Specification: Metrics for Eclipse MicroProfile

Version: {revnumber}

Status: {revremark}

Release: {revdate}

Copyright (c) 2016-2017 Eclipse Microprofile Contributors:
Heiko W. Rupp, Raymond Lam, David Chan, Don Bourne, Antonin Stefanutti, Brennan Nichyporuk, Mike Croft, Werner Keil

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

   http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

1. Introduction

To ensure reliable operation of software it is necessary to monitor essential system parameters. This enhancement proposes the addition of well-known monitoring endpoints and metrics for each process adhering to the Eclipse MicroProfile standard.

This proposal does not talk about health checks. There is a separate specification for Health Checks.

1.1. Motivation

Reliable service of a platform needs monitoring. There is already JMX as standard to expose metrics, but remote-JMX is not easy to deal with and especially does not fit well in a polyglot environment where other services are not running on the JVM. To enable monitoring in an easy fashion it is necessary that all MicroProfile implementations follow a certain standard with respect to (base) API path, data types involved, always available metrics and return codes used.

1.2. Difference to health checks

Health checks are primarily targeted at a quick yes/no response to the question "Is my application still running ok?". Modern systems that schedule the starting of applications (e.g. Kubernetes) use this information to restart the application if the answer is 'no'.

Metrics on the other hand can help to determine the health. Beyond this they serve to pinpoint issues, provide long term trend data for capacity planning and pro-active discovery of issues (e.g. disk usage growing without bounds). Metrics can also help those scheduling systems decide when to scale the application to run on more or fewer machines.

1.3. Major changes to previous versions

  • Since 1.0

    • Improved TCK

    • org.eclipse.microprofile.metrics.MetricRegistry.register(String name, Metric, Metadata) is deprecated. Use org.eclipse.microprofile.metrics.MetricRegistry.register(Metadata, Metric) instead, where Metadata already has a field for the name.

    • Global tags are now supplied via the means of MicroProfile Config (the env variable is still valid).

    • Annotations and Metadata can now have a flag reusable that indicates that the metric name can be registered more than once. Default is false as in Metrics 1.0. See Reusing of Metrics.

2. Architecture

This chapter describes the architectural overview of how metrics are setup, stored and exposed for consumption. This chapter also lists the various scopes of metrics.

See section Required Metrics for more information regarding metrics that are required for each vendor.

See section Application Metrics Programming Model for more information regarding the application metrics programming model.

2.1. Metrics Setup

Metrics that are exposed need to be configured in the server. On top of the pure metrics, metadata needs to be provided.

The following three sets of sub-resource (scopes) are exposed.

  • base: metrics that all MicroProfile vendors have to provide

  • vendor: vendor specific metrics (optional)

  • application: application-specific metrics (optional)

It is expected that a future version of this specification will also have a sub-resource for integrations with other specifications of MicroProfile.

2.1.1. Scopes

Required Base metrics

Required base metrics describe a set of metrics that all MicroProfile-compliant servers have to provide. Each vendor can implement the set-up of the metrics in the base scope in a vendor-specific way. The metrics can be hard coded into the server or read from a configuration file or supplied via the Java-API described in Application Metrics Programming Model. The Appendix shows a possible data format for such a configuration. The configuration and set up of the base scope is thus an implementation detail and is not expected to be portable across vendors.

Section Required Metrics lists the required metrics. This list also includes a few items marked as optional. These are listed here as they are dependent on the underlying JVM and not the server and thus fit better in base scope than the vendor one.

Required base metrics are exposed under /metrics/base.

Application metrics

Application specific metrics can not be baked into the server as they are supposed to be provided by the application at runtime. Therefore a Java API is provided. Application specific metrics are supposed to be portable to other implementations of the MicroProfile. That means that an application written to this specification which exposes metrics, can expose the same metrics on a different compliant server without change.

Details of this Java API are described in Application Metrics Programming Model.

Application specific metrics are exposed under /metrics/application.

Vendor specific Metrics

It is possible for MicroProfile server implementors to supply their specific metrics data on top of the basic set of required metrics. Vendor specific metrics are exposed under /metrics/vendor.

Examples for vendor specific data could be metrics like:

  • OSGi statistics if the MicroProfile-enabled container internally runs on top of OSGi.

  • Statistics of some internal caching modules

Vendor specific metrics are not supposed to be portable between different implementations of MicroProfile servers, even if they are compliant with the same version of this specification.

2.1.2. Supplying of Tags

Tags (or labels) play an important role in modern microservices and microservice scheduling systems (like e.g. Kubernetes). Application code can run on any node and can be re-scheduled to a different node at any time. Each container in such an environment gets its own ID; when the container is stopped and a new one started for the same image, it will get a different id. The classical mapping of host/node and application runtime on it, therefore no longer works.

Tags have taken over the role to, for example, identify an application (app=myShop), the tier inside the application (tier=database or tier=app_server) and also the node/container id. Aggregation of metrics can then work over label queries (Give me the API hit count for app=myShop && tier=app_server).

Tags can be supplied in two ways:

The use of MicroProfile-Config is mandatory in MicroProfile-Metrics 1.1 even if it only serves to provide the global tags that existed already in Metrics 1.0. Future version of the spec will introduce more config options.
Set up global tags via environment
export MP_METRICS_TAGS=app=shop,tier=integration

Global tags and tags set in metric metadata are included in the output returned from the REST API.

2.1.3. Metadata

Metadata can be specified for metrics in any scope. For base metrics, metadata must be provided by the implementation. Metadata is exposed by the REST handler.

While technically it is possible to expose metrics without (some) of the metadata, it helps tooling and also operators when correct metadata is provided, as this helps getting a context and an explanation of the metric.

The Metadata:

  • unit: a fixed set of string units

  • type:

    • counter: an incrementally increasing or decreasing numeric value (e.g. total number of requests received or total number of concurrently active HTTP sessions).

    • gauge: a metric that is sampled to obtain its value (e.g. cpu temperature or disk usage).

    • meter: a metric which tracks mean throughput and one-, five-, and fifteen-minute exponentially-weighted moving average throughput.

    • histogram: a metric which calculates the distribution of a value.

    • timer: a metric which aggregates timing durations and provides duration statistics, plus throughput statistics.

