opentelemetry-sdk-extension-prometheus-multiprocess 1.1.1

No project description provided

Workaround to use prometheus_client multiprocess mode (for e.g. gunicorn, celery) with opentelemetry, where this model currently is not supported, and not even employing an additional otel collector currently solves this.

This package provides a customized MeterProvider, that uses the native prometheus_client metrics implementations, and thus their mechanisms for exporting/exposition (and not the opentelemetry ones).

Usage

• See Multiprocess Mode Documentation for details on how to operate this.

• Set environment variable PROMETHEUS_MULTIPROC_DIR to a directory where prometheus_client can store its metrics state files (that are shared between processes via mmap).

• Configure opentelemetry meter provider:

  from opentelemetry.sdk.extension.prometheus_multiprocess import PrometheusMeterProvider
  import opentelemetry.metrics
  
  provider = PrometheusMeterProvider()
  opentelemetry.metrics.set_meter_provider(provider)

• Set up prometheus metrics exposition, for example:

  registry = prometheus_client.CollectorRegistry()
  prometheus_client.multiprocess.MultiProcessCollector(registry)
  prometheus_client.start_http_server(8080, registry=registry)

  (The package also provides a MultiProcessRegistry to streamline this setup.)

• Use opentelemetry metrics like normal:

  meter = opentelemetry.metrics.get_meter('mypackage', 'myversion')
  HTTP_DURATION = metrics.create_histogram('http.client.duration', unit='ms')
  
  from timeit import default_timer
  start = default_timer()
  # perform request here
  duration = (default_timer() - start) * 1000
  HTTP_DURATION.record(duration, {'http.response.status_code': 200})

Changelog

1.1.1 (2024-11-07)

Changes

• Accept kw to make otel API forward-compatible (fwcompat)

1.1.0 (2024-06-25)

Changes

• naming: Conform to prometheus metric name and unit conventions (naming)

1.0.0 (2024-04-26)

Changes

• Initial release (initial)