The Confluent Cloud Metrics API provides actionable operational metrics about your Confluent Cloud deployment. This is a queryable HTTP API in which the user will POST a query written in JSON and get back a time series of metrics specified by the query.

Comprehensive documentation is available on docs.confluent.io.

Please see the Metrics Reference for a list of available metrics.

This information is also available programmatically via the descriptors endpoint.

Confluent uses API keys for integrating with Confluent Cloud. Applications must be authorized and authenticated before they can access or manage resources in Confluent Cloud. You can manage your API keys in the Confluent Cloud Dashboard or Confluent Cloud CLI.

An API key is owned by a User or Service Account and inherits the permissions granted to the owner.

Today, you can divide API keys into two classes:

• Cloud API Keys - These grant access to the Confluent Cloud Control Plane APIs, such as for Provisioning and Metrics integrations.
• Cluster API Keys - These grant access to a single Confluent cluster, such as a specific Kafka or Schema Registry cluster.

Cloud API Keys are required for the Metrics API. Cloud API Keys can be created using the Confluent Cloud CLI.

ccloud api-key create --resource cloud

All API requests must be made over HTTPS. Calls made over plain HTTP will fail. API requests without authentication will also fail.

The Confluent Cloud Metrics API also supports OAuth 2.0 by allowing Confluent Security Token Service (STS) tokens as credentials to authenticate to metrics API. See steps to [Authenticate access to Confluent Cloud APIs using Confluent STS tokens] (https://docs.confluent.io/cloud/current/access-management/authenticate/oauth/access-rest-apis-sts.html#authenticate- access-to-ccloud-apis-using-confluent-security-token-service-sts-tokens)

Confluent APIs ensure stability for your integrations by avoiding the introduction of breaking changes to customers unexpectedly. Confluent will make non-breaking API changes without advance notice. Thus, API clients must follow the Compatibility Policy below to ensure your ingtegration remains stable. All APIs follow the API Lifecycle Policy described below, which describes the guarantees API clients can rely on.

Breaking changes will be widely communicated in advance in accordance with our Deprecation Policy. Confluent will provide timelines and a migration path for all API changes, where available. Be sure to subscribe to one or more communication channels so you don't miss any updates!

One exception to these guidelines is for critical security issues. We will take any necessary actions to mitigate any critical security issue as soon as possible, which may include disabling the vulnerable functionality until a proper solution is available.

Do not consume any Confluent API unless it is documented in the API Reference. All undocumented endpoints should be considered private, subject to change without notice, and not covered by any agreements.

Note: The "v1" in the URL is not a "major version" in the Semantic Versioning sense. It is a "generational version" or "meta version", as seen in other APIs like Github API or the Stripe API.

2022-12-01

The dataset health-plus is now available in preview.

See the Datasets section for more details.

2022-10-18

cluster_active_link_count metric is deprecated

The io.confluent.kafka.server/cluster_active_link_count is now deprecated. Please use the io.confluent.kafka.server/cluster_link_count metric instead.

New metrics available in /export

The following metrics are now available in the /export endpoint:

• io.confluent.kafka.server/request_bytes
• io.confluent.kafka.server/response_bytes
• io.confluent.kafka.server/cluster_link_destination_response_bytes
• io.confluent.kafka.server/cluster_link_source_response_bytes
• io.confluent.kafka.server/cluster_link_count
• io.confluent.kafka.server/cluster_link_mirror_topic_count
• io.confluent.kafka.server/cluster_link_mirror_topic_offset_lag
• io.confluent.kafka.server/cluster_link_mirror_topic_bytes

2022-10-17

API Version 1 is marked sunset

All API Version 1 endpoints are no longer supported from 2022-10-17. API users should migrate to API Version 2.

2021-09-23

API Version 1 is now deprecated

All API Version 1 endpoints are now deprecated and will be removed on 2022-04-04. API users should migrate to API Version 2.

2021-08-24

Metric-specific aggregation functions

New metrics are being introduced that require alternative aggregation functions (e.g. MAX). When querying those metrics, using agg: "SUM" will return an error. It is recommended that clients omit the agg field in the request such that the required aggregation function for the specific metric is automatically applied on the backend.

Note: The initial version of Metrics API required clients to effectively hardcode agg: "SUM" in all queries. In early 2021, the agg field was made optional, but many clients have not been updated to omit the agg field.

Cursor-based pagination for /query endpoint

The /query endpoint now supports cursor-based pagination similar to the /descriptors and /attributes endpoints.

2021-02-10

API Version 2 is now Generally Available (GA)

See the Version 2 section below for a detailed description of changes and migration guide.

2020-12-04

API Version 2 (Preview)

Version 2 of the Metrics API is now available in Preview. See the Version 2 section below for a detailed description of changes.

