aws-embedded-metrics

Generate CloudWatch metrics embedded within structured log events. The embedded metrics will be extracted so that you can visualize and alarm on them for real-time incident detection. This allows you to monitor aggregated values while preserving the detailed log event context that generates them.

• Use Cases
• Usage
• Graceful Shutdown
• API
• Thread-safety
• Examples
• Development

Use Cases

• Generate custom metrics across compute environments

• Easily generate custom metrics from Lambda functions without requiring custom batching code, making blocking network requests or relying on third-party software.

• Other compute environments (EC2, On-prem, ECS, EKS, and other container environments) are supported by installing the CloudWatch Agent.

  • Examples can be found in examples/README.md

• Link metrics to high cardinality context

  Using the Embedded Metric Format, you will be able to visualize and alarm on custom metrics, but also retain the original, detailed and high-cardinality context which is queryable using CloudWatch Logs Insights. For example, the library automatically injects environment metadata such as Lambda Function version, EC2 instance and image ids into the structured log event data.

Usage

To use a metric logger, you need to manually create and flush the logger.

import software.amazon.cloudwatchlogs.emf.logger.MetricsLogger;
import software.amazon.cloudwatchlogs.emf.model.DimensionSet;
import software.amazon.cloudwatchlogs.emf.model.Unit;

class Example {
	public static void main(String[] args) {
		MetricsLogger metrics = new MetricsLogger();

		try {
			metrics.putDimensions(DimensionSet.of("Service", "Aggregator"));
			metrics.putMetric("ProcessingLatency", 100, Unit.MILLISECONDS, StorageResolution.STANDARD);
			metrics.putMetric("Memory.HeapUsed", 1600424.0, Unit.BYTES, StorageResolution.HIGH);
		} catch (InvalidDimensionException | InvalidMetricException e) {
			log.error(e);
		}

		metrics.putProperty("RequestId", "422b1569-16f6-4a03-b8f0-fe3fd9b100f8");
		metrics.flush();
	}
}

You can find the artifact location and examples of how to include it in your project at Maven Central

Graceful Shutdown

Since: 2.0.0-beta-1

In any environment, other than AWS Lambda, we recommend running an out-of-process agent (the CloudWatch Agent or FireLens / Fluent-Bit) to collect the EMF events. When using an out-of-process agent, this package will buffer the data asynchronously in process to handle any transient communication issues with the agent. This means that when the MetricsLogger gets flushed, data may not be safely persisted yet. To gracefully shutdown the environment, you can call shutdown on the environment's sink. A full example can be found in the examples directory.

// create an environment singleton, this should be re-used across loggers
DefaultEnvironment environment = new DefaultEnvironment(EnvironmentConfigurationProvider.getConfig());

MetricsLogger logger = new MetricsLogger(environment);

try {
	logger.setDimensions(DimensionSet.of("Operation", "ProcessRecords"));
	logger.putMetric("ExampleMetric", 100, Unit.MILLISECONDS);
} catch (InvalidDimensionException | InvalidMetricException e) {
	log.error(e);
}

logger.putProperty("RequestId", "422b1569-16f6-4a03-b8f0-fe3fd9b100f8");
logger.flush();

// flush the sink, waiting up to 10s before giving up
environment.getSink().shutdown().orTimeout(10_000L, TimeUnit.MILLISECONDS);

API

MetricsLogger

The MetricsLogger is the interface you will use to publish embedded metrics.

• MetricsLogger putMetric(String key, double value, Unit unit, StorageResolution storageResolution)
• MetricsLogger putMetric(String key, double value, StorageResolution storageResolution)
• MetricsLogger putMetric(String key, double value, Unit unit)
• MetricsLogger putMetric(String key, double value)

Adds a new metric to the current logger context. Multiple metrics using the same key will be appended to an array of values. Multiple metrics cannot have the same key and different storage resolutions. The Embedded Metric Format supports a maximum of 100 values per key.

Requirements:

• Name Length 1-255 characters

• Name must be ASCII characters only

• Values must be in the range of 8.515920e-109 to 1.174271e+108. In addition, special values (for example, NaN, +Infinity, -Infinity) are not supported.

