Python performance integration guide

Step-by-step instructions for adding performance monitoring and distributed tracing to your Python projects.

BugSnag’s serverside performance monitoring leverages OpenTelemetry. With the wealth of open source OpenTelemetry instrumentation available for Python, you can easily send spans and traces from your service to BugSnag by installing the OpenTelemetry SDK and completing some simple configuration.

OpenTelemetry SDK installation

The easiest way to get your app instrumented is to run it inside the opentelemetry-instrument wrapper. This detects which libraries and frameworks your app is using and automatically instruments them with by manipulating the bytecode.

To automatically instrument your Python app in this way, first install the required packages:

pip install opentelemetry-distro opentelemetry-exporter-otlp

Then use the opentelemetry-bootstrap tool to get the relevant instrumentation libraries, based on the packages your application is using:

opentelemetry-bootstrap >> requirements.txt
pip install -r requirements.txt

Or if you are not using requirements.txt, install the libraries directly:

opentelemetry-bootstrap --action=install

If you would prefer to have more visibility and control over what instrumentation is being added, you can configure instrumentation in code, rather than using opentelemetry-instrument. To do this see the OpenTelemetry Manual Instrumentation guide.

Configuration for BugSnag

Next you need to set some configuration:

export OTEL_EXPORTER_OTLP_ENDPOINT="https://<PROJECT_API_KEY>.otlp.bugsnag.com:4318" 
export OTEL_EXPORTER_METRICS_EXPORTER="none"
export OTEL_EXPORTER_TRACES_EXPORTER="otlp,console"
export OTEL_SERVICE_NAME="your-service-name" 
export OTEL_RESOURCE_ATTRIBUTES="deployment.environment=<RELEASE_STAGE>,service.version=<APP_VERSION>" 

• <PROJECT_API_KEY> can be found in project settings in the BugSnag dashboard.
• OTEL_SERVICE_NAME should uniquely identify your service. We recommend using the same name as your project name.
• <RELEASE_STAGE> and <APP_VERSION> must be the same as you are passing to your BugSnag Error SDK. They are case sensitive.

Traces can be received by BugSnag via either gRPC or HTTP (protobuf or JSON). In most cases the simplest way to send traces to BugSnag is to export an environment variable with your BugSnag project’s dedicated OpenTelemetry endpoint before running your OpenTelemetry instrumented app:

export OTEL_EXPORTER_OTLP_PROTOCOL="grpc"
export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT="https://<PROJECT_API_KEY>.otlp.bugsnag.com:4317"

export OTEL_EXPORTER_OTLP_PROTOCOL="http/protobuf"
export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT="https://<PROJECT_API_KEY>.otlp.bugsnag.com:4318/v1/traces"

Finally, running your app via opentelemetry-instrument will start sending span information to BugSnag:

opentelemetry-instrument python app.py

Sampling

Spans sent from OpenTelemetry SDKs use up your unmanaged quota. No spans will be ingested by BugSnag once your daily quota is exhausted. To ensure you have span coverage throughout the day, we recommend you use sampling.

By default the OpenTelemetry SDK will use the parentbased_always_on sampler, which will sample according to the sample decision from the incoming network request if there is one, and always sample if there is not. To use this requires clients of your Python server be instrumented in such a way that they pass their sampling information.

To apply a fixed sampling rate to all your Python spans, you can use the traceidratio sampler:

export OTEL_TRACES_SAMPLER="traceidratio"
export OTEL_TRACES_SAMPLER_ARG="0.25" # 25% of your traces will be sampled

opentelemetry-instrument python app.py

Alternatively you can use the parentbased_traceidratio sampler. This will sample at a constant rate, unless the trace context was propagated from a client, in which case it will use the same sample decision that the client used:

export OTEL_TRACES_SAMPLER="parentbased_traceidratio"
export OTEL_TRACES_SAMPLER_ARG="0.25" # 25% of traces not started in a client will be sampled

opentelemetry-instrument python app.py

Collectors

An OpenTelemetry Collector is an open source component that OpenTelemetry users can host on their own infrastructure. It can receive telemetry data from multiple sources, process it in various ways, and send it on to multiple telemetry back-ends, including BugSnag.

An important use of Collectors is to enable tail-based sampling, where a trace can be inspected in its entirety before making a sampling decision, for example looking to see whether it contained any errors.

To learn about using a collector with BugSnag, see our dedicated page on using a collector.

Span batch size

The maximum payload size for BugSnag’s OpenTelemetry trace endpoints is 1MB. OpenTelemetry payload sizes can be controlled via the batch size in the SDK or collector configuration.

The appropriate batch size will largely depend on the number of attributes getting added to your spans and therefore you may need to experiment to find the appropriate setting. The batch size can be controlled by setting an environment variable; we generally recommend 200 as a good starting point, which is lower than the default of 512 spans per batch.

export OTEL_BSP_MAX_EXPORT_BATCH_SIZE=200

Custom spans

For higher resolution information into the performance of your application, you can create your own spans whenever you like. To do this, get a tracer:

from opentelemetry import trace

tracer = trace.get_tracer("myapp")

Then wrap some work in a span:

with tracer.start_as_current_span("my-operation") as span:
    print("some interesting work")
    # span closes at end of with statement, timing the operation

For more information see the Python OpenTelemetry instrumentation documentation.

If you would like BugSnag to aggregate your custom span to provide you with summary statistics on its performance, then you need to set a bugsnag.span.first_class attribute on the span. BugSnag will then automatically create a grouping for that span based on its name. This will appear under the ‘Custom Spans’ tab of your Performance dashboard.

span.set_attribute("bugsnag.span.first_class", True)