Web Frameworks¶

Here are some tips for instrumenting your web applications.

Integrations¶

If you're using one of the following libraries, check out the integration docs:

Otherwise, check if your server uses WSGI or ASGI and check the corresponding integration.

Capturing HTTP server request and response headers¶

Some methods (e.g. logfire.instrument_fastapi()) allow you to pass capture_headers=True to record all request and response headers in the spans, and that's all you usually need.

If you want more control, there are three environment variables to tell the OpenTelemetry instrumentation libraries to capture request and response headers:

OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_REQUEST
OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_RESPONSE
OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SANITIZE_FIELDS

Each accepts a comma-separated list of regexes which are checked case-insensitively against header names. The first two determine which request/response headers are captured and added to span attributes. The third determines which headers will have their values redacted.

For example, to capture all headers, set the following:

OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_REQUEST=".*"
OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_RESPONSE=".*"

(this is what capture_headers=True does)

To specifically capture the content-type request header and request headers starting with X-:

OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_REQUEST="content-type,X-.*"

To replace the Authorization header value with [REDACTED] to avoid leaking user credentials:

OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SANITIZE_FIELDS="Authorization"

(although usually it's better to rely on Logfire's scrubbing feature)

Query HTTP requests duration per percentile¶

It's usually interesting to visualize HTTP requests duration per percentile. Instead of having an average, which may be influenced by extreme values, percentiles allow us know the maximum duration for 50%, 90%, 95% or 99% of the requests.

Here is a sample query to compute those percentiles for HTTP requests duration:

WITH dataset AS (
  SELECT
    time_bucket('%time_bucket_duration%', start_timestamp) AS x,
    (extract(epoch from end_timestamp - start_timestamp) * 1000) as duration_ms
  FROM records
  WHERE attributes ? 'http.method'
)
SELECT
  x,
  percentile_cont(0.99) WITHIN GROUP (ORDER BY duration_ms) as percentile_99,
  percentile_cont(0.95) WITHIN GROUP (ORDER BY duration_ms) as percentile_95,
  percentile_cont(0.90) WITHIN GROUP (ORDER BY duration_ms) as percentile_90,
  percentile_cont(0.50) WITHIN GROUP (ORDER BY duration_ms) as percentile_50
FROM dataset
GROUP BY x
ORDER BY x DESC;

Notice how we filtered on records that have the http.method attributes set. It's a good starting point to retrieve traces that are relevant for HTTP requests, but depending on your setup, you might need to add more filters.

This query is a good candidate for a Time Series chart in a dashboard: