What's Changed

• Replace pin-project with pin-project-lite by @FalkWoldmann in #80
• Bump metrics-exporter-prometheus version to 0.17 by @raimannma in #84
• feat: allow patterns by @erwanncloarec in #86
• bump various outdated dependencies by @Ptrskay3 in #87

New Contributors

• @raimannma made their first contribution in #84
• @erwanncloarec made their first contribution in #86

Full Changelog: 0.8.0...0.9.0

0.8.0

Changed

• Compatibility with axum = "0.8". This also updates matchit to 0.8, changing how group patterns are described:
  for example, with_group_patterns_as("/foo", &["/foo/:bar"]) needs to be changed to with_group_patterns_as("/foo", &["/foo/{bar}"]).
  The metrics values are also impacted: for example, the value "/foo/:bar" is now "/foo/{bar}".
  This change bumps MSRV to 1.75. [#69]
• Disable the default features in metrics-exporter-prometheus to skip the binding of port 9000, and the upkeep task is manually spawned. [#75]
• Add a new feature http-listener to enable that exact feature in metrics-exporter-prometheus. [#79]
• Replace the once_cell dependency with std::sync::OnceLock [#78]

Fixed

• Fixed the long-standing pending requests metric leak. [#74]
• Removed the sideeffect of binding port 9000. [#75]

Changed

• MakeDefaultHandle::make_default_handle now takes self as argument. This allows custom implementor structs to hold non-static data. [#49]
• Change the default initialization of PrometheusHandle to prevent unbounded memory growth of histograms. [#52]
• Bump metrics to 0.23, metrics-exporter-prometheus to 0.15. [#52]
• Document MSRV as 1.70 currently. [#52]

Added

• GenericMetricLayer::pair_from to initialize from a concrete struct. GenericMetricLayer::pair now requires that the handle type implements Default. [#49]
• BaseMetricLayer that serves a more lightweight alternative to GenericMetricLayer. [#56]

What's Changed

• Update metrics-exporter-prometheus to 0.14.0 by @gauravkumar37 in #50
• MakeDefaultHandle and other API improvements by @Ptrskay3 in #49
• Change default initialization of PrometheusHandle by @Ptrskay3 in #52
• A backend-agnostic layer by @Ptrskay3 in #56
• Briefly document push gateway mode by @Ptrskay3 in #61
• release/0.7.0 by @Ptrskay3 in #62

New Contributors

• @gauravkumar37 made their first contribution in #50

Full Changelog: 0.6.1...0.7.0

• Disabled the "push-gateway" feature in metrics-exporter-prometheus by default, and added a way to enable it via
  the same name under axum_prometheus. This change ensures that this crate can still be built without openssl support, see here. [#44]
• Update examples to metrics-exporter-prometheus to 0.13 and metrics to 0.22. [#43]

• Update metrics-exporter-prometheus to 0.13 and metrics to 0.22 #39

Added

• Support for response body size metric, which can be turned on via PrometheusMetricLayerBuilder::enable_response_body_size. [#33]
• All metrics now are initialized via the metrics::describe_* functions by default, but can be turned off with PrometheusMetricLayerBuilder::no_initialize_metrics if you wish to do it manually. [#33]
• Compatibility with http-body = "1.0" and axum = "0.7". [#36]

Changed

• The lower-level Lifecycle API has changed: separated the OnBodyChunk trait, which is ran when a response body chunk has been generated. [#33]

Added

• Support for different exporters than Prometheus. Developers now allowed to use their own exporter - as long as it's using the metrics.rs ecosystem. This is meant to be a non-breaking change - if you're using Prometheus, you shouldn't notice any changes in the public API. If you do however, please file an issue! #28
• An example showcasing StatsD exporter #28
• Simple snapshot tests #28
• Utility functions to get metric names at runtime #28

Fixed

• Previous attempts to fix PrometheusMetricBuilder::with_prefix in 0.3.4 were not complete, this is now fully addressed. #28

Fixed

• PrometheusMetricBuilder::with_prefix is now properly setting the metric prefix, and the metric handle also takes that prefix into account. Previously the metric initialization incorrectly ignored the prefix, which caused the requests duration histogram to use quantile instead of le labels. [#26]. (Thanks @yanns!)

• Update metrics-exporter-prometheus to 0.12 and metrics to 0.21.

Added

• The status code of the response is now captured in the total requests counter metric.

Added

• with_prefix to PrometheusMetricLayerBuilder, which can be used to rename the default prefix (axum) for all metrics. This is especially useful when working with cargo workspaces that has more than one axum_prometheus instance (since environment variables don't work there).

0.3.0 - 2023-01-04

Added

• Routing patterns can be ignored, and grouped together when reporting to Prometheus.

• Endpoint label behavior can be altered with the new EndpointLabel enum.

• Added a new builder PrometheusMetricLayerBuilder to easily customize these.

  let (prometheus_layer, metric_handle) = PrometheusMetricLayerBuilder::new()
      // ignore reporting requests that match "/foo" or "/sensitive"
      .with_ignore_patterns(&["/foo", "/sensitive"])
      // if the any of the second argument matches, report them at the `/bar` endpoint
      .with_group_patterns_as("/bar", &["/foo/:bar", "/foo/:bar/:baz"])
      // use `axum::extract::MatchedPath`, and if that fails, use the exact requested URI
      .with_endpoint_label_type(EndpointLabel::MatchedPath)
      .with_default_metrics()
      .build_pair();
  

• A builder-example and an endpoint-type-example.

• The metric names can be changed by setting some environmental variables at compile time. It is best to set these in the config.toml (note this is not the same file as Cargo.toml):

  [env]
  AXUM_HTTP_REQUESTS_TOTAL = "my_app_requests_total"
  AXUM_HTTP_REQUESTS_DURATION_SECONDS = "my_app_requests_duration_seconds"
  AXUM_HTTP_REQUESTS_PENDING = "my_app_requests_pending"
  

0.2.0 - 2022-10-25

Added

• Compatibility with axum-core = "0.3" and thus axum = "0.6".

0.1.0

First version.