Python

def pf_callback(): # Note: in the real world these would be retrieved from the operating system return ( (8, (“pid”, 0), (“bitness”, 64)), (37741921, (“pid”, 4), (“bitness”, 64)), (10465, (“pid”, 880), (“bitness”, 32)), )

meter.create_observable_counter(name=“PF”, description=“process page faults”, pf_callback)


```python
# Python

def pf_callback(result):
    # Note: in the real world these would be retrieved from the operating system
    result.Observe(8,        ("pid", 0),   ("bitness", 64))
    result.Observe(37741921, ("pid", 4),   ("bitness", 64))
    result.Observe(10465,    ("pid", 880), ("bitness", 32))

meter.create_observable_counter(name="PF", description="process page faults", pf_callback)

// C#

// A simple scenario where only one value is reported

interface IAtomicClock
{
    UInt64 GetCaesiumOscillates();
}

IAtomicClock clock = AtomicClock.Connect();

meter.CreateObservableCounter<UInt64>("caesium_oscillates", () => clock.GetCaesiumOscillates());

Asynchronous Counter operations

Asynchronous Counter uses an idiomatic interface for reporting measurements through a callback, which is registered during Asynchronous Counter creation.

For callback functions registered after an asynchronous instrument is created, the API is required to support a mechanism for unregistration. For example, the object returned from register_callback can support an unregister() method directly.

# Python
class Device:
    """A device with one counter"""

    def __init__(self, meter, x):
        self.x = x
        counter = meter.create_observable_counter(name="usage", description="count of items used")
        self.cb = counter.register_callback(self.counter_callback)

    def counter_callback(self, result):
        result.Observe(self.read_counter(), {'x', self.x})

    def read_counter(self):
        return 100  # ...

    def stop(self):
        self.cb.unregister()

Histogram

Histogram is a synchronous Instrument which can be used to report arbitrary values that are likely to be statistically meaningful. It is intended for statistics such as histograms, summaries, and percentile.

Example uses for Histogram:

the request duration
the size of the response payload

Histogram creation

There MUST NOT be any API for creating a Histogram other than with a Meter. This MAY be called CreateHistogram. If strong type is desired, OpenTelemetry API authors MAY decide the language idiomatic name(s), for example CreateUInt64Histogram, CreateDoubleHistogram, CreateHistogram<UInt64>, CreateHistogram<double>.

See the general requirements for synchronous instruments.

Here are some examples that OpenTelemetry API authors might consider:

# Python

http_server_duration = meter.create_histogram(
    name="http.server.duration",
    description="measures the duration of the inbound HTTP request",
    unit="ms",
    value_type=float)

// C#

var httpServerDuration = meter.CreateHistogram<double>(
    "http.server.duration",
    description: "measures the duration of the inbound HTTP request",
    unit: "ms"
    );

Histogram operations

Record

Updates the statistics with the specified amount.

This API SHOULD NOT return a value (it MAY return a dummy value if required by certain programming languages or systems, for example null, undefined).

This API MUST accept the following parameter:

A numeric value to record.

The value needs to be provided by a user. If possible, this API SHOULD be structured so a user is obligated to provide this parameter. If it is not possible to structurally enforce this obligation, this API MUST be documented in a way to communicate to users that this parameter is needed.

The value is expected to be non-negative. This API SHOULD be documented in a way to communicate to users that this value is expected to be non-negative. This API SHOULD NOT validate this value, that is left to implementations of the API.
Attributes to associate with the value.

Users can provide attributes to associate with the value, but it is up to their discretion. Therefore, this API MUST be structured to accept a variable number of attributes, including none.

OpenTelemetry API authors MAY decide to allow flexible attributes to be passed in as individual arguments. OpenTelemetry API authors MAY allow attribute values to be passed in using a more efficient way (e.g. strong typed struct allocated on the callstack, tuple). Here are some examples that OpenTelemetry API authors might consider:

# Python

http_server_duration.Record(50, {"http.request.method": "POST", "url.scheme": "https"})
http_server_duration.Record(100, http_method="GET", http_scheme="http")

// C#

httpServerDuration.Record(50, ("http.request.method", "POST"), ("url.scheme", "https"));
httpServerDuration.Record(100, new HttpRequestAttributes { method = "GET", scheme = "http" });

Gauge

Status: Experimental

Gauge is a synchronous Instrument which can be used to record non-additive value(s) (e.g. the background noise level - it makes no sense to record the background noise level value from multiple rooms and sum them up) when changes occur.