  • description (optional): A human readable description of the metric.

  • displayName (optional): A human readable name of the metric for display purposes if the metric name is not human readable. This could e.g. be the case when the metric name is a uuid.

  • tags (optional): A list of key=value pairs, which are separated by comma. See also Supplying of Tags.

  • reusable (optional): If set to true, then it is allowed to register a metric multiple times under the same name. Note that all such instances must set reusable to true. Default is false. See Reusing of Metrics for more details.

Metadata must not change over the lifetime of a process (i.e. it is not allowed to return the units as seconds in one retrieval and as hours in a subsequent one). The reason behind it is that e.g. a monitoring agent on Kubernetes may read the metadata once it sees the new container and store it. It may not periodically re-query the process for the metadata.

In fact, metadata should not change during the life-time of the whole container image or an application, as all containers spawned from it will be "the same" and form part of an app, where it would be confusing in an overall view if the same metric has different metadata.

2.2. Metric Registry

The MetricRegistry stores the metrics and metadata information. There is one MetricRegistry instance for each of the scopes listed in Scopes.

Metrics can be added to or retrieved from the registry either using the @Metric annotation (see Metrics Annotations) or using the MetricRegistry object directly.

2.2.1. Reusing of Metrics

By default it is not allowed to register more than one metric under a certain name in a scope. This is done to prevent hard to spot copy & paste errors, where for example all methods of a Jax-Rs class are marked with @Timed(name="myApp", absolute=true).

If this behaviour is required, then it is possible to mark all such instances as reusable by passing the respective flag via Metadata or field in the Annotation. Gauges are not reusable.

The implementation must throw an 'IllegalArgumentException' during a metric registration call when the call would result in the reuse of a metric where that metric was either previously declared not reusable or where the registration call itself declares the metric to not be reusable.

Only metrics of the same type can be reused under the same name. Trying to reuse a name for different types will result in an IllegalArgumentException.

If you want to re-use a metric name, then you need to also explicitly set the name field OR set absolute to true and have multiple methods annotated as metric that have the same method name.
Example of reused counters
    @Counted(name = "countMe", absolute = true, reusable = true)
    public void countMeA() { }

    @Counted(name = "countMe", absolute = true, reusable = true)
    public void countMeB() { }

In the above examples both countMeA() and countMeB() will share a single Counter with registered name countMe in application scope.

2.3. Exposing metrics via REST API

Data is exposed via REST over HTTP under the /metrics base path in two different data formats for GET requests:

  • JSON format - used when the HTTP Accept header matches application/json.

  • Prometheus text format - default response format when the HTTP Accept header does not match any more specific media type like application/json.

    Future versions may allow for more export formats that are triggered by their specific media type. The Prometheus text format will stay as fall-back.

Formats are detailed below.

Data access must honour the HTTP response codes, especially

  • 200 for successful retrieval of an object

  • 204 when retrieving a subtree that would exist, but has no content. E.g. when the application-specific subtree has no application specific metrics defined.

  • 404 if a directly-addressed item does not exist. This may be a non-existing sub-tree or non-existing object

  • 406 if the HTTP Accept Header in the request cannot be handled by the server.

  • 500 to indicate that a request failed due to "bad health". The body SHOULD contain details if possible { "details": <text> }

The API MUST NOT return a 500 Internal Server Error code to represent a non-existing resource.

Table 1. Supported REST endpoints
Endpoint Request Type Supported Formats Description

/metrics

GET

JSON, Prometheus

Returns all registered metrics

/metrics/<scope>

GET

JSON, Prometheus

Returns metrics registered for the respective scope. Scopes are listed in Metrics Setup

/metrics/<scope>/<metric_name>

GET

JSON, Prometheus

Returns the metric that matches the metric name for the respective scope

/metrics

OPTIONS

JSON

Returns all registered metrics' metadata

/metrics/<scope>

OPTIONS

JSON

Returns metrics' metadata registered for the respective scope. Scopes are listed in Metrics Setup

/metrics/<scope>/<metric_name>

OPTIONS

JSON

Returns the metric’s metadata that matches the metric name for the respective scope

The implementation must return a 406 response code if the request’s HTTP Accept header for an OPTIONS request does not match application/json.

3. REST endpoints

This section describes the REST-api, that monitoring agents would use to retrieve the collected metrics. (Java-) methods mentioned refer to the respective Objects in the Java API. See also Application Metrics Programming Model

3.1. JSON format

  • When using JSON format, the REST API will respond to GET requests with data formatted in a tree like fashion with sub-trees for the sub-resources. A sub-tree that does not contain data must be omitted.

  • A 'shadow tree' that responds to OPTIONS will provide the metadata.

REST-API Objects

API-objects MAY include one or more metrics as in

{
  "thread.count" : 33,
  "thread.max.count" : 47,
  "memory.maxHeap" : 3817863211,
  "memory.usedHeap" : 16859081,
  "memory.committedHeap" : 64703546
}

or

{
  "hitCount": 45
}

In case /metrics is requested, then the data for the scopes are wrapped in the scope name:

{
  "application": {
    "hitCount": 45
  },
  "base": {
     "thread.count" : 33,
     "thread.max.count" : 47
  },
  "vendor": {...}
}

3.1.1. Gauge JSON Format

The value of the gauge must be equivalent to a call to the instance Gauge’s getValue().

Example Gauge JSON GET Response
{
  "responsePercentage": 48.45632
}

3.1.2. Counter JSON Format

The value of the counter must be equivalent to a call to the instance Counter’s getCount().

Example Counter JSON GET Response
{
  "hitCount": 45
}

3.1.3. Meter JSON Format

Meter is a complex metric type comprised of multiple key/values. The format is specified by the table below.

Table 2. JSON mapping for a Meter metric
JSON Key Value (Equivalent Meter method)

count

getCount()

meanRate

getMeanRate()

oneMinRate

getOneMinuteRate()

fiveMinRate

getFiveMinuteRate()

fifteenMinRate

getFifteenMinuteRate()

Example Meter JSON GET Response
{
  "requests": {
    "count": 29382,
    "meanRate": 12.223,
    "oneMinRate": 12.563,
    "fiveMinRate": 12.364,
    "fifteenMinRate": 12.126,
  }
}

3.1.4. Histogram JSON Format

Histogram is a complex metric type comprised of multiple key/values. The format is specified by the table below.

