[Monitor OpenTelemetry Exporter] Add client-side token bucket rate limiter for telemetry export

sdk/monitor/azure-monitor-opentelemetry-exporter/azure/monitor/opentelemetry/exporter/export/_base.py

@@ -44,12 +44,18 @@
     DropCode,
     _exception_categories,
 )
-from azure.monitor.opentelemetry.exporter._connection_string_parser import ConnectionStringParser
+from azure.monitor.opentelemetry.exporter._connection_string_parser import (
+    ConnectionStringParser,
+)
 from azure.monitor.opentelemetry.exporter._storage import LocalFileStorage
 from azure.monitor.opentelemetry.exporter._utils import (
     _get_auth_policy,
     _get_sha256_hash,
 )
+from azure.monitor.opentelemetry.exporter.export._rate_limiter import (
+    _TokenBucketRateLimiter,
+    _DEFAULT_MAX_ENVELOPES_PER_SECOND,
+)
 from azure.monitor.opentelemetry.exporter.statsbeat._state import (
     get_statsbeat_initial_success,
     get_statsbeat_shutdown,
@@ -70,7 +76,6 @@
     get_customer_stats_manager,
 )
 
-
 logger = logging.getLogger(__name__)
 
 _AZURE_TEMPDIR_PREFIX = "Microsoft-AzureMonitor-"
@@ -86,6 +91,7 @@ class ExportResult(Enum):
 
 # pylint: disable=broad-except
 # pylint: disable=too-many-instance-attributes
+# pylint: disable=too-many-statements
 # pylint: disable=C0301
 class BaseExporter:
     """Azure Monitor base exporter for OpenTelemetry."""
@@ -98,6 +104,7 @@ def __init__(self, **kwargs: Any) -> None:
         :keyword ManagedIdentityCredential/ClientSecretCredential credential: Token credential, such as ManagedIdentityCredential or ClientSecretCredential, used for Azure Active Directory (AAD) authentication. Defaults to None.
         :keyword bool disable_offline_storage: Determines whether to disable storing failed telemetry records for retry. Defaults to `False`.
         :keyword str storage_directory: Storage path in which to store retry files. Defaults to `<tempfile.gettempdir()>/opentelemetry-python-<your-instrumentation-key>`.
+        :keyword int max_envelopes_per_second: Maximum number of telemetry envelopes sent per second. Acts as a client-side safety cap to prevent overloading shared ingestion infrastructure during telemetry bursts. Defaults to 10000. Set to 0 to disable rate limiting.
         :rtype: None
         """
         parsed_connection_string = ConnectionStringParser(kwargs.get("connection_string"))
@@ -139,6 +146,17 @@ def __init__(self, **kwargs: Any) -> None:
             "storage_retention_period", 48 * 60 * 60
         )  # Retention period in seconds (default 48 hrs)
         self._timeout = kwargs.get("timeout", 10.0)  # networking timeout in seconds
+        max_eps = kwargs.get("max_envelopes_per_second", _DEFAULT_MAX_ENVELOPES_PER_SECOND)
+        if max_eps is not None and max_eps < 0:
+            raise ValueError("max_envelopes_per_second must be non-negative (0 disables rate limiting)")
+        # Each exporter instance gets its own rate limiter. This is intentional:
+        # different telemetry types (traces, logs, metrics) have different
+        # ingestion characteristics and burst profiles, so per-exporter caps
+        # provide more predictable behaviour than a shared process-wide bucket.
+        if max_eps and max_eps > 0:
+            self._rate_limiter: Optional[_TokenBucketRateLimiter] = _TokenBucketRateLimiter(max_eps)
+        else:
+            self._rate_limiter = None
         self._distro_version = kwargs.get(
             _AZURE_MONITOR_DISTRO_VERSION_ARG, ""
         )  # If set, indicates the exporter is instantiated via Azure monitor OpenTelemetry distro. Versions corresponds to distro version.
@@ -168,7 +186,10 @@ def __init__(self, **kwargs: Any) -> None:
             policies.append(config.http_logging_policy or HttpLoggingPolicy(**kwargs))
 
         self.client: AzureMonitorClient = AzureMonitorClient(
-            host=self._endpoint, connection_timeout=self._timeout, policies=policies, **kwargs
+            host=self._endpoint,
+            connection_timeout=self._timeout,
+            policies=policies,
+            **kwargs,
         )
         # TODO: Uncomment configuration changes once testing is completed
         # if self._configuration_manager:
@@ -195,7 +216,9 @@ def __init__(self, **kwargs: Any) -> None:
         if self._should_collect_stats():
             try:
                 # Import here to avoid circular dependencies
-                from azure.monitor.opentelemetry.exporter.statsbeat._statsbeat import collect_statsbeat_metrics
+                from azure.monitor.opentelemetry.exporter.statsbeat._statsbeat import (
+                    collect_statsbeat_metrics,
+                )
 
                 collect_statsbeat_metrics(self)
             except Exception as e:  # pylint: disable=broad-except
@@ -205,7 +228,9 @@ def __init__(self, **kwargs: Any) -> None:
 
