perf(spanner): optimize request ID header generation and retry closures

Optimize request ID generation and RPC retry hot paths across Database, Transaction, Snapshot, and Batch:
- Use `int.from_bytes(os.urandom(8), "big")` for process-level random ID generation and pre-format the static process prefix once at import time.
- Cache per-database request ID prefixes directly in `instance.__dict__` via a non-data descriptor (`_CachedPrefixDescriptor`), lazily computing on first access and invalidating when `_channel_id` changes.
- Inline `Database.with_error_augmentation` and construct gRPC metadata lists with unpacking (`[*prior_metadata, ...]`) instead of intermediate list copies and `.append()`.
- Replace thread-safe `AtomicCounter` instances with local integer counters (`attempt = 0`) inside sequential retry closures, and pass active OpenTelemetry spans directly to avoid repeated `contextvars` lookups.
- Eliminate transient `functools.partial` allocations and redundant `*args, **kwargs` unpacking inside retry closures.
- Make `wrap_with_request_id` idempotent on repeated wrapping and replace stale request IDs when retrying failed calls.

Author: olavloite

packages/google-cloud-spanner/google/cloud/spanner_v1/_async/batch.py

@@ -15,7 +15,6 @@
 """Context manager for Cloud Spanner batched writes."""
 
 __CROSS_SYNC_OUTPUT__ = "google.cloud.spanner_v1.batch"
-import functools
 import time
 from typing import List, Optional
 
@@ -25,7 +24,6 @@
 from google.cloud.aio._cross_sync import CrossSync
 from google.cloud.spanner_v1._async._helpers import _retry, _retry_on_aborted_exception
 from google.cloud.spanner_v1._helpers import (
-    AtomicCounter,
     _check_rst_stream_error,
     _make_list_value_pb,
     _make_list_value_pbs,
@@ -341,13 +339,11 @@ async def wrapped_method():
                     metadata,
                     span,
                 )
-                commit_method = functools.partial(
-                    api.commit,
-                    request=commit_request,
-                    metadata=call_metadata,
-                )
                 with error_augmenter:
-                    return await commit_method()
+                    return await api.commit(
+                        request=commit_request,
+                        metadata=call_metadata,
+                    )
 
             response = await _retry_on_aborted_exception(
                 wrapped_method,
@@ -478,27 +474,27 @@ async def batch_write(
             ) as span,
             MetricsCapture(self._resource_info),
         ):
-            attempt = AtomicCounter(0)
+            attempt = 0
             nth_request = getattr(database, "_next_nth_request", 0)
 
             def wrapped_method():
+                nonlocal attempt
+                attempt += 1
                 batch_write_request = BatchWriteRequest(
                     session=session.name,
                     mutation_groups=mutation_groups,
                     request_options=request_options,
                     exclude_txn_from_change_streams=exclude_txn_from_change_streams,
                 )
-                batch_write_method = functools.partial(
-                    api.batch_write,
+                return api.batch_write(
                     request=batch_write_request,
                     metadata=database.metadata_with_request_id(
                         nth_request,
-                        attempt.increment(),
+                        attempt,
                         metadata,
                         span,
                     ),
                 )
-                return batch_write_method()
 
             response = await _retry(
                 wrapped_method,

packages/google-cloud-spanner/google/cloud/spanner_v1/_async/database.py

@@ -58,11 +58,14 @@
     _merge_query_options,
     _metadata_with_leader_aware_routing,
     _metadata_with_prefix,
-    _metadata_with_request_id,
-    _metadata_with_request_id_and_req_id,
 )
 from google.cloud.spanner_v1.keyset import KeySet
 from google.cloud.spanner_v1.merged_result_set import MergedResultSet
+from google.cloud.spanner_v1.request_id_header import (
+    REQ_ID_HEADER_KEY,
+    X_GOOG_SPANNER_REQUEST_ID_SPAN_ATTR,
+    _CachedPrefixDescriptor,
+)
 from google.cloud.spanner_v1.services.spanner.async_client import (
     SpannerAsyncClient as SpannerClient,
 )
@@ -180,6 +183,8 @@ class Database(object):
 
     __transport_lock = threading.Lock()
     __transports_to_channel_id = dict()
+    _channel_id_val = 0
+    _req_id_prefix = _CachedPrefixDescriptor()
 
     def __init__(
         self,
@@ -550,23 +555,17 @@ def spanner_api(self):
 
         return self._spanner_api
 
-    def metadata_with_request_id(
-        self, nth_request, nth_attempt, prior_metadata=[], span=None
-    ):
-        if span is None:
-            span = get_current_span()
+    @property
+    def _channel_id(self):
+        return self._channel_id_val
 
-        return _metadata_with_request_id(
-            self._nth_client_id,
-            self._channel_id,
-            nth_request,
-            nth_attempt,
-            prior_metadata,
-            span,
-        )
+    @_channel_id.setter
+    def _channel_id(self, value):
+        self._channel_id_val = value
+        self.__dict__.pop("_req_id_prefix", None)
 
     def metadata_and_request_id(
-        self, nth_request, nth_attempt, prior_metadata=[], span=None
+        self, nth_request, nth_attempt, prior_metadata=None, span=None
     ):
         """Return metadata and request ID string.
 