Table 3. JSON mapping for a Histogram metric
JSON Key Value (Equivalent Histogram method)

count

getCount()

min

getSnapshot().getMin()

max

getSnapshot().getMax()

mean

getSnapshot().getMean()

stddev

getSnapshot().getStdDev()

p50

getSnapshot().getMedian()

p75

getSnapshot().get75thPercentile()

p95

getSnapshot().get95thPercentile()

p98

getSnapshot().get98thPercentile()

p99

getSnapshot().get99thPercentile()

p999

getSnapshot().get999thPercentile()

Example Histogram JSON GET Response
{
  "daily_value_changes": {
    "count":2,
    "min":-1624,
    "max":26,
    "mean":-799.0,
    "stddev":825.0,
    "p50":26.0,
    "p75":26.0,
    "p95":26.0,
    "p98":26.0,
    "p99":26.0,
    "p999":26.0
  }
}

3.1.5. Timer JSON Format

Timer is a complex metric type comprised of multiple key/values. The format is specified by the table below.

Table 4. JSON mapping for a Timer metric
JSON Key Value (Equivalent Timer method)

count

getCount()

meanRate

getMeanRate()

oneMinRate

getOneMinuteRate()

fiveMinRate

getFiveMinuteRate()

fifteenMinRate

getFifteenMinuteRate()

min

getSnapshot().getMin()

max

getSnapshot().getMax()

mean

getSnapshot().getMean()

stddev

getSnapshot().getStdDev()

p50

getSnapshot().getMedian()

p75

getSnapshot().get75thPercentile()

p95

getSnapshot().get95thPercentile()

p98

getSnapshot().get98thPercentile()

p99

getSnapshot().get99thPercentile()

p999

getSnapshot().get999thPercentile()

Example Timer JSON GET Response
{
  "responseTime": {
    "count": 29382,
    "meanRate":12.185627192860734,
    "oneMinRate": 12.563,
    "fiveMinRate": 12.364,
    "fifteenMinRate": 12.126,
    "min":169916,
    "max":5608694,
    "mean":415041.00024926325,
    "stddev":652907.9633011606,
    "p50":293324.0,
    "p75":344914.0,
    "p95":543647.0,
    "p98":2706543.0,
    "p99":5608694.0,
    "p999":5608694.0
  }
}

3.1.6. Metadata

Metadata is exposed in a tree-like fashion with sub-trees for the sub-resources mentioned previously.

Example:

If GET /metrics/base/fooVal exposes:

{"fooVal": 12345}

then OPTIONS /metrics/base/fooVal will expose:

{
  "fooVal": {
    "unit": "milliseconds",
    "type": "gauge",
    "description": "The size of foo after each request",
    "displayName": "Size of foo",
    "tags": "app=webshop"
  }
}

If GET /metrics/base exposes multiple values like this:

Example of exposed metrics data
{
  "fooVal": 12345,
  "barVal": 42
}

then OPTIONS /metrics/base exposes:

Example of JSON output of Metadata
{
  "fooVal": {
    "unit": "milliseconds",
    "type": "gauge",
    "description": "The average duration of foo requests during last 5 minutes",
    "displayName": "Duration of foo",
    "tags": "app=webshop"
  },
  "barVal": {
    "unit": "megabytes",
    "type": "gauge",
    "tags": "component=backend,app=webshop"
  }
}

3.2. Prometheus format

Data is exposed in the Prometheus text format, version 0.0.4 as described in Prometheus text format.

The metadata will be included as part of the normal Prometheus text format. Unlike the JSON format, the text format does not support OPTIONS requests.

Users that want to write tools to transform the metadata can still request the metadata via OPTIONS request and application/json media type.

The above json example would look like this in Prometheus format

Example of Prometheus output
# TYPE base:foo_val_seconds gauge   (1)
# HELP base:foo_val_seconds The average duration of foo requests during last 5 minutes (2)
base:foo_val_seconds{app="webshop"} 12.345  (3)
# TYPE base:bar_val_bytes gauge (1)
base:bar_val_bytes{component="backend", app="webshop"} 42000 (3)
1 Metric names are turned from camel case into snake_case.
2 The description goes into the HELP line
3 Metric names gets the base unit of the family appended with _ and defined labels. Values are scaled accordingly. See Handling of units

3.2.1. Translation rules for metric names

Prometheus text format does not allow for all characters and adds the base unit of a family to the name.

  • Dot (.), Space ( ), Dash (-) are translated to underscore (_).

  • Scope is always specified at the start of the metric name.

  • Scope and name are separated by colon (:).

  • camelCase is translated to camel_case

  • Double underscore is translated to single underscore

  • Colon-underscore (:_) is translated to single colon

  • The unit is appended to the name, separated by underscore. See Handling of units

3.2.2. Handling of tags

Metric tags are appended to the metric name in curly braces { and } and are separated by comma. Each tag is a key-value-pair in the format of <key>="<value>" (the quotes around the value are required).

3.2.3. Handling of units

The Prometheus text format adheres to using "base units" when creating the HTTP response. Due to the different context of each metric type, certain metrics' values must be converted to the respective "base unit" when responding to Prometheus requests. For example, times in milliseconds must be divided by 1000 and displayed in the base unit (seconds).

The following sections outline how each metric type is handled:

Gauges and Histograms

The metric name and values for Gauge and Histogram are converted to the "base unit" in respect to the unit value in the Metadata.

  • If the Metadata is empty, NONE, or null, the metric name is used as is without appending the unit name and no scaling is applied.

  • If the metric’s metadata contains a known unit, as defined in the MetricUnits class, the Prometheus value should be scaled to the base unit of the respective family. The name of the base unit is appended to the metric name delimited by underscores (_).

  • If the unit is specified and is not defined in the MetricUnits class, the value is not scaled but the unit is still appended to the metric name delimited by underscores (_).

Unit families and their base units are described under Prometheus metric names, Base units.

