Skip to content

Latest commit

 

History

History
952 lines (764 loc) · 51 KB

CHANGELOG.md

File metadata and controls

952 lines (764 loc) · 51 KB

Changelog

vNext

  • Bug fix: ObservableGauge returns data points recorded since previous collection, despite temporality. Other asynchronous (observable) instruments with Cumulative temporality behave as synchronous ones and return data points on every collection. #2213

  • Feature: Introduced a new feature flag, experimental_metrics_disable_name_validation, under the opentelemetry-sdk, which allows disabling the Instrument Name Validation. This is useful in scenarios where you need to use special characters, Windows Perf Counter Wildcard Path, or similar cases. For more details, check #2543.

    WARNING: While this feature provides flexibility, be cautious when using it, as platforms like Prometheus impose restrictions on metric names and labels (e.g., no spaces, capital letters, or certain special characters). Using invalid characters may result in compatibility issues or incorrect behavior. Ensure that instrument names comply with the requirements of your target platform to avoid potential problems.

  • Breaking(Affects custom metric exporter authors only) start_time and time is moved from DataPoints to aggregations (Sum, Gauge, Histogram, ExpoHistogram) see #2377 and #2411, to reduce memory.

  • Breaking start_time is no longer optional for Sum aggregation, see #2367, but is still optional for Gauge aggregation see #2389.

  • Breaking

    • SimpleLogProcessor modified to be generic over LogExporter to avoid dynamic dispatch to invoke exporter. If you were using with_simple_exporter to add LogExporter with SimpleLogProcessor, this is a transparent change. #2338
    • ResourceDetector.detect() no longer supports timeout option.
    • opentelemetry::global::shutdown_tracer_provider() removed from the API, should now use tracer_provider.shutdown() see #2369 for a migration example. "Tracer provider" is cheaply clonable, so users are encouraged to set a clone of it as the global (ex: global::set_tracer_provider(provider.clone())), so that instrumentations and other components can obtain tracers from global::tracer(). The tracer_provider must be kept around to call shutdown on it at the end of application (ex: tracer_provider.shutdown())

  • Feature: Add ResourceBuilder for an easy way to create new Resources

  • Breaking: Remove Resource::{new,empty,from_detectors,new_with_defaults,from_schema_url,merge,default} from public api. To create Resources you should only use Resource::builder() or Resource::builder_empty(). See #2322 for a migration guide. Example Usage:

    // old
Resource::default().with_attributes([
    KeyValue::new("service.name", "test_service"),
    KeyValue::new("key", "value"),
]);

// new
Resource::builder()
    .with_service_name("test_service")
    .with_attribute(KeyValue::new("key", "value"))
    .build();

  • Breaking The LogExporter::export() method no longer requires a mutable reference to self.: Before: async fn export(&mut self, batch: LogBatch<'>) -> LogResult<()> After: async fn export(&self, batch: LogBatch<'>) -> LogResult<()> Custom exporters will need to internally synchronize any mutable state, if applicable.

  • Breaking Removed the following deprecated struct:

    • logs::LogData - Previously deprecated in version 0.27.1 Migration Guidance: This structure is no longer utilized within the SDK, and users should not have dependencies on it.

  • Breaking Removed the following deprecated methods:

    • Logger::provider() : Previously deprecated in version 0.27.1
    • Logger::instrumentation_scope() : Previously deprecated in version 0.27.1. Migration Guidance: - These methods were intended for log appenders. Keep the clone of the provider handle, instead of depending on above methods.

  • Breaking - PeriodicReader Updates

    PeriodicReader no longer requires an async runtime by default. Instead, it now creates its own background thread for execution. This change allows metrics to be used in environments without async runtimes.

    For users who prefer the previous behavior of relying on a specific Runtime, they can do so by enabling the feature flag experimental_metrics_periodicreader_with_async_runtime.

    Migration Guide:

    1. Default Implementation, requires no async runtime (Recommended) The new default implementation does not require a runtime argument. Replace the builder method accordingly: Before:
    let reader = opentelemetry_sdk::metrics::PeriodicReader::builder(exporter, runtime::Tokio).build();
  ```

*After:*

```rust
let reader = opentelemetry_sdk::metrics::PeriodicReader::builder(exporter).build();

    The new PeriodicReader can be used with OTLP Exporter, and supports following exporter features:

    • grpc-tonic: This requires MeterProvider to be created within a tokio runtime.
    • reqwest-blocking-client: Works with a regular main or tokio::main.

    In other words, other clients like reqwest and hyper are not supported.

  1. Async Runtime Support If your application cannot spin up new threads or you prefer using async runtimes, enable the "experimental_metrics_periodicreader_with_async_runtime" feature flag and adjust code as below.

    • Before:

      let reader = opentelemetry_sdk::metrics::PeriodicReader::builder(exporter, runtime::Tokio).build();

    • After:

      let reader = opentelemetry_sdk::metrics::periodic_reader_with_async_runtime::PeriodicReader::builder(exporter, runtime::Tokio).build();

    Requirements:

    • Enable the feature flag: experimental_metrics_periodicreader_with_async_runtime.
    • Continue enabling one of the async runtime feature flags: rt-tokio, rt-tokio-current-thread, or rt-async-std.