2020-07-08

Correction for active_connection_count metric

A bug in the active_connection_count metric that affected a subset of customers was fixed. Customers exporting the metric to an external monitoring system may observe a discontinuity between historical results and current results due to this one-time correction.

2020-04-01

This release includes the following changes from the preview release:

New format request attribute

The /query request now includes a format attribute which controls the result structuring in the response body. See the /query endpoint definition for more details.

New /available endpoint

The new /available endpoint allows determining which metrics are available for a set of resources (defined by labels). This endpoint can be used to determine which subset of metrics are currently available for a specific resource (e.g. a Confluent Cloud Kafka cluster).

Metric type changes

The CUMULATIVE_(INT|DOUBLE) metric type enumeration was changed to COUNTER_(INT|DOUBLE). This was done to better align with OpenTelemetry conventions. In tandem with this change, several metrics that were improperly classified as GAUGEs were re-classified as COUNTERs.

Metric name changes

The /delta suffix has been removed from the following metrics:

• io.confluent.kafka.server/received_bytes/delta
• io.confluent.kafka.server/sent_bytes/delta
• io.confluent.kafka.server/request_count/delta

2020-09-15

Retire /available endpoint

The /available endpoint (which was in Preview status) has been removed from the API. The /descriptors endpoint can still be used to determine the universe of available metrics for Metrics API.

The legacy metric names are deprecated and will stop functioning on 2020-07-01.

The following status labels are applicable to APIs, features, and SDK versions, based on the current support status of each:

• Early Access – May change at any time. Not recommended for production usage. Not officially supported by Confluent. Intended for user feedback only. Users must be granted explicit access to the API by Confluent.
• Preview – Unlikely to change between Preview and General Availability. Not recommended for production usage. Officially supported by Confluent for non-production usage. For Closed Previews, users must be granted explicit access to the API by Confluent.
• Generally Available (GA) – Will not change at short notice. Recommended for production usage. Officially supported by Confluent for non-production and production usage.
• Deprecated – No longer supported. Will be removed in the future at the announced date. Use is discouraged and migration following the upgrade guide is recommended.
• Sunset – Removed, and no longer supported or available.

Resources, operations, and individual fields in the OpenAPI spec will be annotated with x-lifecycle-stage, x-deprecated-at, and x-sunset-at. These annotations will appear in the corresponding API Reference Documentation. An API is "Generally Available" unless explicitly marked otherwise.

Confluent APIs are governed by Confluent Cloud Upgrade Policy in which we will make backward incompatible changes and deprecations approximately once per year, and will provide 180 days notice via email to all registered Confluent Cloud users.

Backward Compatibility

An API version is backwards-compatible if a program written against the previous version of the API will continue to work the same way, without modification, against this version of the API.

Confluent considers the following changes to be backwards-compatible:

• Adding new API resources.
• Adding new optional parameters to existing API requests (e.g., query string or body).
• Adding new properties to existing API responses.
• Changing the order of properties in existing API responses.
• Changing the length or format of object IDs or other opaque strings.
  • Unless otherwise documented, you can safely assume object IDs we generate will never exceed 255 characters, but you should be able to handle IDs of up to that length. If you're using MySQL, for example, you should store IDs in a VARCHAR(255) COLLATE utf8_bin column.
  • This includes adding or removing fixed prefixes (such as lkc- on kafka cluster IDs).
  • This includes API keys, API tokens, and similar authentication mechanisms.
  • This includes all strings described as "opaque" in the docs, such as pagination cursors.
• Omitting properties with null values from existing API responses.

Client Responsibilities

• Resource and rate limits, and the default and maximum sizes of paginated data are not considered part of the API contract and may change (possibly dynamically). It is the client's responsibility to read the road signs and obey the speed limit.
• If a property has a primitive type and the API documentation does not explicitly limit its possible values, clients must not assume the values are constrained to a particular set of possible responses.
• If a property of an object is not explicitly declared as mandatory in the API, clients must not assume it will be present.
• A resource may be modified to return a "redirection" response (e.g. 301, 307) instead of directly returning the resource. Clients must handle HTTP-level redirects, and respect HTTP headers (e.g. Location).

Confluent will announce deprecations at least 180 days in advance of a breaking change and we will continue to maintain the deprecated APIs in their original form during this time.

Exceptions to this policy apply in case of critical security vulnerabilities or functional defects.

Communication

When a deprecation is announced, the details and any relevant migration information will be available on the following channels:

• Publication in the API Changelog
• Lifecycle, deprecation and "x-deprecated-at" annotations in the OpenAPI spec
• Announcements on the Developer Blog, Community Slack (join!), Google Group, the @ConfluentInc twitter account, and similar channels
• Enterprise customers may receive information by email to their specified Confluent contact, if applicable.

