AWS Embedded Metrics Package
Generate CloudWatch Metrics embedded within structured log events. The embedded metrics will be extracted so you can visualize and alarm on them for real-time incident detection. This allows you to monitor aggregated values while preserving the detailed event context that generated them.
Generate custom metrics across compute environments
Linking 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.
pip3 install aws-embedded-metrics
To get a metric logger, you can decorate your function with a metric_scope
:
from aws_embedded_metrics import metric_scope
from aws_embedded_metrics.storage_resolution import StorageResolution
@metric_scope
def my_handler(metrics):
metrics.put_dimensions({"Foo": "Bar"})
metrics.put_metric("ProcessingLatency", 100, "Milliseconds", StorageResolution.STANDARD)
metrics.put_metric("Memory.HeapUsed", 1600424.0, "Bytes", StorageResolution.HIGH)
metrics.set_property("AccountId", "123456789012")
metrics.set_property("RequestId", "422b1569-16f6-4a03")
metrics.set_property("DeviceId", "61270781-c6ac-46f1")
return {"message": "Hello!"}
The MetricsLogger
is the interface you will use to publish embedded metrics.
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 same key and different storage resolution. The Embedded Metric Format supports a maximum of 100 values per key. If more metric values are added than are supported by the format, the logger will be flushed to allow for new metric values to be captured.
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 InvalidMetricError
will be thrown. See MetricDatum for valid values.
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
put_metric("Latency", 200, "Milliseconds")
put_metric("Latency", 201, "Milliseconds", StorageResolution.STANDARD)
# High Resolution example
put_metric("Memory.HeapUsed", 1600424.0, "Bytes", StorageResolution.HIGH)
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:
Examples:
set_property("RequestId", "422b1569-16f6-4a03-b8f0-fe3fd9b100f8")
set_property("InstanceId", "i-1234567890")
set_property("Device", {
"Id": "61270781-c6ac-46f1-baf7-22c808af8162",
"Name": "Transducer",
"Model": "PT-1234"
})
Adds a new set of dimensions that will be associated to all metric values.
WARNING: Every distinct value will result in a new CloudWatch Metric.
If the cardinality of a particular value is expected to be high, you should consider
using setProperty
instead.
Requirements:
InvalidDimensionError
or DimensionSetExceededError
will be thrown. See Dimensions for valid values.Examples:
put_dimensions({ "Operation": "Aggregator" })
put_dimensions({ "Operation": "Aggregator", "DeviceType": "Actuator" })
Explicitly override all dimensions. By default, this will disable the default dimensions, but can be configured using the keyword-only parameter use_default
.
WARNING: Every distinct value will result in a new CloudWatch Metric.
If the cardinality of a particular value is expected to be high, you should consider
using setProperty
instead.
Requirements:
InvalidDimensionError
or DimensionSetExceededError
will be thrown. See Dimensions for valid values.Examples:
set_dimensions(
{ "Operation": "Aggregator" },
{ "Operation": "Aggregator", "DeviceType": "Actuator" }
)
set_dimensions(
{ "Operation": "Aggregator" },
use_default=True # default dimensions would be enabled
)
Explicitly clear all custom dimensions. The behavior of whether default dimensions should be used can be configured with the use_default
parameter.
Examples:
reset_dimensions(False) # this will clear all custom dimensions as well as disable default dimensions
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:
InvalidNamespaceError
will be thrown. See Namespaces for valid values.Examples:
set_namespace("MyApplication")
Sets the timestamp of the metrics. If not set, current time of the client will be used.
Timestamp must meet CloudWatch requirements, otherwise a InvalidTimestampError will be thrown. See Timestamps for valid values.
Examples:
set_timestamp(datetime.datetime.now())
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 not preserved by default, but this behavior can be changed by setting logger.flush_preserve_dimensions = True
, so that custom dimensions would be preserved after each flushing thereafter.
Example:
logger.flush() # only default dimensions will be preserved after each flush()
logger.flush_preserve_dimensions = True
logger.flush() # custom dimensions and default dimensions will be preserved after each flush()
logger.reset_dimensions(False)
logger.flush() # default dimensions are disabled; no dimensions will be preserved after each flush()
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:
Example:
# in process
from aws_embedded_metrics.config import get_config
Config = get_config()
Config.service_name = "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:
Example:
# in process
from aws_embedded_metrics.config import get_config
Config = get_config()
Config.service_type = "NodeJSWebApp"
# environment
AWS_EMF_SERVICE_TYPE = NodeJSWebApp
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:
Requirements:
Example:
# in process
from aws_embedded_metrics.config import get_config
Config = get_config()
Config.log_group_name = "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:
Example:
# in process
from aws_embedded_metrics.config import get_config
Config = get_config()
Config.log_stream_name = "LogStreamName"
# environment
AWS_EMF_LOG_STREAM_NAME = LogStreamName
NameSpace: Overrides the CloudWatch namespace. If not set, a default value of aws-embedded-metrics will be used.
Requirements:
Example:
# in process
from aws_embedded_metrics.config import get_config
Config = get_config()
Config.namespace = "MyApplication"
# environment
AWS_EMF_NAMESPACE = MyApplication
DISABLE_METRIC_EXTRACTION: Disables extraction of metrics by CloudWatch, by omitting EMF metadata from serialized log records.
Example:
# in process
from aws_embedded_metrics.config import get_config
Config = get_config()
Config.disable_metric_extraction = True
# environment
AWS_EMF_DISABLE_METRIC_EXTRACTION = true
Check out the examples directory to get started.
pip install tox
tox
export AWS_ACCESS_KEY_ID=
export AWS_SECRET_ACCESS_KEY=
export AWS_REGION=us-west-2
./bin/run-integ-tests.sh
This project is licensed under the Apache-2.0 License.