    Feature flag "experimental_metrics_periodic_reader_no_runtime" is removed as the PeriodicReader offered under that feature flag is now the default.

  • Bump msrv to 1.75.0.

  • Breaking : #2314

    • The LogRecord struct has been updated:
      • All fields are now pub(crate) instead of pub.
      • Getter methods have been introduced to access field values. This change impacts custom exporter and processor developers by requiring updates to code that directly accessed LogRecord fields. They must now use the provided getter methods (e.g., log_record.event_name() instead of log_record.event_name).

  • Upgrade the tracing crate used for internal logging to version 0.1.40 or later. This is necessary because the internal logging macros utilize the name field as metadata, a feature introduced in version 0.1.40. #2418

  • Breaking - BatchLogProcessor Updates #2436 BatchLogProcessor no longer requires an async runtime by default. Instead, a dedicated background thread is created to do the batch processing and exporting.

    For users who prefer the previous behavior of relying on a specific Runtime, they can do so by enabling the feature flag experimental_logs_batch_log_processor_with_async_runtime.

  1. Default Implementation, requires no async runtime (Recommended) The new default implementation does not require a runtime argument. Replace the builder method accordingly:

    • Before:

      let logger_provider = LoggerProvider::builder()
  .with_log_processor(BatchLogProcessor::builder(exporter, runtime::Tokio).build())
  .build();

    • After:

      let logger_provider = LoggerProvider::builder()
  .with_log_processor(BatchLogProcessor::builder(exporter).build())
  .build();

    The new BatchLogProcessor can be used with OTLP Exporter, and supports following exporter features:

    • grpc-tonic: This requires MeterProvider to be created within a tokio runtime.
    • reqwest-blocking-client: Works with a regular main or tokio::main.

    In other words, other clients like reqwest and hyper are not supported.

  2. Async Runtime Support If your application cannot spin up new threads or you prefer using async runtimes, enable the "experimental_logs_batch_log_processor_with_async_runtime" feature flag and adjust code as below.

    • Before:

      let logger_provider = LoggerProvider::builder()
  .with_log_processor(BatchLogProcessor::builder(exporter, runtime::Tokio).build())
  .build();

    • After:

      let logger_provider = LoggerProvider::builder()
  .with_log_processor(log_processor_with_async_runtime::BatchLogProcessor::builder(exporter, runtime::Tokio).build())
  .build();

    Requirements:

    • Enable the feature flag: experimental_logs_batch_log_processor_with_async_runtime.
    • Continue enabling one of the async runtime feature flags: rt-tokio, rt-tokio-current-thread, or rt-async-std.

  • Breaking - BatchSpanProcessor Updates #2435

    BatchSpanProcessor no longer requires an async runtime by default. Instead, a dedicated background thread is created to do the batch processing and exporting.

    For users who prefer the previous behavior of relying on a specific Runtime, they can do so by enabling the feature flag experimental_trace_batch_span_processor_with_async_runtime.

  1. Default Implementation, requires no async runtime (Recommended) The new default implementation does not require a runtime argument. Replace the builder method accordingly:

    • Before:

      let tracer_provider = TracerProvider::builder()
  .with_span_processor(BatchSpanProcessor::builder(exporter, runtime::Tokio).build())
  .build();

    • After:

      let tracer_provider = TracerProvider::builder()
  .with_span_processor(BatchSpanProcessor::builder(exporter).build())
  .build();

      This implementation does not support multiple concurrent exports (with_max_concurrent_exports is not supported).

    The new BatchLogProcessor can be used with OTLP Exporter, and supports following exporter features:

    • grpc-tonic: This requires MeterProvider to be created within a tokio runtime.
    • reqwest-blocking-client: Works with a regular main or tokio::main.

    In other words, other clients like reqwest and hyper are not supported.

  2. Async Runtime Support If your application cannot spin up new threads or you prefer using async runtimes, enable the "experimental_trace_batch_span_processor_with_async_runtime" feature flag and adjust code as below.

    • Before:

      let tracer_provider = TracerProvider::builder()
  .with_span_processor(BatchSpanProcessor::builder(exporter, runtime::Tokio).build())
  .build();

    • After:

      let tracer_provider = TracerProvider::builder()
  .with_span_processor(span_processor_with_async_runtime::BatchSpanProcessor::builder(exporter, runtime::Tokio).build())
  .build();

    Requirements:

    • Enable the feature flag: experimental_trace_batch_span_processor_with_async_runtime.
    • Continue enabling one of the async runtime feature flags: rt-tokio, rt-tokio-current-thread, or rt-async-std.

  • Bug fix: Empty Tracer names are retained as-is instead of replacing with "rust.opentelemetry.io/sdk/tracer" #2486

  • Update EnvResourceDetector to allow resource attribute values containing equal signs ("="). #2120

  • Breaking Introduced experimental_async_runtime feature for runtime-specific traits.

