v1.1.3

See the 1.1.3 milestone for included items.

Fixes

Includes 1.0.10 fixes up to the point of 1.1.3 release (see diff).

• Wait to publish metrics to Cloudwatch even if application is shutting down
• Handle NaN in various registries
  • LoggingMeterRegistry
  • HumioMeterRegistry
• KafkaConsumerMetrics' "records-lead" metric should not be a TimeGauge
• KafkaConsumerMetrics properly tagged per topic and partition
• StackdriverMeterRegistry handles publishing empty histograms

Micrometer Spring Legacy fixes

• Avoid creating meter binders before registry has been customized

Polish

• Avoid MediaType.get calls in OkHttpSender where possible
• JavaDoc polish for Elasticsearch support
• JavaDoc for RestTemplateExchangeTags.outcome
• Avoid potentially expensive calls when debug logging is not enabled
• Align some test class names with class under test
• Unify access modifiers on across some classes
• Fix setup for some test classes
• Properly set interrupted status for CloudWatchMeterRegistry publish thread

v1.1.2

See the 1.1.2 milestone for included items.

Enhancements

• Add outcome tag to default rest template tags provider in spring legacy
• Reading kafka records-lag metrics with topic and partition name
• cloudwatch: log interruptions as a warn

Fixes

• Includes all 1.0.9 fixes.
• Split Dynatrace messages if they are too large
• Fix handling of infinity and NaN values for some registries
• Address potential NPE in TomcatMetricsBinder when Context is null
• Make checkedForIndexTemplate volatile in ElasticMeterRegistry
• HibernateMetrics fails if Hibernate is not on the classpath
• Exclude libnetty_transport_native_epoll_x86_64.so to prevent PortUnreachableException: recvfrom() failed on Linux
• Consider property LoggingRegistryConfig.logInactive() when logging

v1.0.9

Fixes

• Add access token validation to SignalFxPropertiesConfigAdapter
• Fix description and base unit for executor.queued meter

v1.1.1

Enhancements

• Auto-configure Kafka consumer metrics in micrometer-spring-legacy.
• Remove unused Measurement interface in AppOptics.
• Monitor additivity-disabled loggers for Log4j2.

Fixes

• Includes all 1.0.8 fixes.
• Removed meters are removed from Prometheus registry.
• Remove metrics from Dropwizard MetricRegistry when removing a meter.
• Remove HttpSender.print() usages.
• Fix Kafka consumer metrics.
• micrometer-registry-statsd does not pollute classpath with /META-INF/native/libnetty_transport_native_epoll_x86_64.so from Netty 4.1.29.
• Stop configuring metrics if Hibernate is not available.
• Fixed StatsD Nagle buffering algorithm.
• Fix a wrong index when copying an array in Tags.and(Tag...)
• Jersey server configuration alignment.
• Remove logging from JvmGcMetrics.
• PostgreSQLDatabaseMetrics#runQuery advances JDBC cursor to next row.
• Tag deduplication now does not add extra duplicate tag.
• Handle null suffix in CloudWatchMeterRegistry.Batch.metricDatum().

v1.0.8

Fixes

• TimedAspect does not disrupt the application flow when there is an exception recording a metric.
• micrometer-spring-legacy now respects Actuator endpoint settings.
• Removed extraneous print statements in StatsD registry.
• Exception thrown by MicrometerCollector for a meter with a duplicate name now includes the name in its message.
• Fixed accept/deny filter precedence.
• Add endpoint-enabled condition to Prometheus Actuator endpoint.
• Improved Telegraf sanitization.
• Use Commons Logging for WebMvcMetricsFilter.
• WavefrontPropertiesConfigAdapter now extends StepRegistryPropertiesConfigAdapter.
• Fix spring.aop.auto property reference.

v1.1.0

Enhancements

New registries

• Added new registry implementations for monitoring systems:
  • AppOptics
  • Azure Monitor
  • Dynatrace
  • Elastic
  • Humio
  • KairosDB
  • Stackdriver
  • Sysdig StatsD
• Added a LoggingMeterRegistry that formats metrics as logs periodically.

New meter binders

• Jetty server thread pool metrics.
• Thread state metrics to JvmThreadMetrics.
• PostgreSQL metrics.
• Disk space metrics.
• Log4J 1.2 metrics.
• Kafka consumer metrics.
• Enhance ExecutorServiceMetrics to record time between task submission and task execution start time.
• Hystrix metrics reworked from the ground up.

