Merge pull request #713 from prometheus/beorn7/promauto

Add a "factory" to make promauto work for custom registries
This commit is contained in:
Björn Rabenstein 2020-02-14 16:41:32 +01:00 committed by GitHub
commit b25ce2693a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 246 additions and 95 deletions

View File

@ -84,25 +84,21 @@
// of those four metric types can be found in the Prometheus docs:
// https://prometheus.io/docs/concepts/metric_types/
//
// A fifth "type" of metric is Untyped. It behaves like a Gauge, but signals the
// Prometheus server not to assume anything about its type.
//
// In addition to the fundamental metric types Gauge, Counter, Summary,
// Histogram, and Untyped, a very important part of the Prometheus data model is
// the partitioning of samples along dimensions called labels, which results in
// In addition to the fundamental metric types Gauge, Counter, Summary, and
// Histogram, a very important part of the Prometheus data model is the
// partitioning of samples along dimensions called labels, which results in
// metric vectors. The fundamental types are GaugeVec, CounterVec, SummaryVec,
// HistogramVec, and UntypedVec.
// and HistogramVec.
//
// While only the fundamental metric types implement the Metric interface, both
// the metrics and their vector versions implement the Collector interface. A
// Collector manages the collection of a number of Metrics, but for convenience,
// a Metric can also “collect itself”. Note that Gauge, Counter, Summary,
// Histogram, and Untyped are interfaces themselves while GaugeVec, CounterVec,
// SummaryVec, HistogramVec, and UntypedVec are not.
// a Metric can also “collect itself”. Note that Gauge, Counter, Summary, and
// Histogram are interfaces themselves while GaugeVec, CounterVec, SummaryVec,
// and HistogramVec are not.
//
// To create instances of Metrics and their vector versions, you need a suitable
// …Opts struct, i.e. GaugeOpts, CounterOpts, SummaryOpts, HistogramOpts, or
// UntypedOpts.
// …Opts struct, i.e. GaugeOpts, CounterOpts, SummaryOpts, or HistogramOpts.
//
// Custom Collectors and constant Metrics
//
@ -118,13 +114,16 @@
// existing numbers into Prometheus Metrics during collection. An own
// implementation of the Collector interface is perfect for that. You can create
// Metric instances “on the fly” using NewConstMetric, NewConstHistogram, and
// NewConstSummary (and their respective Must… versions). That will happen in
// the Collect method. The Describe method has to return separate Desc
// instances, representative of the “throw-away” metrics to be created later.
// NewDesc comes in handy to create those Desc instances. Alternatively, you
// could return no Desc at all, which will mark the Collector “unchecked”. No
// checks are performed at registration time, but metric consistency will still
// be ensured at scrape time, i.e. any inconsistencies will lead to scrape
// NewConstSummary (and their respective Must… versions). NewConstMetric is used
// for all metric types with just a float64 as their value: Counter, Gauge, and
// a special “type” called Untyped. Use the latter if you are not sure if the
// mirrored metric is a Counter or a Gauge. Creation of the Metric instance
// happens in the Collect method. The Describe method has to return separate
// Desc instances, representative of the “throw-away” metrics to be created
// later. NewDesc comes in handy to create those Desc instances. Alternatively,
// you could return no Desc at all, which will mark the Collector “unchecked”.
// No checks are performed at registration time, but metric consistency will
// still be ensured at scrape time, i.e. any inconsistencies will lead to scrape
// errors. Thus, with unchecked Collectors, the responsibility to not collect
// metrics that lead to inconsistencies in the total scrape result lies with the
// implementer of the Collector. While this is not a desirable state, it is

View File

@ -11,11 +11,16 @@
// See the License for the specific language governing permissions and
// limitations under the License.
// Package promauto provides constructors for the usual Prometheus metrics that
// return them already registered with the global registry
// (prometheus.DefaultRegisterer). This allows very compact code, avoiding any
// references to the registry altogether, but all the constructors in this
// package will panic if the registration fails.