    • Runtime-specific features (rt-tokio, rt-tokio-current-thread, and rt-async-std) now depend on the experimental_async_runtime feature.
    • For most users, no action is required. Enabling runtime features such as rt-tokio, rt-tokio-current-thread, or rt-async-std will automatically enable the experimental_async_runtime feature.
    • If you're implementing a custom runtime, you must explicitly enable the experimental_async_runtimefeature in your Cargo.toml and implement the requiredRuntime` traits.

  • Removed Metrics Cardinality Limit feature. This was originally introduced in #1066 with a hardcoded limit of 2000 and no ability to change it. This feature will be re-introduced in a future date, along with the ability to change the cardinality limit.

  • Breaking Removed unused opentelemetry_sdk::Error enum.

  • Breaking Resource.get() modified to require reference to Key instead of owned. Replace get(Key::from_static_str("key")) with get(&Key::from_static_str("key"))

  • Breaking (Affects custom Exporter authors only) Moved ExportError trait from opentelemetry::export::ExportError to opentelemetry_sdk::ExportError

  • Breaking (Affects custom SpanExporter, SpanProcessor authors only): Rename namespaces for Span exporter structs/traits before: opentelemetry_sdk::export::spans::{ExportResult, SpanData, SpanExporter}; now: opentelemetry_sdk::spans::{ExportResult, SpanData, SpanExporter};

  • Breaking (Affects custom LogExporter, LogProcessor authors only): Rename namespaces for Log exporter structs/traits. before: opentelemetry_sdk::export::logs::{ExportResult, LogBatch, LogExporter}; now: opentelemetry_sdk::logs::{ExportResult, LogBatch, LogExporter};

  • Breaking opentelemetry_sdk::LogRecord::default() method is removed. The only way to create log record outside opentelemetry_sdk crate is using Logger::create_log_record() method.

  • Rename opentelemetry_sdk::logs::Builder to opentelemetry_sdk::logs::LoggerProviderBuilder.

  • Rename opentelemetry_sdk::trace::Builder to opentelemetry_sdk::trace::SdkTracerProviderBuilder.

  • Breaking: Rename namespaces for InMemoryExporters. (The module is still under "testing" feature flag) before: opentelemetry_sdk::testing::logs::{InMemoryLogExporter, InMemoryLogExporterBuilder}; opentelemetry_sdk::testing::trace::{InMemorySpanExporter, InMemorySpanExporterBuilder}; opentelemetry_sdk::testing::metrics::{InMemoryMetricExporter, InMemoryMetricExporterBuilder}; now: opentelemetry_sdk::logs::{InMemoryLogExporter, InMemoryLogExporterBuilder}; opentelemetry_sdk::trace::{InMemorySpanExporter, InMemorySpanExporterBuilder}; opentelemetry_sdk::metrics::{InMemoryMetricExporter, InMemoryMetricExporterBuilder};

  • Breaking: The BatchLogProcessor no longer supports configuration of max_export_timeout or the OTEL_BLRP_EXPORT_TIMEOUT environment variable. Timeout handling is now the responsibility of the exporter. For example, in the OTLP Logs exporter, the export timeout can be configured using:

  • The environment variables OTEL_EXPORTER_OTLP_TIMEOUT or OTEL_EXPORTER_OTLP_LOGS_TIMEOUT.

  • The opentelemetry_otlp API, via .with_tonic().with_timeout() or .with_http().with_timeout().

Before:

let processor = BatchLogProcessor::builder(exporter)
    .with_batch_config(
        BatchConfigBuilder::default()
            .with_max_queue_size(2048)
            .with_max_export_batch_size(512)
            .with_scheduled_delay(Duration::from_secs(5))
            .with_max_export_timeout(Duration::from_secs(30)) // Previously configurable
            .build(),
    )
    .build();

After:

let processor = BatchLogProcessor::builder(exporter)
    .with_batch_config(
        BatchConfigBuilder::default()
            .with_max_queue_size(2048)
            .with_max_export_batch_size(512)
            .with_scheduled_delay(Duration::from_secs(5)) // No `max_export_timeout`
            .build(),
    )
    .build();

  • Breaking: The BatchSpanProcessor no longer supports configuration of max_export_timeout or the OTEL_BLRP_EXPORT_TIMEOUT environment variable. Timeout handling is now the responsibility of the exporter. For example, in the OTLP Span exporter, the export timeout can be configured using:

    • The environment variables OTEL_EXPORTER_OTLP_TIMEOUT or OTEL_EXPORTER_OTLP_TRACES_TIMEOUT.
    • The opentelemetry_otlp API, via .with_tonic().with_timeout() or .with_http().with_timeout().