Spring Boot 1.5.x legacy module improvements

• Support management.metrics.tags.* in Spring Boot 1.5.x.
• Support distribution's min/max expected values via properties in Spring Boot 1.5.x.
• Autoconfigure Hibernate metrics.

Other improvements

• Added support for buffering StatsD metrics using Nagle's algorithm.
• Add support for MultiGauge, a helper to spawn a series of gauges from a table-like structure that share a name but vary in tags. MultiGauge allows the set of gauges to grow and shrink over time, provided the total set of tags is well-bounded over time.
• Add support for Meter removal (MeterRegistry#remove(Meter.Id) and MeterRegistry#remove(Meter)) and a new MeterRegistry#config().onMeterRemoved(Consumer) method.
• Meter registration memory and CPU performance improvements.
• Support for strong reference gauges.
• Support for defining gauges from Supplier<T>.
• Add "outcome" tag to WebMVC and Jersey metrics that summarizes the HTTP status as informational, success, redirection, client error, server error, or unknown.
• Reduce default push-based registry logging level to DEBUG and add a single INFO-level log message that indicates the push frequency on startup.
• Upgrade to Dropwizard 4.x for Graphite/JMX.
• Eliminate Dropwizard dependency for Ganglia, instead publishing directly through the Ganglia Java client.
• Add a new overload Timer#start() that uses the default system clock.
• Allow for injection of HTTP interface into push-based registries (e.g. DatadogMeterRegistry.builder(..).httpClient(CUSTOM_HTTP_CLIENT).build().
• Drop Autocloseable from Meter and MeterRegistry to improve static code analysis experience (e.g. Sonar).
• Added structural pattern matching for meters (Meter#match(..) and Meter#use(..))
• Make HierarchicalNameMapper injectable for GraphiteMeterRegistry.
• Add exception handling for TimedAspect.
• Add base units to several binders. (Breaking change for Prometheus. See #1063)

Fixes

• Properly handle infinity values in gauges.
• Remove invalid DEFAULT from some MeterRegistryConfig subclasses (when a default isn't possible because there is a required property).
• Always convert Strings to byte array using UTF-8.

v1.0.7

Enhancements

• Base units are set for percentile data on distribution summaries and timers.
• Don't report on close in GraphiteMeterRegistry/GangliaMeterRegistry if registry is not enabled.
• Remove NotificationListeners created by TomcatMetrics.
• Make ServletMetricsConfiguration an auto-configuration class.
• Conditionalize the JMX name that TomcatMetrics looks up depending on whether the app is using embedded Tomcat or not.
• StepMeterRegistry now produces daemon threads.
• Order MeterFilters and MeterRegistryCustomizers.
• Add a switch to disable ScheduledMethodMetrics.
• Limit metrics collection of incoming requests for Servlet.

Fixes

• Don't allow NaN values to be shipped to Cloudwatch/Wavefront.
• Fixed StackOverflowError in Prometheus registry when thousands of tags exist for one meter name.
• Correctly name exception tag for anonymous exception classes in Jersey instrumentation.
• Make GraphiteNamingConvention respect its delegate NamingConvention.
• Make InfluxNamingConvention respect its delegate for tag values.
• Close all registries when closing CompositeMeterRegistry.
• WavefrontMetricRegistry now honors connectTimeout and readTimeout properties.
• WavefrontMeterRegistry does not send metrics when the value is 'NaN'.
• Ensure that Spring Boot CharacterEncodingFilter is ordered first.
• Restore constructors in StepDistributionSummary and StepTimer (fix API backwards incompatibility introduced in 1.0.6).
• DataSourcePoolMetricsAutoConfiguration now properly checks whether DataSource bean is available properly.
• Restore double quotes for database name in InfluxDB database creation.

v1.0.6

Improvements

• Influx retention policies are now fully customizable through InfluxConfig.
• Make DataSourcePoolMetricsAutoConfiguration conditional on DataSource bean.
• Make RestTemplateMetricsConfiguration an auto-configuration.
• Make meterNamePrefix work as a prefix in MeterFilter#maximumAllowableTags.
• Fetch authorization keys for every batch post in push-based registries containing such keys (including for Datadog).
• JCacheMetrics guards against empty cache names, such as those emanating from Apache Ignite.
• Remove extra space in description text for jvm.memory.committed.
• Make step size configurable for SignalFx.
• Added support for Jersey MappableException in DefaultJerseyTagsProvider.