Families and Prometheus base units are:

Family Base unit

Bits

bytes

Bytes

bytes

Time

seconds

Percent

ratio (normally ratio is A_per_B, but there are exceptions like disk_usage_ratio)

Counters

Counter metrics are considered dimensionless. The implementation must not append the unit name to the metric name and must not scale the value.

Meters and Timers

Meter and Timer have fixed units as described below regardless of the unit value in the Metadata.

3.2.4. Gauge Prometheus Text Format

The value of the gauge must be the value of getValue() with appropriate naming/scaling based on Handling of units

Example Prometheus text format for a Gauge in dollars.
# TYPE application:cost_dollars gauge
# HELP application:cost_dollars The running cost of the server in dollars.
application:cost_dollars 80

3.2.5. Counter Prometheus Text Format

The value of the counter must be the value of getCount().

Implementors must not convert the unit of Counters or append the unit suffix to the metric.
Example Prometheus text format for a Counter.
# TYPE application:visitors counter
# HELP application:visitors The number of unique visitors
application:visitors 80

3.2.6. Meter Prometheus Text Format

Meter is a complex metric type comprised of multiple key/values. Each key will require a suffix to be appended to the metric name. The format is specified by the table below.

The # HELP description line is only required for the total value as shown below.

Table 5. Prometheus text mapping for a Meter metric
Suffix{label} TYPE Value (Meter method) Units

total

Counter

getCount()

N/A

rate_per_second

Gauge

getMeanRate()

PER_SECOND

one_min_rate_per_second

Gauge

getOneMinuteRate()

PER_SECOND

five_min_rate_per_second

Gauge

getFiveMinuteRate()

PER_SECOND

fifteen_min_rate_per_second

Gauge

getFifteenMinuteRate()

PER_SECOND

Example Prometheus text format for a Meter
# TYPE application:requests_total counter
# HELP application:requests_total Tracks the number of requests to the server
application:requests_total 29382
# TYPE application:requests_rate_per_second gauge
application:requests_rate_per_second 12.223
# TYPE application:requests_one_min_rate_per_second gauge
application:requests_one_min_rate_per_second 12.563
# TYPE application:requests_five_min_rate_per_second gauge
application:requests_five_min_rate_per_second 12.364
# TYPE application:requests_fifteen_min_rate_per_second gauge
application:requests_fifteen_min_rate_per_second 12.126

3.2.7. Histogram Prometheus Text Format

Histogram is a complex metric type comprised of multiple key/values. Each key will require a suffix to be appended to the metric name with appropriate naming/scaling based on Handling of units. The format is specified by the table below.

The # HELP description line is only required for the summary value as shown below.

Table 6. Prometheus text mapping for a Histogram metric
Suffix{label} TYPE Value (Histogram method) Units

min_<units>

Gauge

getSnapshot().getMin()

<units>1

max_<units>

Gauge

getSnapshot().getMax()

<units>1

mean_<units>

Gauge

getSnapshot().getMean()

<units>1

stddev_<units>

Gauge

getSnapshot().getStdDev()

<units>1

<units>_count2

Summary

getCount()

N/A

<units>{quantile="0.5"}2

Summary

getSnapshot().getMedian()

<units>1

<units>{quantile="0.75"}2

Summary

getSnapshot().get75thPercentile()

<units>1

<units>{quantile="0.95"}2

Summary

getSnapshot().get95thPercentile()

<units>1

<units>{quantile="0.98"}2

Summary

getSnapshot().get98thPercentile()

<units>1

<units>{quantile="0.99"}2

Summary

getSnapshot().get99thPercentile()

<units>1

<units>{quantile="0.999"}2

Summary

getSnapshot().get999thPercentile()

<units>1

1 The implementation is expected to convert the result returned by the Histogram into the base unit (if known). The <unit> represents the base metric unit and is named according to Handling of units.

2 The summary type is a complex metric type for Prometheus which consists of the count and multiple quantile values.

Example Prometheus text format for a Histogram with unit bytes.
# TYPE application:file_sizes_mean_bytes gauge
application:file_sizes_mean_bytes 4738.231
# TYPE application:file_sizes_max_bytes gauge
application:file_sizes_max_bytes 31716
# TYPE application:file_sizes_min_bytes gauge
application:file_sizes_min_bytes 180
# TYPE application:file_sizes_stddev_bytes gauge
application:file_sizes_stddev_bytes 1054.7343037063602
# TYPE application:file_sizes_bytes summary
# HELP application:file_sizes_bytes Users file size
application:file_sizes_bytes_count 2037
application:file_sizes_bytes{quantile="0.5"} 4201
application:file_sizes_bytes{quantile="0.75"} 6175
application:file_sizes_bytes{quantile="0.95"} 13560
application:file_sizes_bytes{quantile="0.98"} 29643
application:file_sizes_bytes{quantile="0.99"} 31716
application:file_sizes_bytes{quantile="0.999"} 31716

3.2.8. Timer Prometheus Text Format

Timer is a complex metric type comprised of multiple key/values. Each key will require a suffix to be appended to the metric name. The format is specified by the table below.

The # HELP description line is only required for the summary value as shown below.

Table 7. Prometheus text mapping for a Timer metric
Suffix{label} TYPE Value (Timer method) Units

rate_per_second

Gauge

getMeanRate()

PER_SECOND

one_min_rate_per_second

Gauge

getOneMinuteRate()

PER_SECOND

five_min_rate_per_second

Gauge

getFiveMinuteRate()

PER_SECOND

fifteen_min_rate_per_second

Gauge

getFifteenMinuteRate()

PER_SECOND

min_seconds

Gauge

getSnapshot().getMin()

SECONDS1

max_seconds

Gauge

getSnapshot().getMax()

SECONDS1

mean_seconds

Gauge

getSnapshot().getMean()

SECONDS1

stddev_seconds

Gauge

getSnapshot().getStdDev()

SECONDS1

seconds_count2

Summary

getCount()

N/A

seconds{quantile="0.5"}2

Summary

getSnapshot().getMedian()

SECONDS1

seconds{quantile="0.75"}2

Summary

getSnapshot().get75thPercentile()

SECONDS1

seconds{quantile="0.95"}2

Summary

getSnapshot().get95thPercentile()

SECONDS1

seconds{quantile="0.98"}2

Summary

getSnapshot().get98thPercentile()