@@ -585,17 +584,38 @@ def metadata_and_request_id(
         if span is None:
             span = get_current_span()
 
-        return _metadata_with_request_id_and_req_id(
-            self._nth_client_id,
-            self._channel_id,
-            nth_request,
-            nth_attempt,
-            prior_metadata,
-            span,
+        req_id = f"{self._req_id_prefix}{nth_request}.{nth_attempt}"
+        metadata = (
+            [*prior_metadata, (REQ_ID_HEADER_KEY, req_id)]
+            if prior_metadata
+            else [(REQ_ID_HEADER_KEY, req_id)]
         )
 
+        if span is not None and span.is_recording():
+            span.set_attribute(X_GOOG_SPANNER_REQUEST_ID_SPAN_ATTR, req_id)
+
+        return metadata, req_id
+
+    def metadata_with_request_id(
+        self, nth_request, nth_attempt, prior_metadata=None, span=None
+    ):
+        if span is None:
+            span = get_current_span()
+
+        req_id = f"{self._req_id_prefix}{nth_request}.{nth_attempt}"
+        metadata = (
+            [*prior_metadata, (REQ_ID_HEADER_KEY, req_id)]
+            if prior_metadata
+            else [(REQ_ID_HEADER_KEY, req_id)]
+        )
+
+        if span is not None and span.is_recording():
+            span.set_attribute(X_GOOG_SPANNER_REQUEST_ID_SPAN_ATTR, req_id)
+
+        return metadata
+
     def with_error_augmentation(
-        self, nth_request, nth_attempt, prior_metadata=[], span=None
+        self, nth_request, nth_attempt, prior_metadata=None, span=None
     ):
         """Context manager for gRPC calls with error augmentation.
 
@@ -614,16 +634,17 @@ def with_error_augmentation(
         if span is None:
             span = get_current_span()
 
-        metadata, request_id = _metadata_with_request_id_and_req_id(
-            self._nth_client_id,
-            self._channel_id,
-            nth_request,
-            nth_attempt,
-            prior_metadata,
-            span,
+        req_id = f"{self._req_id_prefix}{nth_request}.{nth_attempt}"
+        metadata = (
+            [*prior_metadata, (REQ_ID_HEADER_KEY, req_id)]
+            if prior_metadata
+            else [(REQ_ID_HEADER_KEY, req_id)]
         )
 
-        return metadata, _augment_errors_with_request_id(request_id)
+        if span is not None and span.is_recording():
+            span.set_attribute(X_GOOG_SPANNER_REQUEST_ID_SPAN_ATTR, req_id)
+
+        return metadata, _augment_errors_with_request_id(req_id)
 
     def __eq__(self, other):
         if not isinstance(other, self.__class__):
@@ -975,13 +996,13 @@ async def execute_pdml():
     @property
     def _next_nth_request(self):
         if self._instance and self._instance._client:
-            return self._instance._client._next_nth_request
+            return getattr(self._instance._client, "_next_nth_request", 1)
         return 1
 
     @property
     def _nth_client_id(self):
         if self._instance and self._instance._client:
-            return self._instance._client._nth_client_id
+            return getattr(self._instance._client, "_nth_client_id", 0)
         return 0
 
     def session(self, labels=None, database_role=None):

packages/google-cloud-spanner/google/cloud/spanner_v1/_async/snapshot.py

@@ -31,7 +31,6 @@
 from google.cloud.spanner_v1._async._helpers import _retry
 from google.cloud.spanner_v1._async.streamed import StreamedResultSet
 from google.cloud.spanner_v1._helpers import (
-    AtomicCounter,
     _augment_error_with_request_id,
     _check_rst_stream_error,
     _make_value_pb,
@@ -760,20 +759,20 @@ async def partition_read(
             MetricsCapture(self._resource_info),
         ):
             nth_request = getattr(database, "_next_nth_request", 0)
-            attempt = AtomicCounter()
+            attempt = 0
 
             async def attempt_tracking_method():
+                nonlocal attempt
+                attempt += 1
                 all_metadata = database.metadata_with_request_id(
-                    nth_request, attempt.increment(), metadata, span
+                    nth_request, attempt, metadata, span
                 )
-                partition_read_method = functools.partial(
-                    api.partition_read,
+                return await api.partition_read(
                     request=partition_read_request,
                     metadata=all_metadata,
                     retry=retry,
                     timeout=timeout,
                 )
-                return await partition_read_method()
 
             response = await _retry(
                 attempt_tracking_method,
@@ -843,20 +842,20 @@ async def partition_query(
             MetricsCapture(self._resource_info),
         ):
             nth_request = getattr(database, "_next_nth_request", 0)
-            attempt = AtomicCounter()
+            attempt = 0
 
             async def attempt_tracking_method():
+                nonlocal attempt
+                attempt += 1
                 all_metadata = database.metadata_with_request_id(
-                    nth_request, attempt.increment(), metadata, span
+                    nth_request, attempt, metadata, span
                 )
-                partition_query_method = functools.partial(
-                    api.partition_query,
+                return await api.partition_query(
                     request=partition_query_request,
                     metadata=all_metadata,
                     retry=retry,
                     timeout=timeout,
                 )
-                return await partition_query_method()
 
             response = await _retry(
                 attempt_tracking_method,
@@ -917,22 +916,22 @@ async def _begin_transaction(
             MetricsCapture(self._resource_info),
         ):
             nth_request = getattr(database, "_next_nth_request", 0)
-            attempt = AtomicCounter()
+            attempt = 0
 
             async def wrapped_method():
+                nonlocal attempt
+                attempt += 1
                 begin_transaction_request = BeginTransactionRequest(
                     **begin_request_kwargs
                 )
                 call_metadata, error_augmenter = database.with_error_augmentation(
-                    nth_request, attempt.increment(), metadata, span
-                )
-                begin_transaction_method = functools.partial(
-                    api.begin_transaction,
-                    request=begin_transaction_request,
-                    metadata=call_metadata,
+                    nth_request, attempt, metadata, span
                 )
                 with error_augmenter:
-                    return await begin_transaction_method()
+                    return await api.begin_transaction(
+                        request=begin_transaction_request,
+                        metadata=call_metadata,
+                    )
 
             async def before_next_retry(nth_retry, delay_in_seconds):
                 add_span_event(