    Before:

let processor = BatchSpanProcessor::builder(exporter)
    .with_batch_config(
        BatchConfigBuilder::default()
            .with_max_queue_size(2048)
            .with_max_export_batch_size(512)
            .with_scheduled_delay(Duration::from_secs(5))
            .with_max_export_timeout(Duration::from_secs(30)) // Previously configurable
            .build(),
    )
    .build();

After:

let processor = BatchSpanProcessor::builder(exporter)
    .with_batch_config(
        BatchConfigBuilder::default()
            .with_max_queue_size(2048)
            .with_max_export_batch_size(512)
            .with_scheduled_delay(Duration::from_secs(5)) // No `max_export_timeout`
            .build(),
    )
    .build();

  • Breaking: The PeriodicReader no longer supports configuration of export timeout using with_timeout API method. Timeout handling is now the responsibility of the exporter.

    For example, in the OTLP Metrics exporter, the export timeout can be configured using:

    • The environment variables OTEL_EXPORTER_OTLP_TIMEOUT or OTEL_EXPORTER_OTLP_METRICS_TIMEOUT.
    • The opentelemetry_otlp API, via .with_tonic().with_timeout() or .with_http().with_timeout().

  • Breaking

  • The public API changes in the Tracing:

    • Before: 
        fn SpanExporter::export(&mut self, batch: Vec<SpanData>) -> BoxFuture<'static, ExportResult>;
  fn SpanExporter::shutdown(&mut self);
  fn SpanExporter::force_flush(&mut self) -> BoxFuture<'static, ExportResult>
  fn TraerProvider::shutdown(&self) -> TraceResult<()>
  fn TracerProvider::force_flush(&self) -> Vec<TraceResult<()>>
    • After: 
        fn SpanExporter::export(&mut self, batch: Vec<SpanData>) -> BoxFuture<'static, OTelSdkResult>;
  fn SpanExporter::shutdown(&mut self) -> OTelSdkResult;
  fn SpanExporter::force_flush(&mut self) -> BoxFuture<'static, OTelSdkResult>
  fn TraerProvider::shutdown(&self) -> OTelSdkResult;
  fn TracerProvider::force_flush(&self) -> OTelSdkResult;

  • Breaking Renamed LoggerProvider, Logger and LogRecord' to SdkLoggerProvider, SdkLoggerandSdkLogRecord` respectively to avoid name collision with public API types. #2612

  • Breaking Renamed TracerProvider and Tracer to SdkTracerProvider and SdkTracer to avoid name collision with public API types. Tracer is still type-aliased to SdkTracer to keep back-compat with tracing-opentelemetry. #2614

0.27.1

Released 2024-Nov-27

  • DEPRECATED:

    • trace::Config methods are moving onto TracerProvider Builder to be consistent with other signals. See #2303 for migration guide. trace::Config is scheduled to be removed from public API in v0.28.0. example:

      // old
let tracer_provider: TracerProvider = TracerProvider::builder()
    .with_config(Config::default().with_resource(Resource::empty()))
    .build();

// new
let tracer_provider: TracerProvider = TracerProvider::builder()
    .with_resource(Resource::empty())
    .build();

    • logs::LogData struct is deprecated, and scheduled to be removed from public API in v0.28.0.

    • Bug fix: Empty Meter names are retained as-is instead of replacing with "rust.opentelemetry.io/sdk/meter" #2334

    • Bug fix: Empty Logger names are retained as-is instead of replacing with "rust.opentelemetry.io/sdk/logger" #2316

    • Logger::provider: This method is deprecated as of version 0.27.1. To be removed in 0.28.0.

    • Logger::instrumentation_scope: This method is deprecated as of version 0.27.1. To be removed in 0.28.0 Migration Guidance: - These methods are intended for log appenders. Keep the clone of the provider handle, instead of depending on above methods.

    • Bug Fix: Validates the with_boundaries bucket boundaries used in Histograms. The boundaries provided by the user must not contain f64::NAN, f64::INFINITY or f64::NEG_INFINITY and must be sorted in strictly increasing order, and contain no duplicates. Instruments will not record measurements if the boundaries are invalid. #2351

  • Added with_periodic_exporter method to MeterProviderBuilder, allowing users to easily attach an exporter with a PeriodicReader for automatic metric export. Retained with_reader() for advanced use cases where a custom MetricReader configuration is needed. 2597 Example Usage:

    SdkMeterProvider::builder()
    .with_periodic_exporter(exporter)
    .build();

    Using a custom PeriodicReader (advanced use case):

    let reader = PeriodicReader::builder(exporter).build(); SdkMeterProvider::builder() .with_reader(reader) .build();

0.27.0

Released 2024-Nov-11

  • Update opentelemetry dependency version to 0.27

  • Update opentelemetry-http dependency version to 0.27

  • Bump MSRV to 1.70 #2179

  • Implement LogRecord::set_trace_context for LogRecord. Respect any trace context set on a LogRecord when emitting through a Logger.