Fixes

• Only publish on close if the registry implementation is enabled.
• Improve sanitation of name and tags for statsd to replace : characters.
• Record a non-empty tag value for anonymous Exception types thrown in Spring WebMVC endpoints.
• Fixed ConcurrentModificationException in CompositeMeterRegistry.
• Added a base unit to the id on TimeGauge.
• Fixed SLA bug that made the last two buckets always equal in count.
• When using a "maximum allowable tag values" MeterFilter, the tag values seen before the maximum is reached continue to be allowed.
• When the specified tag key was reached in "maximum allowable tag values" MeterFilter, other tag keys were affected too. This is no longer the case.
• Fixed description text that generated errors on Prometheus Pushgateway.
• Metrics.More is now public so less frequently used meter types can be constructed, e.g. Metrics.more().longTaskTimer(..).
• Filters are now only applied once for percentiles, SLAs, and other synthetic meters.
• Fixed regression in 1.0.5 that stopped converting Influx metric names to snakecase.
• Make DatadogNamingConvention#tagKey respect its delegate.
• StepTimer and StepDistributionSummary now update count and sum based on expiry.
• Shade pcollections into micrometer-registry-statsd.

v1.0.5

Notice to StatsD users!

This release mistakenly failed to shade pcollections into micrometer-registry-statsd. For this patch release only, please add pcollections to your runtime classpath:

compile("io.micrometer:micrometer-registry-statsd:1.0.5")
runtime("org.pcollections:pcollections:3.0.1")

This will be resolved in 1.0.6.

Improvements

• Cache StatsD lines for each meter to limit garbage produced by StatsD implementation.
• Make JMX metrics domain configurable by property (backported from Spring Boot 2.0.3).
• Avoid uri tag explosion when use of path variable is undetected (backported from Spring Boot 2.0.3).
• Upgrade to Prometheus simpleclient_common version 0.4.0.
• Memory improvements to camel and upper camel NamingConvention implementations.
• Shutdown thread pool when closing StepMeterRegistry, an abstract base class upon which many implementations are built.

Fixes

• Sanitize Telegraf tags on export.
• HistogramSnapshot methods max(TimeUnit unit) and mean(TimeUnit unit) are appropriately scaled.
• InfluxNamingConvention doesn't apply base convention to tag values.
• GraphiteMeterRegistry passes Clock to self-created GraphiteReporter.
• Use statKeys and statValues in PrometheusMeterRegistry.newMeter().
• Corrected spelling on executor.steals metric name.
• Corrected implementation of maximumAllowableTags to only limit metrics of a particular name.

v1.0.4

Improvements

• Substantially improved meter registration performance.
• Add descriptive text to MeterNotFoundException which is thrown when registry.get("find.me")... fails to find a matching meter:

• All implementation-specific naming conventions that do some sanitization allow choice of base convention through a constructor parameter now.
• Graphite tags-as-prefix are applied in the order in which they are inserted by default.
• Performance of WebMVC tag creation optimized.
• Apply accept/distribution statistic config filters only if meter doesn't already exist (performance nit).
• Added check for empty values for lookup of Spring Boot property-based filter.

Fixes

• Composite meter registries now only update meters once regardless of how many times it appears in the child registry tree.
• We determined that cache miss counts cannot be calculated accurately for Hazelcast, so removed cache.gets with tag result=hit for Hazelcast. A metric cache.partition.gets was added for Hazelcast.
• Properly escape names and tags that are presented to monitoring systems in JSON.
• Explicitly shut down pause detector threads when the registry is closed.
• Preconfigure Atlas registry to rename tags that conflict with reserved names.
• Prometheus Timer and Summary now correctly decay percentiles (this was a regression in 1.0.3).
• Pass percentile precision to child Timer and DistributionSummary.
• Don't report NaN values to Influx.
• Restored a deprecated AbstractTimer constructor which was removed in 1.0.3 (which was a breaking change).

Deprecated

• StatsdConfig.queueSize(), which was unused.