The object model for the Metrics API is designed similarly to the OpenTelemetry standard.

A metric is a numeric attribute of a resource, measured at a specific point in time, labeled with contextual metadata gathered at the point of instrumentation.

There are two types of metrics:

• GAUGE: An instantaneous measurement of a value. Gauge metrics are implicitly averaged when aggregating over time.

  Example: io.confluent.kafka.server/retained_bytes

• COUNTER: The count of occurrences in a single (one minute) sampling interval (unless otherwise stated in the metric description). Counter metrics are implicitly summed when aggregating over time.

  Example: io.confluent.kafka.server/received_bytes

The list of metrics and their labels is available at /docs/descriptors.

A resource represents an entity against which metrics are collected. For example, a Kafka cluster, a Kafka Connector, a ksqlDB application, etc.

Each metric descriptor is associated with one or more resource descriptors, representing the resource types to which that metric can apply. A metric data point is associated with a single resource instance, identified by the resource labels on that metric data point.

For example, metrics emitted by Kafka Connect are associated to the connector resource type. Data points for those metrics include resource labels identifying the specific connector instance that emitted the metric.

The list of resource types and labels are discoverable via the /descriptors/resources endpoint.

A label is a key-value attribute associated with a metric data point.

Labels can be used in queries to filter or group the results. Labels must be prefixed when used in queries:

• metric.<label> (for metric labels), for example metric.topic
• resource.<resource-type>.<label> (for resource labels), for example resource.kafka.id.

The set of valid label keys for a metric include:

• The label keys defined on that metric's descriptor itself
• The label keys defined on the resource descriptor for the metric's associated resource type

For example, the io.confluent.kafka.server/received_bytes metric has the following labels:

• resource.kafka.id - The Kafka cluster to which the metric pertains
• metric.topic - The Kafka topic to which the bytes were produced
• metric.partition - The partition to which the bytes were produced

A dataset is a logical collection of metrics that can be queried together. The dataset is a required URL template parameter for every endpoint in this API. The following datasets are currently available:

To protect the stability of the API and keep it available to all users, Confluent employs multiple safeguards. Users who send many requests in quick succession or perform too many concurrent operations may be throttled or have their requested rejected with an error. When a rate limit is breached, an HTTP 429 Too Many Requests error is returned. The following headers are sent back to provide assistance in dealing with rate limits.

Rate limits are enforced at multiple scopes. You get two sets of the headers above, each specifying the limit of one scope.

Global Rate Limits

A global rate limit of 60 requests per IP address, per minute is enforced.

Per-endpoint Rate Limits

Additionally, some endpoint-specific rate limits are enforced.

Implement retry logic in your client to gracefully handle transient API failures. This should be done by watching for error responses and building in a retry mechanism. This mechanism should follow a capped exponential backoff policy to prevent retry amplification ("retry storms") and also introduce some randomness ("jitter") to avoid the thundering herd effect.

Metric data points are typically available for query in the API within 5 minutes of their origination at the source. This latency can vary based on network conditions and processing overhead. Clients that are polling (or "scraping") metrics into an external monitoring system should account for this latency in their polling requests. API requests that fail to incorporate the latency into the query interval may have incomplete data in the response.

Cursors, tokens, and corresponding pagination links may expire after a short amount of time. In this case, the API will return a 400 Bad Request error and the client will need to restart from the beginning.

The client should have no trouble pausing between rate limiting windows, but persisting cursors for hours or days is not recommended.

Version 2 of the Metrics API adds the ability to query metrics for Kafka Connect, ksqlDB, and Schema Registry.

This capability is enabled by the introduction of a Resource abstraction into the API object model. Resources represent the entity against which metrics are collected.

Migration Guide

The following endpoint URLs have changed in version 2:

The label prefix syntax has changed in version 2:

This example shows a request to /v1/metrics/cloud/query migrated into the new v2 syntax

Version 1 Request

{
  "group_by": [
    "metric.label.topic"
  ],
  "aggregations": [{
    "metric": "io.confluent.kafka.server/received_bytes",
    "agg": "SUM"
  }],
  "filter": {
    "field": "metric.label.cluster_id",
    "op": "EQ",
    "value": "lkc-00000"
  },
  "granularity": "ALL",
  "intervals" : [
    "2020-01-01T00:00:00Z/PT1H"
  ]
}

Version 2 Request

{
  "group_by": [
    "metric.topic"
  ],
  "aggregations": [{
    "metric": "io.confluent.kafka.server/received_bytes",
    "agg": "SUM"
  }],
  "filter": {
    "field": "resource.kafka.id",
    "op": "EQ",
    "value": "lkc-00000"
  },
  "granularity": "ALL",
  "intervals" : [
    "2020-01-01T00:00:00Z/PT1H"
  ]
}