Documentation

Overview

Package metric provides server metrics (a.k.a. transient stats) for a CockroachDB server. These metrics are persisted to the time-series database and are viewable through the web interface and the /_status/metrics/<NODEID> HTTP endpoint.

Adding a new metric

First, add the metric to a Registry.

Next, call methods such as Counter() and Rate() on the Registry to register the metric. For example:

exec := &Executor{
	...
	selectCount: sqlRegistry.Counter("sql.select.count"),
	...
}

This code block registers the metric "sql.select.count" in sql.Registry. The metric can then be accessed through the "selectCount" variable, which can be updated as follows:

func (e *Executor) doSelect() {
	// do the SELECT
	e.selectCount.Inc(1)
}

To add the metric to the web UI, modify the appropriate file in "ui/ts/pages/*.ts". Someone more qualified than me can elaborate, like @maxlang.

Sub-registries

It's common for a Registry to become part of another Registry through the "Add" and "MustAdd" methods.

func NewNodeStatusMonitor(serverRegistry *registry) {
	nodeRegistry := metric.NewRegistry()

	// Add the registry for this node to the root-level server Registry. When
	// accessed from through the serverRegistry, all metrics from the
	// nodeRegistry will have the prefix "cr.node.".
	serverRegistry.MustAdd("cr.node.%s", nodeRegistry)
}

Node-level sub-registries are added by calling:

(*metric.MetricRecorder).AddNodeRegistry(YOUR_NODE_SUBREGISTRY)

Testing

After your test does something to trigger your new metric update, you'll probably want to call methods in TestServer such as MustGetSQLCounter() to verify that the metric was updated correctly. See "sql/metric_test.go" for an example.

Additionally, you can manually verify that your metric is updating by using the metrics endpoint. For example, if you're running the Cockroach DB server with the "--insecure" flag, you can use access the endpoint as follows:

$ curl http://localhost:8080/_status/nodes/1

(some other output)
"cr.node.sql.select.count.1": 5,
(some other output)

Note that a prefix and suffix have been added. The prefix "cr.node." denotes that this metric is node-level. The suffix ".1" specifies that this metric is for node 1.

Index

Constants

const (
	// MaxLatency is the maximum value tracked in latency histograms. Higher
	// values will be recorded as this value instead.
	MaxLatency = 10 * time.Second

	// TestSampleInterval is passed to histograms during tests which don't
	// want to concern themselves with supplying a "correct" interval.
	TestSampleInterval = time.Duration(math.MaxInt64)
)

Variables

var (
	// Scale1M is a 1 minute window for windowed stats (e.g. Rates and Histograms).
	Scale1M = TimeScale{"1m", 1 * time.Minute}

	// Scale10M is a 10 minute window for windowed stats (e.g. Rates and Histograms).
	Scale10M = TimeScale{"10m", 10 * time.Minute}

	// Scale1H is a 1 hour window for windowed stats (e.g. Rates and Histograms).
	Scale1H = TimeScale{"1h", time.Hour}
)

var DefaultTimeScales = []TimeScale{Scale1M, Scale10M, Scale1H}

DefaultTimeScales are the durations used for helpers which create windowed metrics in bulk (such as Rates).


Functions

func TestingSetNow

func TestingSetNow(f func() time.Time) func()

TestingSetNow changes the clock used by the metric system. For use by testing to precisely control the clock.

Types

type Counter

type Counter struct {
	Metadata
	metrics.Counter
}

A Counter holds a single mutable atomic value.

func NewCounter

func NewCounter(metadata Metadata) *Counter

NewCounter creates a counter.

func (*Counter) GetType

func (c *Counter) GetType() *prometheusgo.MetricType

GetType returns the prometheus type enum for this metric.

func (*Counter) Inspect

func (c *Counter) Inspect(f func(interface{}))

Inspect calls the given closure with the empty string and itself.

func (*Counter) MarshalJSON

func (c *Counter) MarshalJSON() ([]byte, error)

MarshalJSON marshals to JSON.

func (*Counter) ToPrometheusMetric

func (c *Counter) ToPrometheusMetric() *prometheusgo.Metric

ToPrometheusMetric returns a filled-in prometheus metric of the right type.

type CounterWithRates

type CounterWithRates struct {
	*Counter
	Rates map[TimeScale]*Rate
}

CounterWithRates is a counter and associated EWMA backed rates at different time scales (which are ignored when visiting the object, and are thus not exported).

func NewCounterWithRates

func NewCounterWithRates(metadata Metadata) *CounterWithRates

NewCounterWithRates registers and returns a new Counter along with some a set of EWMA-based rates at time scales specified by DefaultTimeScales. These rates are provided for convenience and are not exported to metrics frameworks. If they are not required, a Counter should be used instead.

func (*CounterWithRates) Inc

func (c *CounterWithRates) Inc(v int64)

Inc increments the counter.

type Gauge

type Gauge struct {
	Metadata
	// contains filtered or unexported fields
}

A Gauge atomically stores a single integer value.