SECONDS1

seconds{quantile="0.99"}2

Summary

getSnapshot().get99thPercentile()

SECONDS1

seconds{quantile="0.999"}2

Summary

getSnapshot().get999thPercentile()

SECONDS1

1 The implementation is expected to convert the result returned by the Timer into seconds

2 The summary type is a complex metric type for Prometheus which consists of the count and multiple quantile values.

Example Prometheus text format for a Timer
# TYPE application:response_time_rate_per_second gauge
application:response_time_rate_per_second  0.004292520715985437
# TYPE application:response_time_one_min_rate_per_second gauge
application:response_time_one_min_rate_per_second  2.794076465421066E-14
# TYPE application:response_time_five_min_rate_per_second  gauge
application:response_time_five_min_rate_per_second  4.800392614619373E-4
# TYPE application:response_time_fifteen_min_rate_per_second  gauge
application:response_time_fifteen_min_rate_per_second  0.01063191047532505
# TYPE application:response_time_mean_seconds gauge
application:response_time_mean_seconds 0.000415041
# TYPE application:response_time_max_seconds gauge
application:response_time_max_seconds 0.0005608694
# TYPE application:response_time_min_seconds gauge
application:response_time_min_seconds 0.000169916
# TYPE application:response_time_stddev_seconds gauge
application:response_time_stddev_seconds 0.000652907
# TYPE application:response_time_seconds summary
# HELP application:response_time_seconds Server response time for /index.html
application:response_time_seconds_count 80
application:response_time_seconds{quantile="0.5"} 0.0002933240
application:response_time_seconds{quantile="0.75"} 0.000344914
application:response_time_seconds{quantile="0.95"} 0.000543647
application:response_time_seconds{quantile="0.98"} 0.002706543
application:response_time_seconds{quantile="0.99"} 0.005608694
application:response_time_seconds{quantile="0.999"} 0.005608694

3.3. Security

It must be possible to secure the endpoints via the usual means. The definition of 'usual means' is in this version of the specification implementation specific.

In case of a secured endpoint, accessing /metrics without valid credentials must return a 401 Unauthorized header.

A server SHOULD implement TLS encryption by default.

It is allowed to ignore security for trusted origins (e.g. localhost)

4. Required Metrics

Base metrics is a list of metrics that all vendors need to implement. Optional base metrics are recommended to be implemented but are not required. These metrics are exposed under /metrics/base.

The following is a list of required and optional base metrics. All metrics are singletons and have Multi: set to false unless otherwise stated. Visit Metadata for the meaning of each key

4.1. General JVM Stats

UsedHeapMemory

Name

memory.usedHeap

DisplayName

Used Heap Memory

Type

Gauge

Unit

Bytes

Description

Displays the amount of used heap memory in bytes.

MBean

java.lang:type=Memory/HeapMemoryUsage#used

CommittedHeapMemory

Name

memory.committedHeap

DisplayName

Committed Heap Memory

Type

Gauge

Unit

Bytes

Description

Displays the amount of memory in bytes that is committed for the Java virtual machine to use. This amount of memory is guaranteed for the Java virtual machine to use.

MBean

java.lang:type=Memory/HeapMemoryUsage#committed

Notes

Also from JSR 77

MaxHeapMemory

Name

memory.maxHeap

DisplayName

Max Heap Memory

Type

Gauge

Unit

Bytes

Description

Displays the maximum amount of heap memory in bytes that can be used for memory management. This attribute displays -1 if the maximum heap memory size is undefined. This amount of memory is not guaranteed to be available for memory management if it is greater than the amount of committed memory. The Java virtual machine may fail to allocate memory even if the amount of used memory does not exceed this maximum size.

MBean

java.lang:type=Memory/HeapMemoryUsage#max

GCCount

Name

gc.%s.count

DisplayName

Garbage Collection Count

Type

Counter

Unit

None

Multi

true

Description

Displays the total number of collections that have occurred. This attribute lists -1 if the collection count is undefined for this collector.

MBean

java.lang:type=GarbageCollector,name=%s/CollectionCount

Notes

There can be multiple garbage collectors active that are assigned to different memory pools. The %s should be substituted with the name of the garbage collector.

GCTime - Approximate accumulated collection elapsed time in ms

Name

gc.%s.time

DisplayName

Garbage Collection Time

Type

Gauge

Unit

Milliseconds

Multi

true

Description

Displays the approximate accumulated collection elapsed time in milliseconds. This attribute displays -1 if the collection elapsed time is undefined for this collector. The Java virtual machine implementation may use a high resolution timer to measure the elapsed time. This attribute may display the same value even if the collection count has been incremented if the collection elapsed time is very short.

MBean

java.lang:type=GarbageCollector,name=%s/CollectionTime

Notes

There can be multiple garbage collectors active that are assigned to different memory pools. The %s should be substituted with the name of the garbage collector.

JVM Uptime - Up time of the Java Virtual machine

Name

jvm.uptime

DisplayName

JVM Uptime

Type

Gauge

Unit

Milliseconds

Description

Displays the start time of the Java virtual machine in milliseconds. This attribute displays the approximate time when the Java virtual machine started.

MBean

java.lang:type=Runtime/Uptime

Notes

Also from JSR 77

4.2. Thread JVM Stats

ThreadCount

Name

thread.count

DisplayName

Thread Count

Type

Counter

Unit

None

Description

Displays the current number of live threads including both daemon and non-daemon threads

MBean

java.lang:type=Threading/ThreadCount

DaemonThreadCount

Name

thread.daemon.count

DisplayName

Daemon Thread Count

Type

Counter

Unit

None

Description

Displays the current number of live daemon threads.

MBean

java.lang:type=Threading/DaemonThreadCount

PeakThreadCount

Name

thread.max.count

DisplayName

Peak Thread Count

Type

Counter

Unit

None

Description

Displays the peak live thread count since the Java virtual machine started or peak was reset. This includes daemon and non-daemon threads.

MBean

java.lang:type=Threading/PeakThreadCount

4.3. Thread Pool Stats

(Optional) ActiveThreads

Name

threadPool.%s.activeThreads

DisplayName

Active Threads

Type

Gauge

Unit

None

Multi

true

Description

Number of active threads that belong to a specific thread pool. Note: The %s should be substituted with the name of the thread pool. This is a vendor specific attribute/operation that is not defined in java.lang.

