Add OpenTelemetry metrics exporter built on MetricBuffer

CHANGELOG.md

@@ -90,6 +90,10 @@ to include examples, links to docs, or any other relevant information.
   This lets types with transfer type converters delegate their wire representation to the
   configured payload converter, preserving SDK behavior such as serialization
   contexts.
+- Added `temporalio.contrib.opentelemetry.MetricsExporter`, which drains a
+  `temporalio.runtime.MetricBuffer` on a fixed interval and exports through a
+  real OpenTelemetry `MeterProvider`, giving SDK/Core metrics access to
+  standard OTel features (views, resource, exemplars). Experimental.
 - Added `TLSConfig.verification_server_name` to verify the server certificate against a fixed name
   instead of the connection's server name. Unlike `domain`, it does not change the TLS SNI or
   HTTP/2 authority values, which keep following the connected host, so it can be used when the

README.md

@@ -113,6 +113,7 @@ informal introduction to the features and their implementation.
     - [Observability](#observability)
       - [Metrics](#metrics)
       - [OpenTelemetry Tracing](#opentelemetry-tracing)
+      - [OpenTelemetry Metrics](#opentelemetry-metrics)
     - [Protobuf 3.x vs 4.x](#protobuf-3x-vs-4x)
     - [Known Compatibility Issues](#known-compatibility-issues)
       - [gevent Patching](#gevent-patching)
@@ -1990,6 +1991,40 @@ as an interceptor on the `interceptors` argument of `Client.connect`. When set,
 calls and for all activity and workflow invocations on the worker, spans will be created and properly serialized through
 the server to give one proper trace for a workflow execution.
 
+#### OpenTelemetry Metrics
+
+Metrics support also requires the `opentelemetry` extra (see above). If you want your Temporal SDK/Core metrics (and
+any custom metrics recorded via `activity.metric_meter()`/`workflow.metric_meter()`) to flow through the standard
+OpenTelemetry metrics pipeline (views, resources, any OTel-compatible backend) rather than only through
+`PrometheusConfig`/`OpenTelemetryConfig`, set a `temporalio.runtime.MetricBuffer` as the `metrics` on
+`TelemetryConfig`, then drain it into a real OpenTelemetry `MeterProvider` using
+`temporalio.contrib.opentelemetry.MetricsExporter`:
+
+```python
+from datetime import timedelta
+
+from opentelemetry.sdk.metrics import MeterProvider
+from opentelemetry.sdk.metrics.export import ConsoleMetricExporter, PeriodicExportingMetricReader
+
+from temporalio.client import Client
+from temporalio.contrib.opentelemetry import MetricsExporter
+from temporalio.runtime import MetricBuffer, Runtime, TelemetryConfig
+
+buffer = MetricBuffer(10_000)
+runtime = Runtime(telemetry=TelemetryConfig(metrics=buffer))
+meter_provider = MeterProvider(
+    metric_readers=[PeriodicExportingMetricReader(ConsoleMetricExporter(), export_interval_millis=5000)]
+)
+
+async with MetricsExporter(buffer, meter_provider):
+    client = await Client.connect("localhost:7233", runtime=runtime)
+    # ... run workers/workflows while the exporter drains the buffer in the background
+```
+
+`MetricsExporter` must be running (via `async with` or manual `run()`/`shutdown()`) for as long as metrics should be
+exported, since it works by draining the buffer on a fixed interval (`poll_interval`, default one second) -- per the
+warning on `MetricBuffer`, updates are dropped if the buffer isn't drained regularly.
+
 ### Protobuf 3.x vs 4.x
 
 Python currently has two somewhat-incompatible protobuf library versions - the 3.x series and the 4.x series. Python

