autometrics

Breaking changes

• opentelemetry has been updated to v0.24 (#179)
  New feature flags: opentelemetry-0_24, exemplars-tracing-opentelemetry-0_25
  Removed feature flags: opentelemetry-0_21, exemplars-tracing-opentelemetry-0_22

If you are using these metrics library separately in your application in addition to Autometrics, ensure that you match the version that Autometrics uses.

• Update http to 1.0. This fixes compatibility with axum 0.7 (#167)
• Explicitly set default timeout and period for the OTEL push exporter (#168)
• Fix a regression that made all compilation errors in instrumented code appear as located in the #[autometrics] annotation instead of the original location (#170)
• Add missing feature flag for otel-push-exporter (#172)

Breaking changes

• opentelemetry has been updated to v0.21 (#159)
  New feature flags: opentelemetry-0_21, exemplars-tracing-opentelemetry-0_22
  Removed feature flags: opentelemetry-0_20, exemplars-tracing-opentelemetry-0_20, exemplars-tracing-opentelemetry-0_21
• prometheus-client has been updated to v0.22 (#160)
  New feature flag: prometheus-client-0_22
  Removed feature flag: prometheus-client-0_21

If you are using these metrics separately in your application in addition to Autometrics, ensure that you match the version that Autometrics uses.

General changes

• New exporter, otel_push_exporter is now available in addition to the existing prometheus_exporter. It can be used to push metrics in the OTEL format via HTTP or gRPC to a OTEL-collector-(compatible) server.
• Update to syn v2 (#145)
• Add support for tracing-opentelemetry v0.21 (#147)
• Error messages when using #[autometrics] together with #[async_trait] has been improved (#149)
• Fixed missing feature flag for opentelemetry-otlp when autometrics feature otel-push-exporter is enabled
• Fixed incorrect duration being recorded when using #[async_trait] together with #[autometrics] (#161)
  Please note that the #[autometrics] macro needs to be defined BEFORE #[async_trait].
• Fixed value of the result label being empty when the function is annotated with #[async_trait] (#161)
• Fixed #[async_trait] not being correctly detected if its re-exported in another crate (#164)

Autometrics 1.0 compliance

The Autometrics specification v1.0.0 has been released. The following changes are made in order to bring the release in compliance with the spec:

• Added repository.url and repository.provider as build_info labels. They can be configured either through environment variables or automatically based on your Cargo.toml package.repository value (only GitHub, GitLab and BitBucket) (#152)
• Added autometrics.version as build_info label, specifying the autometrics spec version that we are targeting. It is currently hardcoded to 1.0.0 (#154)
• Environment variables for repository.url and repository.provider will now be read at both compile time and runtime (#156)

Added

• Autometrics settings can now be configured via settings::AutometricsSettings::builder()
• A custom Registry can be used to collect metrics. This may be used to add custom metrics alongside those produced by Autometrics, as well as to export the metrics without using the provided prometheus_exporter
• All metrics now have a service.name label attached. This is set via the settings, via runtime environment variable (AUTOMETRICS_SERVICE_NAME or OTEL_SERVICE_NAME), or it falls back to the crate name defined in the Cargo.toml
• Function counters are initialized to zero in debug builds. This exposes details of instrumented functions to Prometheus and visualization tools built on top of it, before the functions have been called.
• Basic benchmarking

Changed

• Renamed the function.calls.count metric to function.calls (which is exported to Prometheus as function_calls_total) to be in line with OpenTelemetry and OpenMetrics naming conventions. Dashboards and alerting rules must be updated.
• When the function.calls.duration histogram is exported to Prometheus, it now includes the units (function_calls_duration_seconds) to be in line with Prometheus/OpenMetrics naming conventions. Dashboards and alerting rules must be updated.
• Struct methods now have the struct name as part of the function label on the metrics (for example, MyStruct::my_method)
• The caller label on the function.calls metric was replaced with caller.function and caller.module
• The custom-objective-latency feature can now be used with the prometheus-exporter, as well as with the prometheus and prometheus-client crates, because the histogram buckets can now be configured via the settings
• Upgraded opentelemetry to v0.20

Deprecated

• metrics feature flag (replaced with metrics-0_21)
• opentelemetry feature flag (replaced with opentelemetry-0_20)
• prometheus feature flag (replaced with prometheus-0_13)
• prometheus-client feature flag (replaced with prometheus-client-0_21)
• exemplars-tracing-opentelemetry feature flag (replaced with exemplars-tracing-opentelemetry-0_20)

Removed

• encode_global_metrics was removed and replaced by prometheus_exporter::encode_to_string
• global_metrics_exporter was removed and replaced by prometheus_exporter::init
• backends::prometheus_client::REGISTRY was removed. The Registry used with the prometheus-client backend can now be accessed via AutometricsSettings::prometheus_client_registry

Fixed

• Return types on functions annotated with #[autometrics] containing generic impl types in their type arguments (fn() -> Result<impl ToString, impl std::error::Error>) no longer fail to compile.

Added

• Support the official prometheus-client crate for producing metrics
• Support exemplars when using the feature flags exemplars-tracing or exemplars-tracing-opentelemetry. Autometrics can now extract fields from the current span and attach them as exemplars on the counter and histogram metrics
• ResultLabels derive macro allows to specify on an enum whether variants should always be "ok", or "error" for the success rate metrics of functions using them. (#61)
• The prometheus_exporter module contains all functions related to the prometheus-exporter feature
• prometheus_exporter::encode_http_response function returns an http::Response with the metrics. This is especially recommended when using exemplars, because it automatically uses the OpenMetrics Content-Type header, which is required for Prometheus to scrape metrics with exemplars
• AUTOMETRICS_DISABLE_DOCS environment variable can be set to disable doc comment generation (this is mainly for use with editor extensions that generate doc comments themselves)

Changed

• Users must configure the metrics library they want to use autometrics with, unless the prometheus-exporter feature is enabled. In that case, the official prometheus-client will be used
• Metrics library feature flags are now mutually exclusive (previously, autometrics would only produce metrics using a single metrics library if multiple feature flags were enabled, using a prioritization order defined internally)
• GetLabels trait (publicly exported but meant for internal use) changed the signature of its function to accomodate the new ResultLabels macro. This change is not significant if you never imported autometrics::__private manually (#61)
• When using the opentelemetry together with the prometheus-exporter, it will no longer use the default registry provided by the prometheus crate. It will instead use a new registry
• Updated opentelemetry dependencies to v0.19. This means that users using autometrics with opentelemetry but not using the prometheus-exporter must update the opentelemetry to use v0.19.

Deprecated

• global_metrics_exporter and encode_global_metrics have been deprecated and replaced by prometheus_exporter::init and prometheus_exporter::encode_to_string, respectively

Removed

• opentelemetry is no longer used by default
• GetLabelsForResult trait (publicly exported but meant for internal use) was removed to accomodate the new ResultLabels macro. This change is not significant if you never imported autometrics::__private manually (#61)

Fixed

• #[autometrics] now works on functions that use type inference in their return statement (#74, #61)

Changed

• Overhaul documentation

Fixed

• Generated latency query needs additional labels to show lines for 95th and 99th percentiles

Added

• build_info metric tracks software version and commit
• Queries now use build_info metric to correlate version info with problems

Fixed

• Prometheus rules handle when the counter metric name has a _total suffix

Added

• Alerts have minimum traffic threshold of 1 request / minute

Fixed

• Latency SLO total query in alerting rules