  • Improved LoggerProvider shutdown handling to prevent redundant shutdown calls when drop is invoked. #2195

  • When creating new metric instruments by calling build(), SDK would return a no-op instrument if the validation fails (eg: Invalid metric name). #2166

  • BREAKING for Metrics users:

    • Replaced
      • (#2217): Removed {Delta,Cumulative}TemporalitySelector::new() in favor of directly using Temporality enum to simplify the configuration of MetricsExporterBuilder with different temporalities.
    • Renamed
      • (#2232): The init method used to create instruments has been renamed to build. Before: 
        let counter = meter.u64_counter("my_counter").init();
        Now: 
        let counter = meter.u64_counter("my_counter").build();
      • (#2255): de-pluralize Metric types.
        • PushMetricsExporter -> PushMetricExporter
        • InMemoryMetricsExporter -> InMemoryMetricExporter
        • InMemoryMetricsExporterBuilder -> InMemoryMetricExporterBuilder

  • BREAKING: #2220

    • Removed InstrumentationLibrary re-export and its Scope alias, use opentelemetry::InstrumentationLibrary instead.
    • Unified builders across signals
      • Removed deprecated LoggerProvider::versioned_logger, TracerProvider::versioned_tracer
      • Removed MeterProvider::versioned_meter
      • Replaced these methods with LoggerProvider::logger_with_scope, TracerProvider::logger_with_scope, MeterProvider::meter_with_scope

  • #2272

    • Pin url version to 2.5.2. The higher version breaks the build refer: servo/rust-url#992. The url crate is used when jaeger_remote_sampler feature is enabled.

  • BREAKING: #2266

    • Moved ExportError trait from opentelemetry::ExportError to opentelemetry_sdk::export::ExportError

    • Moved LogError enum from opentelemetry::logs::LogError to opentelemetry_sdk::logs::LogError

    • Moved LogResult type alias from opentelemetry::logs::LogResult to opentelemetry_sdk::logs::LogResult

    • Renamed opentelemetry::metrics::Result type alias to opentelemetry::metrics::MetricResult

    • Renamed opentelemetry::metrics::MetricsError enum to opentelemetry::metrics::MetricError

    • Moved MetricError enum from opentelemetry::metrics::MetricError to opentelemetry_sdk::metrics::MetricError

    • Moved MetricResult type alias from opentelemetry::metrics::MetricResult to opentelemetry_sdk::metrics::MetricResult

    • Users calling public APIs that return these constructs (e.g, LoggerProvider::shutdown(), MeterProvider::force_flush()) should now import them from the SDK instead of the API.

    • Developers creating custom exporters should ensure they import these constructs from the SDK, not the API.

    • 2291 Rename logs_level_enabled flag to spec_unstable_logs_enabled. Please enable this updated flag if the feature is needed. This flag will be removed once the feature is stabilized in the specifications.

  • BREAKING: Temporality enum moved from opentelemetry_sdk::metrics::data::Temporality to opentelemetry_sdk::metrics::Temporality.

  • BREAKING: Views are now an opt-in ONLY feature. Please include the feature spec_unstable_metrics_views to enable Views. It will be stabilized post 1.0 stable release of the SDK. #2295

  • Added a new PeriodicReader implementation (PeriodicReaderWithOwnThread) that does not rely on an async runtime, and instead creates own Thread. This is under feature flag "experimental_metrics_periodic_reader_no_runtime". The functionality maybe moved into existing PeriodReader or even removed in the future. As of today, this cannot be used as-is with OTLP Metric Exporter or any exporter that require an async runtime.

v0.26.0

Released 2024-Sep-30

  • Update opentelemetry dependency version to 0.26

  • BREAKING Public API changes:

    • Removed: SdkMeter struct #2113. This API is only meant for internal use.
    • Removed: AggregationSelector trait and DefaultAggregationSelector struct #2085. This API was unnecessary. The feature to customize aggregation for instruments should be offered by Views API.

  • Update async-std dependency version to 1.13

  • Breaking - Remove support for MetricProducer which allowed metrics from external sources to be sent through OpenTelemetry. #2105

  • Feature: SimpleSpanProcessor::new is now public #2119

  • For Delta Temporality, exporters are not invoked unless there were new measurements since the last collect/export. #2153

  • MeterProvider modified to not invoke shutdown on Drop, if user has already called shutdown(). #2156

v0.25.0

  • Update opentelemetry dependency version to 0.25

  • Starting with this version, this crate will align with opentelemetry crate on major,minor versions.

  • Perf improvements for all metric instruments (except ExponentialHistogram) that led to faster metric updates and higher throughput #1740:

    • Zero allocations when recording measurements: Once a measurement for a given attribute combination is reported, the SDK would not allocate additional memory for subsquent measurements reported for the same combination.
    • Minimized thread contention: Threads reporting measurements for the same instrument no longer contest for the same Mutex. The internal aggregation data structure now uses a combination of RwLock and atomics. Consequently, threads reporting measurements now only have to acquire a read lock.
    • Lock-free floating point updates: Measurements reported for f64 based metrics no longer need to acquire a Mutex to update the f64 value. They use a CAS-based loop instead.