• Metrics must meet CloudWatch Metrics requirements, otherwise a InvalidMetricException will be thrown. See MetricDatum for valid values.

• Storage Resolution

An OPTIONAL value representing the storage resolution for the corresponding metric. Setting this to High specifies this metric as a high-resolution metric, so that CloudWatch stores the metric with sub-minute resolution down to one second. Setting this to Standard specifies this metric as a standard-resolution metric, which CloudWatch stores at 1-minute resolution. If a value is not provided, then a default value of Standard is assumed. See Cloud Watch High-Resolution metrics

Examples:

// Standard Resolution example
putMetric("Latency", 200, Unit.MILLISECONDS)
putMetric("Latency", 201, Unit.MILLISECONDS, StorageResolution.STANDARD)

// High Resolution example
putMetric("Memory.HeapUsed", 1600424.0, Unit.BYTES, StorageResolution.HIGH);

• MetricsLogger putProperty(String key, Object value )

Adds or updates the value for a given property on this context. This value is not submitted to CloudWatch Metrics but is searchable by CloudWatch Logs Insights. This is useful for contextual and potentially high-cardinality data that is not appropriate for CloudWatch Metrics dimensions.

Requirements:

• Length 1-255 characters

Examples:

putProperty("RequestId", "422b1569-16f6-4a03-b8f0-fe3fd9b100f8");
putProperty("InstanceId", "i-1234567890");
putProperty("Device", new HashMap<String, String>() {{
		put("Id", "61270781-c6ac-46f1-baf7-22c808af8162");
		put("Name", "Transducer");
		put("Model", "PT-1234");
	}}
);

• MetricsLogger putDimensions(DimensionSet dimensions)

Adds a new set of dimensions that will be associated with all metric values.

WARNING: Each dimension set will result in a new CloudWatch metric (even dimension sets with the same values). If the cardinality of a particular value is expected to be high, you should consider using setProperty instead.

Requirements:

• Length 1-255 characters
• ASCII characters only
• Dimensions must meet CloudWatch Dimension requirements, otherwise a InvalidDimensionException or DimensionSetExceededException will be thrown. See Dimension for valid values.

Examples:

putDimensions(DimensionSet.of("Operation", "Aggregator"))
putDimensions(DimensionSet.of("Operation", "Aggregator", "DeviceType", "Actuator"))

• MetricsLogger setDimensions(DimensionSet... dimensionSets)
• MetricsLogger setDimensions(boolean useDefault, DimensionSet... dimensionSets)

Explicitly override all dimensions. This will remove the default dimensions unless useDefault is set to true.

WARNING:Each dimension set will result in a new CloudWatch metric (even dimension sets with the same values). If the cardinality of a particular value is expected to be high, you should consider using setProperty instead.

Requirements:

• Length 1-255 characters
• ASCII characters only
• Dimensions must meet CloudWatch Dimension requirements, otherwise a InvalidDimensionException or DimensionSetExceededException will be thrown. See Dimension for valid values.

Examples:

setDimensions(DimensionSet.of(
	"Operation", "Aggregator",
	"DeviceType", "Actuator")
)

To create an aggregate metric across all dimensions, you can use an empty DimensionsSet:

setDimensions(
	DimensionSet.of(
		"Operation", "Aggregator",
		"DeviceType", "Actuator"),
	new DimensionSet()
)

• MetricsLogger resetDimensions(boolean useDefault)

Explicitly clear all custom dimensions. The behavior of whether default dimensions should be used can be configured by the input parameter.

Examples:

resetDimensions(false)  // this will clear all custom dimensions as well as disable default dimensions

• MetricsLogger setNamespace(String value)

Sets the CloudWatch namespace that extracted metrics should be published to. If not set, a default value of aws-embedded-metrics will be used.

Requirements:

• Name Length 1-255 characters
• Name must be ASCII characters only
• Namespace must meet CloudWatch requirements, otherwise a InvalidNamespaceException will be thrown. See Namespaces for valid values.

