Prometheus

5 minute read

The prometheus adapter collects Istio metrics and makes them available to Prometheus.

This adapter supports the metric template.

Params

Configuration format for the prometheus adapter.

Field Type Description

Field	Type	Description
`metrics`	`Params.MetricInfo[]`	The set of metrics to represent in Prometheus. If a metric is defined in Istio but doesn’t have a corresponding shape here, it will not be populated at runtime.
`metricsExpirationPolicy`	`Params.MetricsExpirationPolicy`	Optional. The rate at which to expire metrics from the adapter. This option controls the amount of metric data that the adapter will maintain over its lifetime. EXPERIMENTAL: This feature should only be used in advanced cases. Example config stanza: `metricsExpirationPolicy: metricsExpiryDuration: "5s" expiryCheckIntervalDuration: "1s"`

metrics

Params.MetricInfo[]

The set of metrics to represent in Prometheus. If a metric is defined in Istio but doesn’t have a corresponding shape here, it will not be populated at runtime.

metricsExpirationPolicy

Params.MetricsExpirationPolicy

Optional. The rate at which to expire metrics from the adapter. This option controls the amount of metric data that the adapter will maintain over its lifetime.

EXPERIMENTAL: This feature should only be used in advanced cases.

Example config stanza:

metricsExpirationPolicy:
  metricsExpiryDuration: "5s"
  expiryCheckIntervalDuration: "1s"

Params.MetricInfo

Describes how a metric should be represented in Prometheus.

Field	Type	Description
`namespace`	`string`	Optional. The namespace is used as a prefix on the metric names. An example: for a metric named `requests_total` with a namespace of `istio`, the full metric name becomes `istio_requests_total`. NOTE: The specified namespace should be a prometheus-safe name. If not, the adapter will attempt to convert the namespace to a prometheus-safe name. NOTE: Changing the value for this will potentially impact downstream integrations and should be used with caution. Default value: `istio`.
`name`	`string`	Recommended. The name is used to register the prometheus metric. It must be unique across all prometheus metrics as prometheus does not allow duplicate names. If name is not specified a sanitized version of instance_name is used.
`instanceName`	`string`	Required. The name is the fully qualified name of the Istio metric instance that this MetricInfo processes.
`description`	`string`	Optional. A human readable description of this metric.
`kind`	`Params.MetricInfo.Kind`
`buckets`	`Params.MetricInfo.BucketsDefinition`	For metrics with a metric kind of DISTRIBUTION, this provides a mechanism for configuring the buckets that will be used to store the aggregated values. This field must be provided for metrics declared to be of type DISTRIBUTION. This field will be ignored for non-distribution metric kinds.
`labelNames`	`string[]`	The names of labels to use: these need to match the dimensions of the Istio metric. TODO: see if we can remove this and rely on only the dimensions in the future.

Params.MetricInfo.BucketsDefinition

Describes buckets for DISTRIBUTION kind metrics.

Field	Type	Description
`linearBuckets`	`Params.MetricInfo.BucketsDefinition.Linear (oneof)`	The linear buckets.
`exponentialBuckets`	`Params.MetricInfo.BucketsDefinition.Exponential (oneof)`	The exponential buckets.
`explicitBuckets`	`Params.MetricInfo.BucketsDefinition.Explicit (oneof)`	The explicit buckets.

Params.MetricInfo.BucketsDefinition.Explicit

Specifies a set of buckets with arbitrary widths.

There are size(bounds) + 1 (= N) buckets. Bucket i has the following boundaries:

Upper bound (0 <= i < N-1): bounds[i]
Lower bound (1 <= i < N): bounds[i - 1]

The bounds field must contain at least one element. If bounds has only one element, then there are no finite buckets, and that single element is the common boundary of the overflow and underflow buckets.

Field	Type	Description
`bounds`	`double[]`	The values must be monotonically increasing.

Params.MetricInfo.BucketsDefinition.Exponential

Specifies an exponential sequence of buckets that have a width that is proportional to the value of the lower bound. Each bucket represents a constant relative uncertainty on a specific value in the bucket.

There are num_finite_buckets + 2 (= N) buckets. The two additional buckets are the underflow and overflow buckets.

Bucket i has the following boundaries:

Upper bound (0 <= i < N-1): scale * (growth_factor ^ i)
Lower bound (1 <= i < N): scale * (growth_factor ^ (i - 1))

Field	Type	Description
`numFiniteBuckets`	`int32`	Must be greater than 0.
`growthFactor`	`double`	Must be greater than 1.
`scale`	`double`	Must be greater than 0.

Params.MetricInfo.BucketsDefinition.Linear

Specifies a linear sequence of buckets that all have the same width (except overflow and underflow). Each bucket represents a constant absolute uncertainty on the specific value in the bucket.

There are num_finite_buckets + 2 (= N) buckets. The two additional buckets are the underflow and overflow buckets.

Bucket i has the following boundaries:

Upper bound (0 <= i < N-1): offset + (width * i)
Lower bound (1 <= i < N): offset + (width * (i - 1))

Field	Type	Description
`numFiniteBuckets`	`int32`	Must be greater than 0.
`width`	`double`	Must be greater than 0.
`offset`	`double`	Lower bound of the first bucket.

Params.MetricInfo.Kind

Describes what kind of metric this is.

Name	Description
`UNSPECIFIED`
`GAUGE`
`COUNTER`
`DISTRIBUTION`

Params.MetricsExpirationPolicy

Describes the expiration policy for metrics generated by a prometheus handler.

Example: A Metrics Expiration Policy of { metrics_expiry_duration: "10m", expiry_check_interval_duration: "1m" } would configure the handler to delete all metrics that have received no updtaes for 10 minutes. Metrics would be checked every minute to determine whether or not they should be expired.

Field	Type	Description
`metricsExpiryDuration`	`google.protobuf.Duration`	Required. Describes the desired lifetime of a metric. If the metric is not updated at any point during this duration, it will be removed from the set of metrics exported by the handler.
`expiryCheckIntervalDuration`	`google.protobuf.Duration`	Optional. Describes the interval in which metrics will be checked to see if they have been stale for longer that the configured `metrics_expiry_duration`. This should be generally be set to a duration much shorter than the configured `metrics_expiry_duration`. This bounds the total amount of additional time (beyond the desired lifetime) that a metric may be exported. If a value is not explicitly provided, this value will default to half of the configured `metrics_expiry_duration`.