  • opentelemetry_sdk::logs::record::LogRecord and opentelemetry_sdk::logs::record::TraceContext derive from PartialEq to facilitate Unit Testing.

  • Fixed an issue causing a panic during shutdown when using the TokioCurrentThread in BatchExportProcessor for traces and logs. #1964 #1973

  • Fix BatchExportProcessor for traces and logs to trigger first export at the first interval instead of doing it right away. #1970 #1973

    • Breaking #1985 Hide LogRecord attributes Implementation Details from processors and exporters. The custom exporters and processors can't directly access the LogData::LogRecord::attributes, as these are private to opentelemetry-sdk. Instead, they would now use LogRecord::attributes_iter() method to access them.

  • Fixed various Metric aggregation bug related to ObservableCounter,UpDownCounter including #1517. #2004

  • Fixed a bug related to cumulative aggregation of Gauge measurements. #1975. #2021

  • Provide default implementation for event_enabled method in LogProcessor trait that returns true always.

  • Breaking #2041 and #2057

    • The Exporter::export() interface is modified as below: Previous Signature:

      async fn export<'a>(&mut self, batch: Vec<Cow<'a, LogData>>) -> LogResult<()>;

      Updated Signature:

      async fn export(&mut self, batch: LogBatch<'_>) -> LogResult<()>;

      where

      pub struct LogBatch<'a> {

  data: &'a [(&'a LogRecord, &'a InstrumentationLibrary)],
}

      This change enhances performance by reducing unnecessary heap allocations and maintains object safety, allowing for more efficient handling of log records. It also simplifies the processing required by exporters. Exporters no longer need to determine if the LogData is borrowed or owned, as they now work directly with references. As a result, exporters must explicitly create a copy of LogRecord and/or InstrumentationLibrary when needed, as the new interface only provides references to these structures.

v0.24.1

  • Add hidden method to support tracing-opentelemetry

v0.24.0

  • Add "metrics", "logs" to default features. With this, default feature list is "trace", "metrics" and "logs".

  • Add with_resource on Builder for LoggerProvider, replacing the with_config method. Instead of using .with_config(Config::default().with_resource(RESOURCE::default())) users must now use .with_resource(RESOURCE::default()) to configure Resource on logger provider.

  • Removed dependency on ordered-float.

  • Removed XrayIdGenerator, which was marked deprecated since 0.21.3. Use opentelemetry-aws, version 0.10.0 or newer.

  • Performance Improvement - Counter/UpDownCounter instruments internally use RwLock instead of Mutex to reduce contention

  • Breaking 1726 Update LogProcessor::emit() method to take mutable reference to LogData. This is breaking change for LogProcessor developers. If the processor needs to invoke the exporter asynchronously, it should clone the data to ensure it can be safely processed without lifetime issues. Any changes made to the log data before cloning in this method will be reflected in the next log processor in the chain, as well as to the exporter.

  • Breaking 1726 Update LogExporter::export() method to accept a batch of log data, which can be either a reference or ownedLogData. If the exporter needs to process the log data asynchronously, it should clone the log data to ensure it can be safely processed without lifetime issues.

  • Clean up public methods in SDK.

    • [TracerProvider::span_processors] and [TracerProvider::config] was removed as it's not part of the spec.
    • Added non_exhaustive annotation to [trace::Config]. Marked [config] as deprecated since it's only a wrapper for Config::default
    • Removed [Tracer::tracer_provder] and [Tracer::instrument_libraries] as it's not part of the spec.

  • Breaking #1830 [Traces SDK] Improves performance by sending Resource information to processors (and exporters) once, instead of sending with every log. If you are an author of Processor, Exporter, the following are BREAKING changes.

    • Implement set_resource method in your custom SpanProcessor, which invokes exporter's set_resource.
    • Implement set_resource method in your custom SpanExporter. This method should save the resource object in original or serialized format, to be merged with every span event during export.
    • SpanData doesn't have the resource attributes. The SpanExporter::export() method needs to merge it with the earlier preserved resource before export.

  • Breaking 1836 SpanProcessor::shutdown now takes an immutable reference to self. Any reference can call shutdown on the processor. After the first call to shutdown the processor will not process any new spans.

  • Breaking [1850] (#1850) LoggerProvider::log_processors() and LoggerProvider::resource() are not public methods anymore. They are only used within the opentelemetry-sdk crate.

  • 1857 Fixed an issue in Metrics SDK which prevented export errors from being send to global error handler. With the fix, errors occurring during export like OTLP Endpoint unresponsive shows up in stderr by default.

  • 1869 Added a target field to LogRecord structure, populated by opentelemetry-appender-tracing and opentelemetry-appender-log appenders.

async fn export<'a>(&mut self, batch: Vec<Cow<'a, LogData>>) -> LogResult<()>;

where LogRecord within LogData now includes:

LogData {
  LogRecord {
    event_name,
    target,  // newly added
    timestamp,
    observed_timestamp,
    trace_context,
    trace_context,
    severity_number,
    body,
    attributes
  }
  Instrumentation {
    name,
    version,
    schema_url,
    version
  }
}

The LogRecord::target field contains the actual target/component emitting the logs, while the Instrumentation::name contains the name of the OpenTelemetry appender.