// Package promauto provides alternative constructors for the fundamental
// Prometheus metric types and their …Vec and …Func variants. The difference to
// their counterparts in the prometheus package is that the promauto
// constructors return Collectors that are already registered with a
// registry. There are two sets of constructors. The constructors in the first
// set are top-level functions, while the constructors in the other set are
// methods of the Factory type. The top-level function return Collectors
// registered with the global registry (prometheus.DefaultRegisterer), while the
// methods return Collectors registered with the registry the Factory was
// constructed with. All constructors panic if the registration fails.
//
// The following example is a complete program to create a histogram of normally
// distributed random numbers from the math/rand package:
@ -79,51 +84,78 @@
// http.ListenAndServe(":1971", nil)
// }
//
// A Factory is created with the With(prometheus.Registerer) function, which
// enables two usage pattern. With(prometheus.Registerer) can be called once per
// line:
//
// var (
// reg = prometheus.NewRegistry()
// randomNumbers = promauto.With(reg).NewHistogram(prometheus.HistogramOpts{
// Name: "random_numbers",
// Help: "A histogram of normally distributed random numbers.",
// Buckets: prometheus.LinearBuckets(-3, .1, 61),
// })
// requestCount = promauto.With(reg).NewCounterVec(
// prometheus.CounterOpts{
// Name: "http_requests_total",
// Help: "Total number of HTTP requests by status code end method.",
// },
// []string{"code", "method"},
// )
// )
//
// Or it can be used to create a Factory once to be used multiple times:
//
// var (
// reg = prometheus.NewRegistry()
// factory = promauto.With(reg)
// randomNumbers = factory.NewHistogram(prometheus.HistogramOpts{
// Name: "random_numbers",
// Help: "A histogram of normally distributed random numbers.",
// Buckets: prometheus.LinearBuckets(-3, .1, 61),
// })
// requestCount = factory.NewCounterVec(
// prometheus.CounterOpts{
// Name: "http_requests_total",
// Help: "Total number of HTTP requests by status code end method.",
// },
// []string{"code", "method"},
// )
// )
//
// This appears very handy. So why are these constructors locked away in a
// separate package? There are two caveats:
// separate package?
//
// First, in more complex programs, global state is often quite problematic.
// That's the reason why the metrics constructors in the prometheus package do
// not interact with the global prometheus.DefaultRegisterer on their own. You
// are free to use the Register or MustRegister functions to register them with
// the global prometheus.DefaultRegisterer, but you could as well choose a local
// Registerer (usually created with prometheus.NewRegistry, but there are other
// scenarios, e.g. testing).
// The main problem is that registration may fail, e.g. if a metric inconsistent
// with the newly to be registered one is already registered. Therefore, the
// Register method in the prometheus.Registerer interface returns an error, and
// the same is the case for the top-level prometheus.Register function that
// registers with the global registry. The prometheus package also provides
// MustRegister versions for both. They panic if the registration fails, and
// they clearly call this out by using the Must… idiom. Panicking is a bit
// problematic here because it doesn't just happen on input provided by the
// caller that is invalid on its own. Things are a bit more subtle here: Metric
// creation and registration tend to be spread widely over the codebase. It can
// easily happen that an incompatible metric is added to an unrelated part of
// the code, and suddenly code that used to work perfectly fine starts to panic
// (provided that the registration of the newly added metric happens before the
// registration of the previously existing metric). This may come as an even
// bigger surprise with the global registry, where simply importing another
// package can trigger a panic (if the newly imported package registers metrics
// in its init function). At least, in the prometheus package, creation of
// metrics and other collectors is separate from registration. You first create
// the metric, and then you decide explicitly if you want to register it with a
// local or the global registry, and if you want to handle the error or risk a
// panic. With the constructors in the promauto package, registration is
// automatic, and if it fails, it will always panic. Furthermore, the
// constructors will often be called in the var section of a file, which means
// that panicking will happen as a side effect of merely importing a package.
//
// The second issue is that registration may fail, e.g. if a metric inconsistent
// with the newly to be registered one is already registered. But how to signal
// and handle a panic in the automatic registration with the default registry?