         # customer sdkstats initialization
         if self._should_collect_customer_sdkstats():
-            from azure.monitor.opentelemetry.exporter.statsbeat.customer import collect_customer_sdkstats
+            from azure.monitor.opentelemetry.exporter.statsbeat.customer import (
+                collect_customer_sdkstats,
+            )
 
             # Collect customer sdkstats metrics
             collect_customer_sdkstats(self)
@@ -260,6 +285,39 @@ def _transmit(self, envelopes: List[TelemetryItem]) -> ExportResult:
         :rtype: ~azure.monitor.opentelemetry.exporter.export._base._ExportResult
         """
         if len(envelopes) > 0:
+            # Client-side rate limiting: cap send rate to protect shared ingestion infrastructure.
+            # Stats exporters bypass rate limiting to ensure observability data is not lost.
+            if self._rate_limiter and not self._is_stats_exporter() and not self._is_customer_sdkstats_exporter():
+                granted = self._rate_limiter.try_consume(len(envelopes))
+                if granted == 0:
+                    logger.warning(
+                        "Rate limiter rejected entire batch of %d envelopes. Routing to local storage for retry.",
+                        len(envelopes),
+                    )
+                    return ExportResult.FAILED_RETRYABLE
+                if granted < len(envelopes):
+                    # Send what we can, route the rest to local storage.
+                    # We mutate the list in-place so that the caller's reference
+                    # (used later in _handle_transmit_from_storage) only sees
+                    # the admitted envelopes, preventing double-persist of the
+                    # overflow on a subsequent retryable failure.
+                    overflow = envelopes[granted:]
+                    del envelopes[granted:]
+                    logger.info(
+                        "Rate limiter admitted %d of %d envelopes; %d envelopes deferred to local storage.",
+                        granted,
+                        granted + len(overflow),
+                        len(overflow),
+                    )
+                    if self.storage:
+                        self.storage.put([x.as_dict() for x in overflow])
+                    else:
+                        logger.warning(
+                            "Rate limiter deferred %d envelopes but offline "
+                            "storage is disabled; these envelopes are dropped.",
+                            len(overflow),
+                        )
+
             result = ExportResult.SUCCESS
             # Track whether or not exporter has successfully reached ingestion
             # Currently only used for statsbeat exporter to detect shutdown cases
@@ -295,7 +353,7 @@ def _transmit(self, envelopes: List[TelemetryItem]) -> ExportResult:
                                 logger.info(
                                     "Data dropped due to ingestion sampling: %s %s.",
                                     error.message,
-                                    envelopes[error.index] if error.index is not None else "",
+                                    (envelopes[error.index] if error.index is not None else ""),
                                 )
                         elif _is_retryable_code(error.status_code):
                             resend_envelopes.append(envelopes[error.index])  # type: ignore
@@ -317,7 +375,7 @@ def _transmit(self, envelopes: List[TelemetryItem]) -> ExportResult:
                                     "Data drop %s: %s %s.",
                                     error.status_code,
                                     error.message,
-                                    envelopes[error.index] if error.index is not None else "",
+                                    (envelopes[error.index] if error.index is not None else ""),
                                 )
                     if self.storage and resend_envelopes:
                         envelopes_to_store = [x.as_dict() for x in resend_envelopes]
@@ -399,7 +457,9 @@ def _transmit(self, envelopes: List[TelemetryItem]) -> ExportResult:
                             # Track dropped items in customer sdkstats, non-retryable scenario
                             if self._should_collect_customer_sdkstats():
                                 track_dropped_items(
-                                    envelopes, DropCode.CLIENT_EXCEPTION, _exception_categories.CLIENT_EXCEPTION.value
+                                    envelopes,
+                                    DropCode.CLIENT_EXCEPTION,
+                                    _exception_categories.CLIENT_EXCEPTION.value,
                                 )
                             logger.error(
                                 "Error sending telemetry because of circular redirects. "
@@ -461,7 +521,9 @@ def _transmit(self, envelopes: List[TelemetryItem]) -> ExportResult:
                 # Track dropped items in customer sdkstats for general exceptions
                 if self._should_collect_customer_sdkstats():
                     track_dropped_items(
-                        envelopes, DropCode.CLIENT_EXCEPTION, _exception_categories.CLIENT_EXCEPTION.value
+                        envelopes,
+                        DropCode.CLIENT_EXCEPTION,
+                        _exception_categories.CLIENT_EXCEPTION.value,
                     )
 
                 if self._should_collect_stats():
@@ -595,7 +657,9 @@ def _is_sampling_rejection(message: Optional[str]) -> bool:
 