(Optional) PoolSize

Name

threadPool.%s.size

DisplayName

Thread Pool Size

Type

Gauge

Unit

None

Multi

true

Description

The size of a specific thread pool. Note: The %s should be substituted with the name of the thread pool. This is a vendor specific attribute/operation that is not defined in java.lang.

4.4. ClassLoading JVM Stats

LoadedClassCount

Name

classloader.currentLoadedClass.count

DisplayName

Current Loaded Class Count

Type

Counter

Unit

None

Description

Displays the number of classes that are currently loaded in the Java virtual machine.

MBean

java.lang:type=ClassLoading/LoadedClassCount

TotalLoadedClassLoaded

Name

classloader.totalLoadedClass.count

DisplayName

Total Loaded Class Count

Type

Counter

Unit

None

Description

Displays the total number of classes that have been loaded since the Java virtual machine has started execution.

MBean

java.lang:type=ClassLoading/TotalLoadedClassCount

UnloadedClassCount

Name

classloader.totalUnloadedClass.count

DisplayName

Total Unloaded Class Count

Type

Counter

Unit

None

Description

Displays the total number of classes unloaded since the Java virtual machine has started execution.

MBean

java.lang:type=ClassLoading/UnloadedClassCount

4.5. Operating System

AvailableProcessors

Name

cpu.availableProcessors

DisplayName

Available Processors

Type

Gauge

Unit

None

Description

Displays the number of processors available to the Java virtual machine. This value may change during a particular invocation of the virtual machine.

MBean

java.lang:type=OperatingSystem/AvailableProcessors

(Optional) SystemLoadAverage

Name

cpu.systemLoadAverage

DisplayName

System Load Average

Type

Gauge

Unit

None

Description

Displays the system load average for the last minute. The system load average is the sum of the number of runnable entities queued to the available processors and the number of runnable entities running on the available processors averaged over a period of time. The way in which the load average is calculated is operating system specific but is typically a damped time-dependent average. If the load average is not available, a negative value is displayed. This attribute is designed to provide a hint about the system load and may be queried frequently. The load average may be unavailable on some platforms where it is expensive to implement this method.

MBean

java.lang:type=OperatingSystem/SystemLoadAverage

(Optional) ProcessCpuLoad

Name

cpu.processCpuLoad

DisplayName

Process CPU Load

Type

Gauge

Unit

Percent

Description

Displays the "recent cpu usage" for the Java Virtual Machine process

MBean

java.lang:type=OperatingSystem (com.sun.management.UnixOperatingSystemMXBean for Oracle Java, similar one exists for IBM Java: com.ibm.lang.management.ExtendedOperatingSystem) Note: This is a vendor specific attribute/operation that is not defined in java.lang

5. Application Metrics Programming Model

MicroProfile Metrics provides a way to register Application-specific metrics to allow applications to expose metrics in the application scope (see Scopes for the description of scopes).

Metrics and their metadata are added to a Metric Registry upon definition and can afterwards have their values set and retrieved via the Java-API and also be exposed via the REST-API (see Exposing metrics via REST API).

Implementors of this specification can use the Java API to also expose metrics for base and vendor scope by using the respective Metric Registry.

In order to make setting the values easier, Annotations are made available.

Example set-up of a Gauge metric. No unit is given, so MetricUnits.NONE is used, an explicit name is provided
@Gauge(unit = MetricUnits.NONE, name = "queueSize")
public int getQueueSize() {
  return queue.size;
}
The programming API follows Dropwizard Metrics 3.2.3 API, but with local changes. It is expected that many existing DropWizard Metrics based applications can easily be ported over by exchanging the package names.

It will be possible to use the non-annotations API, but using the annotations will generally be easier for developers.

5.1. Responsibility of the MicroProfile Metrics implementation

  • The implementation must scan the application at deploy time for Annotations and register the Metrics along with their metadata in the application MetricsRegistry.

  • The implementation must watch the annotated objects and update internal data structures when the values of the annotated objects change.

  • The implementation must expose the values of the objects registered in the MetricsRegistry via REST-API as described in Exposing metrics via REST API.

  • Metrics registered via non-annotations API need their values be set via updates from the application code.

  • The implementation must flag duplicate metrics upon registration and reject the duplicate unless the metric is explicitly marked as reusable upon first registration and in all subsequent registrations.

    • A duplicate metric is a metric that has the same scope and name as an existing one.

    • The implementation must throw an IllegalArgumentException when the metric is rejected.

    • It is not allowed to reuse a metric (name) for metrics of different types. The implementation must throw an IllegalArgumentException if such a mismatch is detected.

    • See Reusing of Metrics for more details.

5.2. Base Package

All Java-Classes are in the top-level package org.eclipse.microprofile.metrics or one of its sub-packages.

The org.eclipse.microprofile.metrics package was influenced by the Drop Wizard Metrics project release 3.2.3.

Implementors can consult this project for implementation ideas.

See References for more information.

5.3. Annotations

All Annotations are in the org.eclipse.microprofile.metrics.annotation package

These annotations include interceptor bindings as defined by the Java Interceptors specification.

CDI leverages on the Java Interceptors specification to provide the ability to associate interceptors to beans via typesafe interceptor bindings, as a mean to separate cross-cutting concerns, like Metrics annotations instrumentation, from the application business logic.

Both the Java Interceptors and CDI specifications set restrictions about the type of bean to which an interceptor can be bound.

That implies only managed beans whose bean types are proxyable can be instrumented using the Metrics annotations.

The org.eclipse.microprofile.metrics.annotation package was influenced by the CDI extension for Dropwizard Metric project release 1.4.0.

Implementors can consult this project for implementation ideas.

See References for more information.

The following Annotations exist, see below for common fields:

Annotation Applies to Description Default Unit

@Counted

M, C, T

Denotes a counter, which counts the invocations of the annotated object.

MetricUnits.NONE

@Gauge

M

Denotes a gauge, which samples the value of the annotated object.

none Must be supplied by the user

@Metered

M, C, T

Denotes a meter, which tracks the frequency of invocations of the annotated object.