// The only way is panicking. While panicking on invalid input provided by the
// programmer is certainly fine, things are a bit more subtle in this case: You
// might just add another package to the program, and that package (in its init
// function) happens to register a metric with the same name as your code. Now,
// all of a sudden, either your code or the code of the newly imported package
// panics, depending on initialization order, without any opportunity to handle
// the case gracefully. Even worse is a scenario where registration happens
// later during the runtime (e.g. upon loading some kind of plugin), where the
// panic could be triggered long after the code has been deployed to
// production. A possibility to panic should be explicitly called out by the
// Must… idiom, cf. prometheus.MustRegister. But adding a separate set of
// constructors in the prometheus package called MustRegisterNewCounterVec or
// similar would be quite unwieldy. Adding an extra MustRegister method to each
// metric, returning the registered metric, would result in nice code for those
// using the method, but would pollute every single metric interface for
// everybody avoiding the global registry.
// A separate package allows conservative users to entirely ignore it. And
// whoever wants to use it, will do so explicitly, with an opportunity to read
// this warning.
//
// To address both issues, the problematic auto-registering and possibly
// panicking constructors are all in this package with a clear warning
// ahead. And whoever cares about avoiding global state and possibly panicking
// function calls can simply ignore the existence of the promauto package
// altogether.
//
// A final note: There is a similar case in the net/http package of the standard
// library. It has DefaultServeMux as a global instance of ServeMux, and the
// Handle function acts on it, panicking if a handler for the same pattern has
// already been registered. However, one might argue that the whole HTTP routing
// is usually set up closely together in the same package or file, while
// Prometheus metrics tend to be spread widely over the codebase, increasing the
// chance of surprising registration failures. Furthermore, the use of global
// state in net/http has been criticized widely, and some avoid it altogether.
// Enjoy promauto responsibly!
package promauto
import "github.com/prometheus/client_golang/prometheus"
@ -132,9 +164,7 @@ import "github.com/prometheus/client_golang/prometheus"
// but it automatically registers the Counter with the
// prometheus.DefaultRegisterer. If the registration fails, NewCounter panics.
func NewCounter(opts prometheus.CounterOpts) prometheus.Counter {
c := prometheus.NewCounter(opts)
prometheus.MustRegister(c)
return c
return With(prometheus.DefaultRegisterer).NewCounter(opts)
}
// NewCounterVec works like the function of the same name in the prometheus
@ -142,9 +172,7 @@ func NewCounter(opts prometheus.CounterOpts) prometheus.Counter {
// prometheus.DefaultRegisterer. If the registration fails, NewCounterVec
// panics.
func NewCounterVec(opts prometheus.CounterOpts, labelNames []string) *prometheus.CounterVec {
c := prometheus.NewCounterVec(opts, labelNames)
prometheus.MustRegister(c)
return c
return With(prometheus.DefaultRegisterer).NewCounterVec(opts, labelNames)
}
// NewCounterFunc works like the function of the same name in the prometheus
@ -152,45 +180,35 @@ func NewCounterVec(opts prometheus.CounterOpts, labelNames []string) *prometheus
// prometheus.DefaultRegisterer. If the registration fails, NewCounterFunc
// panics.
func NewCounterFunc(opts prometheus.CounterOpts, function func() float64) prometheus.CounterFunc {
g := prometheus.NewCounterFunc(opts, function)
prometheus.MustRegister(g)
return g
return With(prometheus.DefaultRegisterer).NewCounterFunc(opts, function)
}
// NewGauge works like the function of the same name in the prometheus package
// but it automatically registers the Gauge with the
// prometheus.DefaultRegisterer. If the registration fails, NewGauge panics.
func NewGauge(opts prometheus.GaugeOpts) prometheus.Gauge {
g := prometheus.NewGauge(opts)
prometheus.MustRegister(g)
return g
return With(prometheus.DefaultRegisterer).NewGauge(opts)
}
// NewGaugeVec works like the function of the same name in the prometheus
// package but it automatically registers the GaugeVec with the
// prometheus.DefaultRegisterer. If the registration fails, NewGaugeVec panics.