Examples:

setNamespace("MyApplication")

• MetricsLogger setTimestamp(Instant timestamp)

Sets the timestamp of the metrics. If not set, current time of the client will be used.

Timestamp must meet CloudWatch requirements, otherwise a InvalidTimestampException will be thrown. See Timestamps for valid values.

Examples:

setTimestamp(Instant.now())

• flush()

Flushes the current MetricsContext to the configured sink and resets all properties and metric values. The namespace and default dimensions will be preserved across flushes. Custom dimensions are preserved by default, but this behavior can be disabled by invoking setFlushPreserveDimensions(false), so that no custom dimensions would be preserved after each flushing thereafter.

Example:

flush();  // default dimensions and custom dimensions will be preserved after each flush()

setFlushPreserveDimensions(false);
flush();  // only default dimensions will be preserved after each flush()

setFlushPreserveDimensions(false);
flush();
resetDimensions(false);  // default dimensions are disabled; no dimensions will be preserved after each flush()

Configuration

All configuration values can be set using environment variables with the prefix (AWS_EMF_). Configuration should be performed as close to application start up as possible.

ServiceName: Overrides the name of the service. For services where the name cannot be inferred (e.g. Java process running on EC2), a default value of Unknown will be used if not explicitly set.

Requirements:

• Name Length 1-255 characters
• Name must be ASCII characters only

Example:

# in process
import software.amazon.cloudwatchlogs.emf.config.Configuration;
import software.amazon.cloudwatchlogs.emf.config.EnvironmentConfigurationProvider;

Configuration config = EnvironmentConfigurationProvider.getConfig();
config.setServiceName("MyApp")

# environment
AWS_EMF_SERVICE_NAME="MyApp"

ServiceType: Overrides the type of the service. For services where the type cannot be inferred (e.g. Java process running on EC2), a default value of Unknown will be used if not explicitly set.

Requirements:

• Name Length 1-255 characters
• Name must be ASCII characters only

Example:

# in process
import software.amazon.cloudwatchlogs.emf.config.Configuration;
import software.amazon.cloudwatchlogs.emf.config.EnvironmentConfigurationProvider;

Configuration config = EnvironmentConfigurationProvider.getConfig();
config.setServiceType("JavaWebApp")

# environment
AWS_EMF_SERVICE_TYPE="JavaWebApp"

LogGroupName: For agent-based platforms, you may optionally configure the destination log group that metrics should be delivered to. This value will be passed from the library to the agent in the Embedded Metric payload. If a LogGroup is not provided, the default value will be derived from the service name: -metrics

Requirements:

• Name Length 1-512 characters
• Log group names consist of the following characters: a-z, A-Z, 0-9, '_' (underscore), '-' (hyphen), '/' (forward slash), and '.' (period). Pattern: [.-_/#A-Za-z0-9]+

Example:

# in process
import software.amazon.cloudwatchlogs.emf.config.Configuration;
import software.amazon.cloudwatchlogs.emf.config.EnvironmentConfigurationProvider;

Configuration config = EnvironmentConfigurationProvider.getConfig();
config.setLogGroupName("LogGroupName")

# environment
AWS_EMF_LOG_GROUP_NAME="LogGroupName"

LogStreamName: For agent-based platforms, you may optionally configure the destination log stream that metrics should be delivered to. This value will be passed from the library to the agent in the Embedded Metric payload. If a LogGroup is not provided, the default value will be derived by the agent (this will likely be the hostname).

Requirements:

• Name Length 1-512 characters
• The ':' (colon) and '*' (asterisk) characters are not allowed. Pattern: [^:]*

Example:

# in process
import software.amazon.cloudwatchlogs.emf.config.Configuration;
import software.amazon.cloudwatchlogs.emf.config.EnvironmentConfigurationProvider;

Configuration config = EnvironmentConfigurationProvider.getConfig();
config.setLogStreamName(LogStreamName))

# environment
AWS_EMF_LOG_STREAM_NAME="LogStreamName"

EnvironmentOverride: Short circuit auto-environment detection by explicitly defining how events should be sent. This is not supported through programmatic access.