  • Breaking #1674 Update to http v1 types (via opentelemetry-http update)
  • Update opentelemetry dependency version to 0.24
  • Update opentelemetry-http dependency version to 0.13

v0.23.0

  • Fix SimpleSpanProcessor to be consistent with log counterpart. Also removed dependency on crossbeam-channel. 1612
  • #1422 Fix metrics aggregation bug when using Views to drop attributes.
  • #1766 Fix Metrics PeriodicReader to trigger first collect/export at the first interval instead of doing it right away.
  • #1623 Add Drop implementation for SdkMeterProvider, which shuts down metricreaders, thereby allowing metrics still in memory to be flushed out.
  • Breaking #1624 Remove OsResourceDetector and ProcessResourceDetector resource detectors, use the opentelemetry-resource-detector instead.
  • #1636 [Logs SDK] Improves performance by sending Resource information to processors (and exporters) once, instead of sending with every log. If you are an author of Processor, Exporter, the following are BREAKING changes.
    • Implement set_resource method in your custom LogProcessor, which invokes exporter's set_resource.
    • Implement set_resource method in your custom LogExporter. This method should save the resource object in original or serialized format, to be merged with every log event during export.
    • LogData doesn't have the resource attributes. The LogExporter::export() method needs to merge it with the earlier preserved resource before export.
  • Baggage propagation error will be reported to global error handler #1640
  • Improves shutdown behavior of LoggerProvider and LogProcessor #1643.
    • shutdown can be called by any clone of the LoggerProvider without the need of waiting on all Logger drops. Thus, try_shutdown has been removed.
    • shutdown methods in LoggerProvider and LogProcessor now takes a immutable reference
    • After shutdown, LoggerProvider will return noop Logger
    • After shutdown, LogProcessor will not process any new logs
  • Moving LogRecord implementation to the SDK. 1702.
    • Relocated LogRecord struct to SDK, as an implementation for the trait in the API.
  • Breaking #1729
    • Update the return type of TracerProvider.span_processors() from &Vec<Box<dyn SpanProcessor>> to &[Box<dyn SpanProcessor>].
    • Update the return type of LoggerProvider.log_processors() from &Vec<Box<dyn LogProcessor>> to &[Box<dyn LogProcessor>].
  • Update opentelemetry dependency version to 0.23
  • Update opentelemetry-http dependency version to 0.12
  • Breaking #1750
    • Update the return type of LoggerProvider.shutdown() from Vec<LogResult<()>> to LogResult<()>.

v0.22.1

Fixed

  • #1576 Fix Span kind is always set to "internal".

v0.22.0

Deprecated

  • XrayIdGenerator in the opentelemetry-sdk has been deprecated and moved to version 0.10.0 of the opentelemetry-aws crate.

Added

  • #1410 Add experimental synchronous gauge

  • #1471 Configure batch log record processor via OTEL_BLRP_* environment variables and via OtlpLogPipeline::with_batch_config

  • #1503 Make the documentation for In-Memory exporters visible.

  • #1526 Performance Improvement : Creating Spans and LogRecords are now faster, by avoiding expensive cloning of Resource for every Span/LogRecord.

Changed

  • Breaking #1313 #1350 Changes how Span links/events are stored to achieve performance gains. See below for details:

    Behavior Change: When enforcing max_links_per_span, max_events_per_span from SpanLimits, links/events are kept in the first-come order. The previous "eviction" based approach is no longer performed.

    Breaking Change Affecting Exporter authors:

    SpanData now stores links as SpanLinks instead of EvictedQueue where SpanLinks is a struct with a Vec of links and dropped_count.

    SpanData now stores events as SpanEvents instead of EvictedQueue where SpanEvents is a struct with a Vec of events and dropped_count.

  • Breaking Remove TextMapCompositePropagator #1373. Use TextMapCompositePropagator in opentelemetry API.

  • #1375 Fix metric collections during PeriodicReader shutdown

  • Breaking #1480 Remove fine grained BatchConfig configurations from BatchLogProcessorBuilder and BatchSpanProcessorBuilder. Use BatchConfigBuilder to construct a BatchConfig instance and pass it using BatchLogProcessorBuilder::with_batch_config or BatchSpanProcessorBuilder::with_batch_config.

  • Breaking #1480 Remove mutating functions from BatchConfig, use BatchConfigBuilder to construct a BatchConfig instance.

  • Breaking #1495 Remove Batch LogRecord&Span Processor configuration via non-standard environment variables. Use the following table to migrate from the no longer supported non-standard environment variables to the standard ones.