MetricUnits.PER_SECOND

@Metric

M, F, P

An annotation that contains the metadata information when requesting a metric to be injected or produced. This annotation can be used on fields of type Meter, Timer, Counter, and Histogram. For Gauge, the @Metric annotation can only be used on producer methods/fields.

MetricUnits.NONE

@Timed

M, C, T

Denotes a timer, which tracks duration of the annotated object.

MetricUnits.NANOSECONDS

(C=Constructor, F=Field, M=Method, P=Parameter, T=Type)

Annotation Description Default

@RegistryType

Qualifies the scope of Metric Registry to inject when injecting a MetricRegistry.

application (scope)

5.3.1. Fields

All annotations (Except RegistryType) have the following fields that correspond to the metadata fields described in Metadata.

String name

Optional. Sets the name of the metric. If not explicitly given the name of the annotated object is used.

boolean absolute

If true, uses the given name as the absolute name of the metric. If false, prepends the package name and class name before the given name. Default value is false.

String displayName

Optional. A human readable display name for metadata.

String description

Optional. A description of the metric.

String unit

Unit of the metric. For @Gauge no default is provided. Check the MetricUnits class for a set of pre-defined units.

String[] tags

Optional. Array of Strings in the <key>=<value> format to supply special tags to a metric.

boolean reusable

Denotes if a metric with a certain name can be registered in more than one place. Does not apply to gauges.

Implementors are encouraged to issue warnings in the server log if metadata is missing. Implementors MAY stop the deployment of an application if Metadata is missing.

5.3.2. Annotated Naming Convention

Annotated metrics are registered into the application MetricRegistry with the name based on the annotation’s name and absolute fields.

Example of annotated metric names
package com.example;

import javax.inject.Inject;
import org.eclipse.microprofile.metrics.Counter;
import org.eclipse.microprofile.metrics.annotation.Metric;

public class Colours {

  @Inject
  @Metric
  Counter redCount;

  @Inject
  @Metric(name="blue")
  Counter blueCount;

  @Inject
  @Metric(absolute=true)
  Counter greenCount;

  @Inject
  @Metric(name="purple", absolute=true)
  Counter purpleCount;
}

The above bean would produce the following entries in the MetricRegistry

com.example.Colours.redCount
com.example.Colours.blue
greenCount
purple

5.3.3. @Counted

An annotation for marking a method, constructor, or type as a counter.

The implementation must support the following annotation targets:

  • CONSTRUCTOR

  • METHOD

  • TYPE

In addition to the metadata fields, @Counted has the following field:

monotonic

If false (default), the counter is incremented before the annotated method is invoked and decremented after the annotated method returns, counting current invocations of the annotated method.

If true, the counter increases monotonically, counting total invocations of the annotated method.

The intuitive expectation is that each invocation of a method marked with @Counted would result in the respective counter being increased by one. This is not the case with the default semantics of the monotonic flag.
If you want the counter being increased on each invocation you need to use @Counted(monotonic = true).

The following lists the behavior for each annotation target.

CONSTRUCTOR

When a constructor is annotated, the implementation must register a counter for the constructor using the Annotated Naming Convention. The counter is treated according to the monotonic field when the constructor is invoked.

Example of an annotated constructor
@Counted
public CounterBean() {
}
METHOD

When a method is annotated, the implementation must register a counter for the method using the Annotated Naming Convention. The counter is treated according to the monotonic field when the method is invoked.

Example of an annotated method
@Counted
public void run() {
}
TYPE

When a type/class is annotated, the implementation must register a counter for each of the constructors and methods using the Annotated Naming Convention. The counters are treated according to the monotonic field when the corresponding constructor/method is invoked.

Example of an annotated type/class
@Counted
public class CounterBean {

  public void countMethod1() {}
  public void countMethod2() {}

}

5.3.4. @Gauge

An annotation for marking a method as a gauge. No default MetricUnit is supplied, so the unit must always be specified explicitly.

The implementation must support the following annotation target:

  • METHOD

The following lists the behavior for each annotation target.

METHOD

When a method is annotated, the implementation must register a gauge for the method using the Annotated Naming Convention. The gauge value and type is equal to the annotated method return value and type.

Example of an annotated method
@Gauge(unit = MetricUnits.NONE)
public long getValue() {
  return value;
}

5.3.5. @Metered

An annotation for marking a constructor or method as metered. The meter counts the invocations of the constructor or method and tracks how frequently they are called.

The implementation must support the following annotation targets:

  • CONSTRUCTOR

  • METHOD

  • TYPE

The following lists the behavior for each annotation target.

CONSTRUCTOR

When a constructor is annotated, the implementation must register a meter for the constructor using the Annotated Naming Convention. The meter is marked each time the constructor is invoked.

Example of an annotated constructor
@Metered
public MeteredBean() {
}
METHOD

When a method is annotated, the implementation must register a meter for the method using the Annotated Naming Convention. The meter is marked each time the method is invoked.

Example of an annotated method
@Metered
public void run() {
}
TYPE

When a type/class is annotated, the implementation must register a meter for each of the constructors and methods using the Annotated Naming Convention. The meters are marked each time the corresponding constructor/method is invoked.

Example of an annotated type/class
@Metered
public class MeteredBean {

  public void meteredMethod1() {}
  public void meteredMethod2() {}

}

5.3.6. @Timed

An annotation for marking a constructor or method of an annotated object as timed. The metric of type Timer tracks how frequently the annotated object is invoked, and tracks how long it took the invocations to complete.

The implementation must support the following annotation targets:

  • CONSTRUCTOR

  • METHOD

  • TYPE

The following lists the behavior for each annotation target.

CONSTRUCTOR

When a constructor is annotated, the implementation must register a timer for the constructor using the Annotated Naming Convention. Each time the constructor is invoked, the execution will be timed.

Example of an annotated constructor
@Timed
public TimedBean() {
}
METHOD

When a method is annotated, the implementation must register a timer for the method using the Annotated Naming Convention. Each time the method is invoked, the execution will be timed.

