cloud._helpers import _date_from_iso8601_date
 from google.cloud.spanner_v1 import TypeCode
 from google.cloud.spanner_v1 import ExecuteSqlRequest
 from google.cloud.spanner_v1 import JsonObject
 from google.cloud.spanner_v1.request_id_header import with_request_id
+from google.rpc.error_details_pb2 import RetryInfo
+
+import random
 
 # Validation error messages
 NUMERIC_MAX_SCALE_ERR_MSG = (
@@ -466,13 +470,19 @@ def _retry(
     delay=2,
     allowed_exceptions=None,
     beforeNextRetry=None,
+    deadline=None,
 ):
     """
-    Retry a function with a specified number of retries, delay between retries, and list of allowed exceptions.
+    Retry a specified function with different logic based on the type of exception raised.
+
+    If the exception is of type google.api_core.exceptions.Aborted,
+    apply an alternate retry strategy that relies on the provided deadline value instead of a fixed number of retries.
+    For all other exceptions, retry the function up to a specified number of times.
 
     Args:
         func: The function to be retried.
         retry_count: The maximum number of times to retry the function.
+        deadline: This will be used in case of Aborted transactions.
         delay: The delay in seconds between retries.
         allowed_exceptions: A tuple of exceptions that are allowed to occur without triggering a retry.
                             Passing allowed_exceptions as None will lead to retrying for all exceptions.
@@ -481,13 +491,21 @@ def _retry(
         The result of the function if it is successful, or raises the last exception if all retries fail.
     """
     retries = 0
-    while retries <= retry_count:
+    while True:
         if retries > 0 and beforeNextRetry:
             beforeNextRetry(retries, delay)
 
         try:
             return func()
         except Exception as exc:
+            if isinstance(exc, Aborted) and deadline is not None:
+                if (
+                    allowed_exceptions is not None
+                    and allowed_exceptions.get(exc.__class__) is not None
+                ):
+                    retries += 1
+                    _delay_until_retry(exc, deadline=deadline, attempts=retries)
+                    continue
             if (
                 allowed_exceptions is None or exc.__class__ in allowed_exceptions
             ) and retries < retry_count:
@@ -529,6 +547,60 @@ def _metadata_with_leader_aware_routing(value, **kw):
     return ("x-goog-spanner-route-to-leader", str(value).lower())
 
 
+def _delay_until_retry(exc, deadline, attempts):
+    """Helper for :meth:`Session.run_in_transaction`.
+
+    Detect retryable abort, and impose server-supplied delay.
+
+    :type exc: :class:`google.api_core.exceptions.Aborted`
+    :param exc: exception for aborted transaction
+
+    :type deadline: float
+    :param deadline: maximum timestamp to continue retrying the transaction.
+
+    :type attempts: int
+    :param attempts: number of call retries
+    """
+
+    cause = exc.errors[0]
+    now = time.time()
+    if now >= deadline:
+        raise
+
+    delay = _get_retry_delay(cause, attempts)
+    if delay is not None:
+        if now + delay > deadline:
+            raise
+
+        time.sleep(delay)
+
+
+def _get_retry_delay(cause, attempts):
+    """Helper for :func:`_delay_until_retry`.
+
+    :type exc: :class:`grpc.Call`
+    :param exc: exception for aborted transaction
+
+    :rtype: float
+    :returns: seconds to wait before retrying the transaction.
+
+    :type attempts: int
+    :param attempts: number of call retries
+    """
+    if hasattr(cause, "trailing_metadata"):
+        metadata = dict(cause.trailing_metadata())
+    else:
+        metadata = {}
+    retry_info_pb = metadata.get("google.rpc.retryinfo-bin")
+    if retry_info_pb is not None:
+        retry_info = RetryInfo()
+        retry_info.ParseFromString(retry_info_pb)
+        nanos = retry_info.retry_delay.nanos
+        return retry_info.retry_delay.seconds + nanos / 1.0e9
+
+    return 2**attempts + random.random()
+
+
 class AtomicCounter:
     def __init__(self, start_value=0):
         self.__lock = threading.Lock()
diff --git a/google/cloud/spanner_v1/batch.py b/google/cloud/spanner_v1/batch.py
@@ -31,6 +31,10 @@
 from google.cloud.spanner_v1._helpers import _retry
 from google.cloud.spanner_v1._helpers import _check_rst_stream_error
 from google.api_core.exceptions import InternalServerError
+from google.api_core.exceptions import Aborted
+import time
+
+DEFAULT_RETRY_TIMEOUT_SECS = 30
 
 
 class _BatchBase(_SessionWrapper):
@@ -162,6 +166,7 @@ def commit(
         request_options=None,
         max_commit_delay=None,
         exclude_txn_from_change_streams=False,
+        **kwargs,
     ):
         """Commit mutations to the database.
 
@@ -227,9 +232,16 @@ def commit(
                 request=request,
                 metadata=metadata,
             )
+            deadline = time.time() + kwargs.get(
+                "timeout_secs", DEFAULT_RETRY_TIMEOUT_SECS
+            )
             response = _retry(
                 method,
-                allowed_exceptions={InternalServerError: _check_rst_stream_error},
+                allowed_exceptions={
+                    InternalServerError: _check_rst_stream_error,
+                    Aborted: no_op_handler,
+                },
+                deadline=deadline,
             )
         self.committed = response.commit_timestamp
         self.commit_stats = response.commit_stats
@@ -293,7 +305,9 @@ def group(self):
         self._mutation_groups.append(mutation_group)
         return MutationGroup(self._session, mutation_group.mutations)
 
-    def batch_write(self, request_options=None, exclude_txn_from_change_streams=False):
+    def batch_write(
+        self, request_options=None, exclude_txn_from_change_streams=False, **kwargs
+    ):
         """Executes batch_write.
 