Valid values include:

• Local: no decoration and sends over stdout
• Lambda: decorates logs with Lambda metadata and sends over stdout
• Agent: no decoration and sends over TCP
• EC2: decorates logs with EC2 metadata and sends over TCP
• ECS: decorates logs with ECS metadata and sends over TCP

Example:

AWS_EMF_ENVIRONMENT="Local"

AgentEndpoint: For agent-based platforms, you may optionally configure the endpoint to reach the agent on.

Example:

// in process
import software.amazon.cloudwatchlogs.emf.config.Configuration;
import software.amazon.cloudwatchlogs.emf.config.EnvironmentConfigurationProvider;

Configuration config = EnvironmentConfigurationProvider.getConfig();
config.setAgentEndpoint("udp://127.0.0.1:1000");

// environment
AWS_EMF_AGENT_ENDPOINT="udp://127.0.0.1:1000"

WriteToStdout: For agent-based platforms, setting this configuration to true will make the MetricsLogger write to stdout rather than sending them to the agent. The default value for this configuration is false. This configuration has no effect for non-agent-based platforms.

If an EnvironmentOverride is provided, this configuration will apply to the overriden environment if the environment is an agent-based platform

Example:

// in process
import software.amazon.cloudwatchlogs.emf.config.Configuration;
import software.amazon.cloudwatchlogs.emf.config.EnvironmentConfigurationProvider;

Configuration config = EnvironmentConfigurationProvider.getConfig();
config.setShouldWriteToStdout(true);

// environment
AWS_EMF_WRITE_TO_STDOUT="true"

Thread-safety

Internal Synchronization

The MetricsLogger class is thread-safe. Specifically, the generalized multi-threading use cases for this library are:

1. Collect some metrics or metadata on a single MetricsLogger; Pass the logger into one or more async contexts where new metrics or metadata can be added concurrently; Join the async contexts (e.g. Future.get()) and flush the metrics.
2. Collect some metrics or metadata on a single MetricsLogger; Pass the logger into an async context; Flush from the async context concurrently.

Thread-safety for the first use case is achieved by introducing concurrent internal data structures and atomic operations associated with these models, to ensure the access to shared mutable resources are always synchronized.

Thread-safety for the second use case is achieved by using a ReentrantReadWriteLock. This lock is used to create an internal sync context for flush() method in multi-threading situations. flush() acquires write lock, while other methods (which have access to mutable shared data with flush()) acquires read lock. This makes sure flush() is always executed exclusively, while other methods can be executed concurrently.

Use Cases that are Not Covered

With all the internal synchronization measures, however, there're still certain multi-threading use cases that are not covered by this library, which might require external synchronizations or other protection measures. This is due to the fact that the execution order of APIs are not determined in async contexts. For example, if user needs to associate a given set of properties with a metric in each thread, the results are not guaranteed since the execution order of putProperty() is not determined across threads. In such cases, we recommend using a different MetricsLogger instance for different threads, so that no resources are shared and no thread-safety problem would ever happen. Note that this can often be simplified by using a ThreadLocal variable.

Examples

Check out the examples directory to get started.

Development

Building

Gradle is used to build the project. Run this command to build the project:

./gradlew build

Testing

We have 2 different types of tests:

1. Unit tests. The command to run these tests

   ./gradlew test

2. Integration tests. These tests require Docker to run the CloudWatch Agent and valid AWS credentials. Tests can be run by:

   export AWS_ACCESS_KEY_ID=YOUR_ACCESS_KEY_ID
   export AWS_SECRET_ACCESS_KEY=YOUR_ACCESS_KEY
   export AWS_REGION=us-west-2
   ./gradlew integ

   NOTE: You need to replace the access key id and access key with your own AWS credentials.

Formatting

We use Spotless plugin for style-checking. To auto fix code style, run

./gradlew :spotlessApply

Benchmark

We use JMH as our framework for concurrency performance benchmarking. Benchmarks can be run by:

./gradlew jmh

To run a single benchmark, consider using JMH plugins. For example, JMH plugin for IntelliJ IDEA

License

This project is licensed under the Apache-2.0 License.