pyproject.toml

@@ -26,7 +26,7 @@ classifiers = [
 
 [project.optional-dependencies]
 grpc = ["grpcio>=1.48.2,<2"]
-opentelemetry = ["opentelemetry-api>=1.11.1,<2", "opentelemetry-sdk>=1.11.1,<2"]
+opentelemetry = ["opentelemetry-api>=1.12.0,<2", "opentelemetry-sdk>=1.12.0,<2"]
 pydantic = ["pydantic>=2.0.0,<3"]
 openai-agents = ["openai-agents>=0.17.5", "mcp>=1.9.4, <2"]
 google-adk = ["google-adk>=2.2.0,<3", "mcp>=1.24,<2"]
@@ -38,8 +38,8 @@ deepagents = [
   "langchain-core>=1.4.8,<2; python_version >= '3.11'",
 ]
 lambda-worker-otel = [
-  "opentelemetry-api>=1.11.1,<2",
-  "opentelemetry-sdk>=1.11.1,<2",
+  "opentelemetry-api>=1.12.0,<2",
+  "opentelemetry-sdk>=1.12.0,<2",
   "opentelemetry-exporter-otlp-proto-grpc>=1.11.1,<2",
   "opentelemetry-semantic-conventions>=0.40b0,<1",
   "opentelemetry-sdk-extension-aws>=2.0.0,<3",

temporalio/contrib/opentelemetry/README.md

@@ -1,6 +1,29 @@
 # OpenTelemetry Integration for Temporal Python SDK
 
-This package provides OpenTelemetry tracing integration for Temporal workflows, activities, and other operations. It includes automatic span creation and propagation for distributed tracing across your Temporal applications.
+This package provides OpenTelemetry tracing and metrics integration for Temporal workflows, activities, and other operations. It includes automatic span creation and propagation for distributed tracing, and a `MetricsExporter` for exporting Temporal SDK/Core metrics through the standard OpenTelemetry metrics API, across your Temporal applications.
+
+## Metrics
+
+`MetricsExporter` drains a `temporalio.runtime.MetricBuffer` into an OpenTelemetry `MeterProvider`, so Temporal's own SDK/Core metrics (and any custom metrics recorded via `activity.metric_meter()`/`workflow.metric_meter()`) can be exported through the standard OpenTelemetry metrics pipeline (views, resources, any OTel-compatible backend) instead of only through `PrometheusConfig`/`OpenTelemetryConfig`.
+
+```python
+from opentelemetry.sdk.metrics import MeterProvider
+from opentelemetry.sdk.metrics.export import ConsoleMetricExporter, PeriodicExportingMetricReader
+from temporalio.contrib.opentelemetry import MetricsExporter
+from temporalio.runtime import MetricBuffer, Runtime, TelemetryConfig
+
+buffer = MetricBuffer(10_000)
+runtime = Runtime(telemetry=TelemetryConfig(metrics=buffer))
+meter_provider = MeterProvider(
+    metric_readers=[PeriodicExportingMetricReader(ConsoleMetricExporter())]
+)
+
+async with MetricsExporter(buffer, meter_provider):
+    client = await Client.connect("localhost:7233", runtime=runtime)
+    ...
+```
+
+**Note:** the `Runtime` must be constructed with the buffer attached *before* `MetricsExporter` is started, and the exporter must keep running (it polls on a fixed interval) for as long as metrics should be exported.
 
 ## Overview
 

temporalio/contrib/opentelemetry/__init__.py

@@ -1,14 +1,17 @@
 """OpenTelemetry v2 integration for Temporal SDK.
 
-This package provides OpenTelemetry tracing integration for Temporal workflows,
-activities, and other operations. It includes automatic span creation and
-propagation for distributed tracing.
+This package provides OpenTelemetry tracing and metrics integration for
+Temporal workflows, activities, and other operations. It includes automatic
+span creation and propagation for distributed tracing, and a
+:py:class:`MetricsExporter` for exporting Temporal SDK/Core metrics through
+the standard OpenTelemetry metrics API.
 """
 
 from temporalio.contrib.opentelemetry._interceptor import (
     TracingInterceptor,
     TracingWorkflowInboundInterceptor,
 )
+from temporalio.contrib.opentelemetry._metrics_exporter import MetricsExporter
 from temporalio.contrib.opentelemetry._otel_interceptor import OpenTelemetryInterceptor
 from temporalio.contrib.opentelemetry._plugin import OpenTelemetryPlugin
 from temporalio.contrib.opentelemetry._tracer_provider import create_tracer_provider
@@ -18,5 +21,6 @@
     "TracingWorkflowInboundInterceptor",
     "OpenTelemetryInterceptor",
     "OpenTelemetryPlugin",
+    "MetricsExporter",
     "create_tracer_provider",
 ]