No longer supported Standard equivalent
OTEL_BLRP_SCHEDULE_DELAY_MILLIS OTEL_BLRP_SCHEDULE_DELAY
OTEL_BLRP_EXPORT_TIMEOUT_MILLIS OTEL_BLRP_EXPORT_TIMEOUT
OTEL_BSP_SCHEDULE_DELAY_MILLIS OTEL_BSP_SCHEDULE_DELAY
OTEL_BSP_EXPORT_TIMEOUT_MILLIS OTEL_BSP_EXPORT_TIMEOUT

  • Breaking #1455 Make the LoggerProvider Owned

    • Logger now takes an Owned Logger instead of a Weak<LoggerProviderInner>
    • LoggerProviderInner is no longer pub (crate)
    • Logger.provider() now returns &LoggerProvider instead of an Option<LoggerProvider>

  • #1519 Performance improvements when calling Counter::add() and UpDownCounter::add() with an empty set of attributes (e.g. counter.Add(5, &[]))

  • Breaking Renamed MeterProvider and Meter to SdkMeterProvider and SdkMeter respectively to avoid name collision with public API types. #1328

Fixed

  • #1481 Fix error message caused by race condition when using PeriodicReader

v0.21.2

Fixed

  • Fix delta aggregation metric reuse. #1434
  • Fix max_scale validation of exponential histogram configuration. #1452

v0.21.1

Fixed

  • Fix metric export corruption if gauges have not received a last value. #1363
  • Return consistent Meter for a given scope from MeterProvider. #1351

v0.21.0

Added

  • Log warning if two instruments have the same name with different #1266 casing
  • Log warning if view is created with empty criteria #1266
  • Add exponential histogram support #1267
  • Add opentelemetry::sdk::logs::config() for parity with opentelemetry::sdk::trace::config() #1197

Changed

  • Bump MSRV to 1.65 #1318

  • Default Resource (the one used when no other Resource is explicitly provided) now includes TelemetryResourceDetector, populating "telemetry.sdk.*" attributes. #1194.

  • Bump MSRV to 1.64 #1203

  • Add unit/doc tests for MeterProvider #1220

  • Changed dependency from opentelemetry_api to opentelemetry as the latter is now the API crate. #1226

  • Add in memory span exporter #1216

  • Add in memory log exporter #1231

  • Add Sync bound to the SpanExporter and LogExporter traits #1240

  • Move MetricsProducer config to builders to match other config #1266

  • Return error earlier if readers are shut down #1266

  • Add / to valid characters for instrument names #1269

  • Increase instrument name maximum length from 63 to 255 #1269

  • Updated crate documentation and examples. #1256

  • Replace regex with glob #1301

  • Breaking #1293 makes few breaking changes with respect to how Span attributes are stored to achieve performance gains. See below for details:

    Behavior Change:

    SDK will no longer perform de-duplication of Span attribute Keys. Please share feedback here, if you are affected.

    Breaking Change Affecting Exporter authors:

    SpanData now stores attributes as Vec<KeyValue> instead of EvictedHashMap. SpanData now expose dropped_attributes_count as a separate field.

    Breaking Change Affecting Sampler authors:

    should_sample changes attributes from OrderMap<Key, Value> to Vec<KeyValue>.

  • Breaking Move type argument from RuntimeChannel<T> to associated types #1314

Removed

  • Remove context from Metric force_flush #1245
  • Remove logs::BatchMessage and trace::BatchMessage types #1314

Fixed

  • Fix metric instrument name validation to include _ #1274

v0.20.0

Added

  • Implement cardinality limits for metric streams #1066.
  • Propagate shutdown calls from PeriodicReader to metrics exporter #1138.
  • Add in memory metrics exporter #1017

Changed

  • New metrics SDK #1000
  • Use Cow<'static, str> instead of &'static str #1018
  • Unify trace and logs runtime extensions traits. #1067

Changed

  • Fix EvictedQueue bug when capacity is set to 0 #1151.

Removed

  • Samplers no longer has access to InstrumentationLibrary as one of parameters to should_sample. #1041.
  • Synchronous instruments no longer accepts Context while reporting measurements. #1076.
  • Don't use CARGO_BIN_NAME for service name #1150

Fixed

  • Wait for exports on the simple span processor's ForceFlush #1030

v0.19.0

Added

  • Add instrument validation to InstrumentBuilder #884.
  • Add TelemetryResourceDetector #899.
  • Add support for instrumentation scope attributes #1021.

Changed

  • Update to opentelemetry_api v0.19.
  • Update to opentelemetry_http v0.8.
  • Bump MSRV to 1.57 #953.
  • Fix doc in ShouldSample trait #951
  • Only run ParentBased delegate sampler when there is no parent #948.
  • Improve SdkProvidedResourceDetector's doc #964.
  • Update dependencies and bump MSRV to 1.60 #969.
  • Use CARGO_BIN_NAME as default service name #991.

Removed

  • Remove in_memory settings #946.

main

Changed

  • Update the Number in the SDK API to support min and max. #989

v0.18.0

Changed

  • BREAKING structs which implement ShouldSample a.k.a Custom Samplers must now implement Clone. This enables (#833)
  • SDK split from opentelemetry crate