         :type request_options:
@@ -346,9 +360,16 @@ def batch_write(self, request_options=None, exclude_txn_from_change_streams=Fals
                 request=request,
                 metadata=metadata,
             )
+            deadline = time.time() + kwargs.get(
+                "timeout_secs", DEFAULT_RETRY_TIMEOUT_SECS
+            )
             response = _retry(
                 method,
-                allowed_exceptions={InternalServerError: _check_rst_stream_error},
+                allowed_exceptions={
+                    InternalServerError: _check_rst_stream_error,
+                    Aborted: no_op_handler,
+                },
+                deadline=deadline,
             )
         self.committed = True
         return response
@@ -372,3 +393,8 @@ def _make_write_pb(table, columns, values):
     return Mutation.Write(
         table=table, columns=columns, values=_make_list_value_pbs(values)
     )
+
+
+def no_op_handler(exc):
+    # No-op (does nothing)
+    pass
diff --git a/google/cloud/spanner_v1/database.py b/google/cloud/spanner_v1/database.py
@@ -775,6 +775,7 @@ def batch(
         request_options=None,
         max_commit_delay=None,
         exclude_txn_from_change_streams=False,
+        **kw,
     ):
         """Return an object which wraps a batch.
 
@@ -805,7 +806,7 @@ def batch(
         :returns: new wrapper
         """
         return BatchCheckout(
-            self, request_options, max_commit_delay, exclude_txn_from_change_streams
+            self, request_options, max_commit_delay, exclude_txn_from_change_streams, **kw
         )
 
     def mutation_groups(self):
@@ -1166,6 +1167,7 @@ def __init__(
         request_options=None,
         max_commit_delay=None,
         exclude_txn_from_change_streams=False,
+        **kw
     ):
         self._database = database
         self._session = self._batch = None
@@ -1177,6 +1179,7 @@ def __init__(
             self._request_options = request_options
         self._max_commit_delay = max_commit_delay
         self._exclude_txn_from_change_streams = exclude_txn_from_change_streams
+        self._kw = kw
 
     def __enter__(self):
         """Begin ``with`` block."""
@@ -1197,6 +1200,7 @@ def __exit__(self, exc_type, exc_val, exc_tb):
                     request_options=self._request_options,
                     max_commit_delay=self._max_commit_delay,
                     exclude_txn_from_change_streams=self._exclude_txn_from_change_streams,
+                    **self._kw
                 )
         finally:
             if self._database.log_commit_stats and self._batch.commit_stats:
diff --git a/google/cloud/spanner_v1/session.py b/google/cloud/spanner_v1/session.py
@@ -15,15 +15,15 @@
 """Wrapper for Cloud Spanner Session objects."""
 
 from functools import total_ordering
-import random
 import time
 from datetime import datetime
 
 from google.api_core.exceptions import Aborted
 from google.api_core.exceptions import GoogleAPICallError
 from google.api_core.exceptions import NotFound
 from google.api_core.gapic_v1 import method
-from google.rpc.error_details_pb2 import RetryInfo
+from google.cloud.spanner_v1._helpers import _delay_until_retry
+from google.cloud.spanner_v1._helpers import _get_retry_delay
 
 from google.cloud.spanner_v1 import ExecuteSqlRequest
 from google.cloud.spanner_v1 import CreateSessionRequest
@@ -554,57 +554,3 @@ def run_in_transaction(self, func, *args, **kw):
                             extra={"commit_stats": txn.commit_stats},
                         )
                     return return_value
-
-
-# Rational:  this function factors out complex shared deadline / retry
-#            handling from two `except:` clauses.
-def _delay_until_retry(exc, deadline, attempts):
-    """Helper for :meth:`Session.run_in_transaction`.
-
-    Detect retryable abort, and impose server-supplied delay.
-
-    :type exc: :class:`google.api_core.exceptions.Aborted`
-    :param exc: exception for aborted transaction
-
-    :type deadline: float
-    :param deadline: maximum timestamp to continue retrying the transaction.
-
-    :type attempts: int
-    :param attempts: number of call retries
-    """
-    cause = exc.errors[0]
-
-    now = time.time()
-
-    if now >= deadline:
-        raise
-
-    delay = _get_retry_delay(cause, attempts)
-    if delay is not None:
-        if now + delay > deadline:
-            raise
-
-        time.sleep(delay)
-
-
-def _get_retry_delay(cause, attempts):
-    """Helper for :func:`_delay_until_retry`.
-
-    :type exc: :class:`grpc.Call`
-    :param exc: exception for aborted transaction
-
-    :rtype: float
-    :returns: seconds to wait before retrying the transaction.
-
-    :type attempts: int
-    :param attempts: number of call retries
-    """
-    metadata = dict(cause.trailing_metadata())
-    retry_info_pb = metadata.get("google.rpc.retryinfo-bin")
-    if retry_info_pb is not None:
-        retry_info = RetryInfo()
-        retry_info.ParseFromString(retry_info_pb)
-        nanos = retry_info.retry_delay.nanos
-        return retry_info.retry_delay.seconds + nanos / 1.0e9
-
-    return 2**attempts + random.random()
diff --git a/tests/unit/test_batch.py b/tests/unit/test_batch.py