func NewGaugeVec(opts prometheus.GaugeOpts, labelNames []string) *prometheus.GaugeVec {
g := prometheus.NewGaugeVec(opts, labelNames)
prometheus.MustRegister(g)
return g
return With(prometheus.DefaultRegisterer).NewGaugeVec(opts, labelNames)
}
// NewGaugeFunc works like the function of the same name in the prometheus
// package but it automatically registers the GaugeFunc with the
// prometheus.DefaultRegisterer. If the registration fails, NewGaugeFunc panics.
func NewGaugeFunc(opts prometheus.GaugeOpts, function func() float64) prometheus.GaugeFunc {
g := prometheus.NewGaugeFunc(opts, function)
prometheus.MustRegister(g)
return g
return With(prometheus.DefaultRegisterer).NewGaugeFunc(opts, function)
}
// NewSummary works like the function of the same name in the prometheus package
// but it automatically registers the Summary with the
// prometheus.DefaultRegisterer. If the registration fails, NewSummary panics.
func NewSummary(opts prometheus.SummaryOpts) prometheus.Summary {
s := prometheus.NewSummary(opts)
prometheus.MustRegister(s)
return s
return With(prometheus.DefaultRegisterer).NewSummary(opts)
}
// NewSummaryVec works like the function of the same name in the prometheus
@ -198,18 +216,14 @@ func NewSummary(opts prometheus.SummaryOpts) prometheus.Summary {
// prometheus.DefaultRegisterer. If the registration fails, NewSummaryVec
// panics.
func NewSummaryVec(opts prometheus.SummaryOpts, labelNames []string) *prometheus.SummaryVec {
s := prometheus.NewSummaryVec(opts, labelNames)
prometheus.MustRegister(s)
return s
return With(prometheus.DefaultRegisterer).NewSummaryVec(opts, labelNames)
}
// NewHistogram works like the function of the same name in the prometheus
// package but it automatically registers the Histogram with the
// prometheus.DefaultRegisterer. If the registration fails, NewHistogram panics.
func NewHistogram(opts prometheus.HistogramOpts) prometheus.Histogram {
h := prometheus.NewHistogram(opts)
prometheus.MustRegister(h)
return h
return With(prometheus.DefaultRegisterer).NewHistogram(opts)
}
// NewHistogramVec works like the function of the same name in the prometheus
@ -217,7 +231,144 @@ func NewHistogram(opts prometheus.HistogramOpts) prometheus.Histogram {
// prometheus.DefaultRegisterer. If the registration fails, NewHistogramVec
// panics.
func NewHistogramVec(opts prometheus.HistogramOpts, labelNames []string) *prometheus.HistogramVec {
h := prometheus.NewHistogramVec(opts, labelNames)
prometheus.MustRegister(h)
return With(prometheus.DefaultRegisterer).NewHistogramVec(opts, labelNames)
}
// NewUntypedFunc works like the function of the same name in the prometheus
// package but it automatically registers the UntypedFunc with the
// prometheus.DefaultRegisterer. If the registration fails, NewUntypedFunc
// panics.
func NewUntypedFunc(opts prometheus.UntypedOpts, function func() float64) prometheus.UntypedFunc {
return With(prometheus.DefaultRegisterer).NewUntypedFunc(opts, function)
}
// Factory provides factory methods to create Collectors that are automatically
// registered with a Registerer. Create a Factory with the With function,
// providing a Registerer to auto-register created Collectors with. The zero
// value of a Factory creates Collectors that are not registered with any
// Registerer. All methods of the Factory panic if the registration fails.
type Factory struct {
r prometheus.Registerer
}
// With creates a Factory using the provided Registerer for registration of the
// created Collectors.
func With(r prometheus.Registerer) Factory { return Factory{r} }
// NewCounter works like the function of the same name in the prometheus package
// but it automatically registers the Counter with the Factory's Registerer.
func (f Factory) NewCounter(opts prometheus.CounterOpts) prometheus.Counter {
c := prometheus.NewCounter(opts)
if f.r != nil {
f.r.MustRegister(c)
}
return c
}
// NewCounterVec works like the function of the same name in the prometheus
// package but it automatically registers the CounterVec with the Factory's
// Registerer.