func NewFunctionalGauge

func NewFunctionalGauge(metadata Metadata, f func() int64) *Gauge

NewFunctionalGauge creates a Gauge metric whose value is determined when asked for by calling the provided function. Note that Update, Inc, and Dec should NOT be called on a Gauge returned from NewFunctionalGauge.

func NewGauge

func NewGauge(metadata Metadata) *Gauge

NewGauge creates a Gauge.

func (*Gauge) Dec

func (g *Gauge) Dec(i int64)

Dec decrements the gauge's value.

func (*Gauge) GetType

func (g *Gauge) GetType() *prometheusgo.MetricType

GetType returns the prometheus type enum for this metric.

func (*Gauge) Inc

func (g *Gauge) Inc(i int64)

Inc increments the gauge's value.

func (*Gauge) Inspect

func (g *Gauge) Inspect(f func(interface{}))

Inspect calls the given closure with the empty string and itself.

func (*Gauge) MarshalJSON

func (g *Gauge) MarshalJSON() ([]byte, error)

MarshalJSON marshals to JSON.

func (*Gauge) Snapshot

func (g *Gauge) Snapshot() metrics.Gauge

Snapshot returns a read-only copy of the gauge.

func (*Gauge) ToPrometheusMetric

func (g *Gauge) ToPrometheusMetric() *prometheusgo.Metric

ToPrometheusMetric returns a filled-in prometheus metric of the right type.

func (*Gauge) Update

func (g *Gauge) Update(v int64)

Update updates the gauge's value.

func (*Gauge) Value

func (g *Gauge) Value() int64

Value returns the gauge's current value.

type GaugeFloat64

type GaugeFloat64 struct {
	Metadata
	metrics.GaugeFloat64
}

A GaugeFloat64 atomically stores a single float64 value.

func NewGaugeFloat64

func NewGaugeFloat64(metadata Metadata) *GaugeFloat64

NewGaugeFloat64 creates a GaugeFloat64.

func (*GaugeFloat64) GetType

func (g *GaugeFloat64) GetType() *prometheusgo.MetricType

GetType returns the prometheus type enum for this metric.

func (*GaugeFloat64) Inspect

func (g *GaugeFloat64) Inspect(f func(interface{}))

Inspect calls the given closure with the empty string and itself.

func (*GaugeFloat64) MarshalJSON

func (g *GaugeFloat64) MarshalJSON() ([]byte, error)

MarshalJSON marshals to JSON.

func (*GaugeFloat64) ToPrometheusMetric

func (g *GaugeFloat64) ToPrometheusMetric() *prometheusgo.Metric

ToPrometheusMetric returns a filled-in prometheus metric of the right type.

type Histogram

type Histogram struct {
	Metadata
	// contains filtered or unexported fields
}

A Histogram collects observed values by keeping bucketed counts. For convenience, internally two sets of buckets are kept: A cumulative set (i.e. data is never evicted) and a windowed set (which keeps only recently collected samples).

Top-level methods generally apply to the cumulative buckets; the windowed variant is exposed through the Windowed method.

func NewHistogram

func NewHistogram(metadata Metadata, duration time.Duration, maxVal int64, sigFigs int) *Histogram

NewHistogram initializes a given Histogram. The contained windowed histogram rotates every 'duration'; both the windowed and the cumulative histogram track nonnegative values up to 'maxVal' with 'sigFigs' decimal points of precision.

func NewLatency

func NewLatency(metadata Metadata, histogramWindow time.Duration) *Histogram

NewLatency is a convenience function which returns a histogram with suitable defaults for latency tracking. Values are expressed in ns, are truncated into the interval [0, MaxLatency] and are recorded with one digit of precision (i.e. errors of <10ms at 100ms, <6s at 60s).

The windowed portion of the Histogram retains values for approximately histogramWindow.

func (*Histogram) GetType

func (h *Histogram) GetType() *prometheusgo.MetricType

GetType returns the prometheus type enum for this metric.

func (*Histogram) Inspect

func (h *Histogram) Inspect(f func(interface{}))

Inspect calls the closure with the empty string and the receiver.

func (*Histogram) Min

func (h *Histogram) Min() int64

Min returns the minimum.

func (*Histogram) RecordValue

func (h *Histogram) RecordValue(v int64)

RecordValue adds the given value to the histogram. Recording a value in excess of the configured maximum value for that histogram results in recording the maximum value instead.

func (*Histogram) Snapshot

func (h *Histogram) Snapshot() *hdrhistogram.Histogram

Snapshot returns a copy of the cumulative (i.e. all-time samples) histogram data.

func (*Histogram) ToPrometheusMetric

func (h *Histogram) ToPrometheusMetric() *prometheusgo.Metric

ToPrometheusMetric returns a filled-in prometheus metric of the right type.

func (*Histogram) TotalCount

func (h *Histogram) TotalCount() int64

TotalCount returns the (cumulative) number of samples.

func (*Histogram) Windowed

func (h *Histogram) Windowed() (*hdrhistogram.Histogram, time.Duration)

Windowed returns a copy of the current windowed histogram data and its rotation interval.