Example of an annotated method
@Timed
public void run() {
}
TYPE

When a type/class is annotated, the implementation must register a timer for each of the constructors and methods using the Annotated Naming Convention. Each time a constructor/method is invoked, the execution will be timed with the corresponding timer.

Example of an annotated type/class
@Timed
public class TimedBean {

  public void timedMethod1() {}
  public void timedMethod2() {}

}

5.3.7. @Metric

An annotation requesting that a metric should be injected or registered.

The implementation must support the following annotation targets:

  • FIELD

  • METHOD

  • PARAMETER

The following lists the behavior for each annotation target.

FIELD

When a metric producer field is annotated, the implementation must register the metric to the application MetricRegistry (using the Annotated Naming Convention). If a metric with the given name already exist (created by another @Produces for example), an java.lang.IllegalArgumentException must be thrown.

Example of a producer field
@Produces
@Metric(name="hitPercentage")
@ApplicationScoped
Gauge<Double> hitPercentage = new Gauge<Double>() {

  @Override
  public Double getValue() {
      return hits / total;
  }
};

When a metric injected field is annotated, the implementation must provide the registered metric with the given name (using the Annotated Naming Convention) if the metric already exist. If no metric exists with the given name then the implementation must produce and register the requested metric. @Metric can only be used on injected fields of type Meter, Timer, Counter, and Histogram.

Example of an injected field
@Inject
@Metric("applicationCount")
Counter count;
METHOD

When a metric producer method is annotated, the implementation must register the metric produced by the method using the Annotated Naming Convention.

Example of a producer method
@Produces
@Metric(name = "hitPercentage")
@ApplicationScoped
protected Gauge<Double> createHitPercentage() {
    return new Gauge<Double>() {

      @Override
      public Double getValue() {
          return hits / total;
      }
    };
}
PARAMETER

When a metric parameter is annotated, the implementation must provide the registered metric with the given name (using the Annotated Naming Convention) if the metric already exist. If no metric exists with the given name then the implementation must produce and register the requested metric.

Example of an annotated parameter
@Inject
public void init(@Metric(name="instances") Counter instances) {
    instances.inc();
}

5.4. Metric Registries

The MetricRegistry is used to maintain a collection of metrics along with their metadata. There is one shared singleton of the MetricRegistry per scope (application, base, and vendor). When metrics are registered using annotations, the metrics are registered in the application MetricRegistry (and thus the application scope).

When injected, the @RegistryType is used as a qualifier to selectively inject either the APPLICATION, BASE, or VENDOR registry. If no qualifier is used, the default MetricRegistry returned is the APPLICATION registry.

Implementations may choose to use a Factory class to produce the injectable MetricRegistry bean via CDI. See Example Metric Registry Factory. Note: The factory would be an internal class and not exposed to the application.

5.4.1. @RegistryType

The @RegistryType can be used to retrieve the MetricRegistry for a specific scope. The implementation must produce the corresponding MetricRegistry specified by the RegistryType.

The implementor can optionally provide a read_only copy of the MetricRegistry for base and vendor scopes.

5.4.2. Application Metric Registry

The implementation must produce the application MetricRegistry when no RegistryType is provided (@Default) or when the RegistryType is APPLICATION.

Example of the application injecting the application registry
@Inject
MetricRegistry metricRegistry;
is equivalent to
@Inject
@RegistryType(type=MetricRegistry.Type.APPLICATION)
MetricRegistry metricRegistry;

5.4.3. Base Metric Registry

The implementation must produce the base MetricRegistry when the RegistryType is BASE. The base MetricRegistry must contain the required metrics specified in Required Metrics.

Example of the application injecting the base registry
@Inject
@RegistryType(type=MetricRegistry.Type.BASE)
MetricRegistry baseRegistry;

5.4.4. Vendor Metric Registry

The implementation must produce the vendor MetricRegistry when the RegistryType is VENDOR. The vendor MetricRegistry must contain any vendor specific metrics.

Example of the application injecting the vendor registry
@Inject
@RegistryType(type=MetricRegistry.Type.VENDOR)
MetricRegistry vendorRegistry;

6. Appendix

6.1. Alternatives considered

Jolokia JMX-HTTP bridge. Using this for application specific metrics would require those metrics to be exposed to JMX first, which many users are not familiar with.

6.3. Example configuration format for base and vendor-specific data

The following is an example configuration in YAML format.

base:
  - name: "thread-count"
    mbean: "java.lang:type=Threading/ThreadCount"
    description: "Number of currently deployed threads"
    unit: "none"
    type: "gauge"
    displayName: "Current Thread count"
  - name: "peak-thread-count"
    mbean: "java.lang:type=Threading/PeakThreadCount"
    description: "Max number of threads"
    unit: "none"
    type: "gauge"
  - name: "total-started-thread-count"
    mbean: "java.lang:type=Threading/TotalStartedThreadCount"
    description: "Number of threads started for this server"
    unit: "none"
    type: "counter"
  - name: "max-heap"
    mbean: "java.lang:type=Memory/HeapMemoryUsage#max"
    description: "Number of threads started for this server"
    unit: "bytes"
    type: "counter"
    tags: "kind=memory"

vendor:
  - name: "msc-loaded-modules"
    mbean: "jboss.modules:type=ModuleLoader,name=BootModuleLoader-2/LoadedModuleCount"
    description: "Number of loaded modules"
    unit: "none"
    type: "gauge"

This configuration can be backed into the runtime or be provided via an external configuration file.

6.4. Example Metric Registry Factory

Sample skeleton factory class to produce MetricRegistry via CDI
@ApplicationScoped
public class MetricRegistryFactory {

    @Produces
    public static MetricRegistry getDefaultRegistry() {
        return getApplicationRegistry();
    }

    @Produces
    @RegistryType(type = Type.APPLICATION)
    public static MetricRegistry getApplicationRegistry() {
        // Returns the static instance of the Application MetricRegistry
        [...]
    }

    @Produces
    @RegistryType(type = Type.BASE)
    public static MetricRegistry getBaseRegistry() {
        // Returns the static instance of the Base MetricRegistry
        [...]
    }

    @Produces
    @RegistryType(type = Type.VENDOR)
    public static MetricRegistry getVendorRegistry() {
        // Returns the static instance of the Vendor MetricRegistry
        [...]
    }

}