func (f Factory) NewCounterVec(opts prometheus.CounterOpts, labelNames []string) *prometheus.CounterVec {
c := prometheus.NewCounterVec(opts, labelNames)
if f.r != nil {
f.r.MustRegister(c)
}
return c
}
// NewCounterFunc works like the function of the same name in the prometheus
// package but it automatically registers the CounterFunc with the Factory's
// Registerer.
func (f Factory) NewCounterFunc(opts prometheus.CounterOpts, function func() float64) prometheus.CounterFunc {
c := prometheus.NewCounterFunc(opts, function)
if f.r != nil {
f.r.MustRegister(c)
}
return c
}
// NewGauge works like the function of the same name in the prometheus package
// but it automatically registers the Gauge with the Factory's Registerer.
func (f Factory) NewGauge(opts prometheus.GaugeOpts) prometheus.Gauge {
g := prometheus.NewGauge(opts)
if f.r != nil {
f.r.MustRegister(g)
}
return g
}
// NewGaugeVec works like the function of the same name in the prometheus
// package but it automatically registers the GaugeVec with the Factory's
// Registerer.
func (f Factory) NewGaugeVec(opts prometheus.GaugeOpts, labelNames []string) *prometheus.GaugeVec {
g := prometheus.NewGaugeVec(opts, labelNames)
if f.r != nil {
f.r.MustRegister(g)
}
return g
}
// NewGaugeFunc works like the function of the same name in the prometheus
// package but it automatically registers the GaugeFunc with the Factory's
// Registerer.
func (f Factory) NewGaugeFunc(opts prometheus.GaugeOpts, function func() float64) prometheus.GaugeFunc {
g := prometheus.NewGaugeFunc(opts, function)
if f.r != nil {
f.r.MustRegister(g)
}
return g
}
// NewSummary works like the function of the same name in the prometheus package
// but it automatically registers the Summary with the Factory's Registerer.
func (f Factory) NewSummary(opts prometheus.SummaryOpts) prometheus.Summary {
s := prometheus.NewSummary(opts)
if f.r != nil {
f.r.MustRegister(s)
}
return s
}
// NewSummaryVec works like the function of the same name in the prometheus
// package but it automatically registers the SummaryVec with the Factory's
// Registerer.
func (f Factory) NewSummaryVec(opts prometheus.SummaryOpts, labelNames []string) *prometheus.SummaryVec {
s := prometheus.NewSummaryVec(opts, labelNames)
if f.r != nil {
f.r.MustRegister(s)
}
return s
}
// NewHistogram works like the function of the same name in the prometheus
// package but it automatically registers the Histogram with the Factory's
// Registerer.
func (f Factory) NewHistogram(opts prometheus.HistogramOpts) prometheus.Histogram {
h := prometheus.NewHistogram(opts)
if f.r != nil {
f.r.MustRegister(h)
}
return h
}
// NewHistogramVec works like the function of the same name in the prometheus
// package but it automatically registers the HistogramVec with the Factory's
// Registerer.
func (f Factory) NewHistogramVec(opts prometheus.HistogramOpts, labelNames []string) *prometheus.HistogramVec {
h := prometheus.NewHistogramVec(opts, labelNames)
if f.r != nil {
f.r.MustRegister(h)
}
return h
}
// NewUntypedFunc works like the function of the same name in the prometheus
// package but it automatically registers the UntypedFunc with the Factory's
// Registerer.
func (f Factory) NewUntypedFunc(opts prometheus.UntypedOpts, function func() float64) prometheus.UntypedFunc {
u := prometheus.NewUntypedFunc(opts, function)
if f.r != nil {
f.r.MustRegister(u)
}
return u
}

View File

@ -28,7 +28,8 @@ import (
// ValueType is an enumeration of metric types that represent a simple value.
type ValueType int
// Possible values for the ValueType enum.
// Possible values for the ValueType enum. Use UntypedValue to mark a metric
// with an unknown type.
const (
_ ValueType = iota
CounterValue