type Iterable

type Iterable interface {
	// GetName returns the fully-qualified name of the metric.
	GetName() string
	// GetHelp returns the help text for the metric.
	GetHelp() string
	// Inspect calls the given closure with each contained item.
	Inspect(func(interface{}))
}

Iterable provides a method for synchronized access to interior objects.

type Metadata

type Metadata struct {
	Name, Help string
	// contains filtered or unexported fields
}

Metadata holds metadata about a metric. It must be embedded in each metric object.

func (*Metadata) AddLabel

func (m *Metadata) AddLabel(name, value string)

AddLabel adds a label/value pair for this metric.

func (*Metadata) GetHelp

func (m *Metadata) GetHelp() string

GetHelp returns the metric's help string.

func (*Metadata) GetLabels

func (m *Metadata) GetLabels() []*prometheusgo.LabelPair

GetLabels returns the metric's labels.

func (*Metadata) GetName

func (m *Metadata) GetName() string

GetName returns the metric's name.

type PrometheusExportable

type PrometheusExportable interface {
	// GetName is a method on Metadata
	GetName() string
	// GetHelp is a method on Metadata
	GetHelp() string
	// GetType returns the prometheus type enum for this metric.
	GetType() *prometheusgo.MetricType
	// GetLabels is a method on Metadata
	GetLabels() []*prometheusgo.LabelPair
	// ToPrometheusMetric returns a filled-in prometheus metric of the right type
	// for the given metric. It does not fill in labels.
	// The implementation must return thread-safe data to the caller, i.e.
	// usually a copy of internal state.
	ToPrometheusMetric() *prometheusgo.Metric
}

PrometheusExportable is the standard interface for an individual metric that can be exported to prometheus.

type PrometheusExporter

type PrometheusExporter struct {
	// contains filtered or unexported fields
}

PrometheusExporter contains a map of metric families (a metric with multiple labels). It initializes each metric family once and reuses it for each prometheus scrape. It is NOT thread-safe. TODO(marc): we should really keep our metric objects here so we can avoid creating new prometheus.Metric every time we are scraped. see: https://github.com/cockroachdb/cockroach/issues/9326

pe := MakePrometheusExporter()
pe.AddMetricsFromRegistry(nodeRegistry)
pe.AddMetricsFromRegistry(storeOneRegistry)
...
pe.AddMetricsFromRegistry(storeNRegistry)
pe.Export(w)

func MakePrometheusExporter

func MakePrometheusExporter() PrometheusExporter

MakePrometheusExporter returns an initialized prometheus exporter.

func (*PrometheusExporter) PrintAsText

func (pm *PrometheusExporter) PrintAsText(w io.Writer) error

PrintAsText writes all metrics in the families map to the io.Writer in prometheus' text format. It removes individual metrics from the families as it goes, readying the families for another found of registry additions.

func (*PrometheusExporter) ScrapeRegistry

func (pm *PrometheusExporter) ScrapeRegistry(registry *Registry)

ScrapeRegistry scrapes all metrics contained in the registry to the metric family map, holding on only to the scraped data (which is no longer connected to the registry and metrics within) when returning from the the call. It creates new families as needed.

type Rate

type Rate struct {
	// contains filtered or unexported fields
}

A Rate is a exponential weighted moving average.

func NewRate

func NewRate(timescale time.Duration) *Rate

NewRate creates an EWMA rate on the given timescale. Timescales at or below 2s are illegal and will cause a panic.

func (*Rate) Add

func (e *Rate) Add(v float64)

Add adds the given measurement to the Rate.

func (*Rate) Value

func (e *Rate) Value() float64

Value returns the current value of the Rate.

type Registry

type Registry struct {
	syncutil.Mutex
	// contains filtered or unexported fields
}

A Registry is a list of metrics. It provides a simple way of iterating over them, can marshal into JSON, and generate a prometheus format.

A registry can have label pairs that will be applied to all its metrics when exported to prometheus.

func NewRegistry

func NewRegistry() *Registry

NewRegistry creates a new Registry.

func (*Registry) AddLabel

func (r *Registry) AddLabel(name, value string)

AddLabel adds a label/value pair for this registry.

func (*Registry) AddMetric

func (r *Registry) AddMetric(metric Iterable)

AddMetric adds the passed-in metric to the registry.

func (*Registry) AddMetricStruct

func (r *Registry) AddMetricStruct(metricStruct interface{})

AddMetricStruct examines all fields of metricStruct and adds all Iterable or metricGroup objects to the registry.

func (*Registry) Each

func (r *Registry) Each(f func(name string, val interface{}))

Each calls the given closure for all metrics.

func (*Registry) MarshalJSON

func (r *Registry) MarshalJSON() ([]byte, error)

MarshalJSON marshals to JSON.

type Struct

type Struct interface {
	MetricStruct()
}

Struct can be implemented by the types of members of a metric container so that the members get automatically registered.

type TimeScale

type TimeScale struct {
	// contains filtered or unexported fields
}

A TimeScale is a named duration.