 
 # mypy: disable-error-code="union-attr"
-def _get_authentication_credential(**kwargs: Any) -> Optional[ManagedIdentityCredential]:
+def _get_authentication_credential(
+    **kwargs: Any,
+) -> Optional[ManagedIdentityCredential]:
     if "credential" in kwargs:
         return kwargs.get("credential")
     auth_string = os.getenv(_APPLICATIONINSIGHTS_AUTHENTICATION_STRING, "")
@@ -612,7 +676,9 @@ def _get_authentication_credential(**kwargs: Any) -> Optional[ManagedIdentityCre
                 return credential
     except ValueError as exc:
         logger.error(  # pylint: disable=do-not-log-exceptions-if-not-debug
-            "APPLICATIONINSIGHTS_AUTHENTICATION_STRING, %s, has invalid format: %s", auth_string, exc
+            "APPLICATIONINSIGHTS_AUTHENTICATION_STRING, %s, has invalid format: %s",
+            auth_string,
+            exc,
         )
     except Exception as e:
         logger.error(  # pylint: disable=do-not-log-exceptions-if-not-debug
@@ -675,6 +741,8 @@ def _safe_psutil_call(func, default=""):
     )
     subdirectory = _get_sha256_hash(hash_input)
     storage_directory = os.path.join(
-        shared_root, _AZURE_TEMPDIR_PREFIX + subdirectory, _TEMPDIR_PREFIX + instrumentation_key
+        shared_root,
+        _AZURE_TEMPDIR_PREFIX + subdirectory,
+        _TEMPDIR_PREFIX + instrumentation_key,
     )
     return storage_directory

sdk/monitor/azure-monitor-opentelemetry-exporter/azure/monitor/opentelemetry/exporter/export/_rate_limiter.py

@@ -0,0 +1,64 @@
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+
+import logging
+import threading
+import time
+
+logger = logging.getLogger(__name__)
+
+# Default maximum envelopes per second across all telemetry types.
+# This is a client-side safety cap to prevent self-inflicted overload
+# of shared ingestion infrastructure during telemetry bursts.
+_DEFAULT_MAX_ENVELOPES_PER_SECOND = 10000
+
+# Minimum allowed value to prevent misconfiguration
+_MIN_MAX_ENVELOPES_PER_SECOND = 1
+
+
+class _TokenBucketRateLimiter:
+    """Thread-safe token bucket rate limiter for outbound telemetry.
+
+    The bucket refills at ``max_per_second`` tokens per second and holds
+    at most ``max_per_second`` tokens (i.e. one second of burst capacity).
+
+    :param float max_per_second: Maximum tokens (envelopes) allowed per second.
+    """
+
+    def __init__(self, max_per_second: float) -> None:
+        if max_per_second < _MIN_MAX_ENVELOPES_PER_SECOND:
+            raise ValueError(f"max_per_second must be at least {_MIN_MAX_ENVELOPES_PER_SECOND}")
+        self._max_per_second = float(max_per_second)
+        self._tokens = self._max_per_second  # start full
+        self._last_refill = time.monotonic()
+        self._lock = threading.Lock()
+
+    def try_consume(self, count: int) -> int:
+        """Try to consume *count* tokens from the bucket.
+
+        Returns the number of tokens actually consumed (i.e. how many
+        envelopes may be sent).  The caller should handle the remainder
+        (e.g. store for retry or drop).
+
+        :param int count: Number of tokens requested.
+        :return: Number of tokens granted (<= *count*).
+        :rtype: int
+        """
+        if count <= 0:
+            return 0
+
+        with self._lock:
+            now = time.monotonic()
+            elapsed = now - self._last_refill
+            self._last_refill = now
+
+            # Refill tokens based on elapsed time, capped at bucket capacity
+            self._tokens = min(
+                self._max_per_second,
+                self._tokens + elapsed * self._max_per_second,
+            )
+
+            granted = min(count, int(self._tokens))
+            self._tokens -= granted
+
+        return granted