Note: If the values are additive (e.g. the process heap size - it makes sense to report the heap size from multiple processes and sum them up, so we get the total heap usage), use UpDownCounter.

Note: Synchronous Gauge is normally used when the measurements are exposed via a subscription to change events ( i.e. backgroundNoiseLevel.onChange(value -> gauge.record(value))). If the measurement is exposed via an accessor, use Asynchronous Gauge to invoke the accessor in a callback function ( i.e. createObservableGauge(observable -> observable.record(backgroundNoiseLevel.getCurrentValue())).

Example uses for Gauge:

subscribe to change events for the background noise level
subscribe to change events for the CPU fan speed

Gauge creation

There MUST NOT be any API for creating a Gauge other than with a Meter. This MAY be called CreateGauge. If strong type is desired, OpenTelemetry API authors MAY decide the language idiomatic name(s), for example CreateUInt64Gauge, CreateDoubleGauge, CreateGauge<UInt64>, CreateGauge<double>.

See the general requirements for synchronous instruments.

Here are some examples that OpenTelemetry API authors might consider:

// Java

DoubleGauge backgroundNoiseLevel = meter.gaugeBuilder("facility.noise.level")
    .setDescription("Background noise level of rooms")
    .setUnit("B")
    .build();

Gauge operations

Record

Record the Gauge current value.

This API SHOULD NOT return a value (it MAY return a dummy value if required by certain programming languages or systems, for example null, undefined).

This API MUST accept the following parameter:

A numeric value. The current absolute value.

The value needs to be provided by a user. If possible, this API SHOULD be structured so a user is obligated to provide this parameter. If it is not possible to structurally enforce this obligation, this API MUST be documented in a way to communicate to users that this parameter is needed.
Attributes to associate with the value.

Users can provide attributes to associate with the value, but it is up to their discretion. Therefore, this API MUST be structured to accept a variable number of attributes, including none.

The OpenTelemetry API authors MAY decide to allow flexible attributes to be passed in as arguments. If the attribute names and types are provided during the gauge creation, the OpenTelemetry API authors MAY allow attribute values to be passed in using a more efficient way (e.g. strong typed struct allocated on the callstack, tuple). The API MUST allow callers to provide flexible attributes at invocation time rather than having to register all the possible attribute names during the instrument creation. Here are some examples that OpenTelemetry API authors might consider:

// Java
Attributes roomA = Attributes.builder().put("room.id", "Rack A");
Attributes roomB = Attributes.builder().put("room.id", "Rack B");

backgroundNoiseLevel.record(4.3, roomA);
backgroundNoiseLevel.record(2.5, roomB);

Asynchronous Gauge

Asynchronous Gauge is an asynchronous Instrument which reports non-additive value(s) (e.g. the room temperature - it makes no sense to report the temperature value from multiple rooms and sum them up) when the instrument is being observed.

Note: if the values are additive (e.g. the process heap size - it makes sense to report the heap size from multiple processes and sum them up, so we get the total heap usage), use Asynchronous Counter or Asynchronous UpDownCounter.

Example uses for Asynchronous Gauge:

the current room temperature
the CPU fan speed

Asynchronous Gauge creation

There MUST NOT be any API for creating an Asynchronous Gauge other than with a Meter. This MAY be called CreateObservableGauge. If strong type is desired, OpenTelemetry API authors MAY decide the language idiomatic name(s), for example CreateUInt64ObservableGauge, CreateDoubleObservableGauge, CreateObservableGauge<UInt64>, CreateObservableGauge<double>.

It is highly recommended that implementations use the name ObservableGauge (or any language idiomatic variation, e.g. observable_gauge) unless there is a strong reason not to do so. Please note that the name has nothing to do with asynchronous pattern and observer pattern.

See the general requirements for asynchronous instruments.

Here are some examples that OpenTelemetry API authors might consider:

# Python

def cpu_frequency_callback():
    # Note: in the real world these would be retrieved from the operating system
    return (
        (3.38, ("cpu", 0), ("core", 0)),
        (3.51, ("cpu", 0), ("core", 1)),
        (0.57, ("cpu", 1), ("core", 0)),
        (0.56, ("cpu", 1), ("core", 1)),
    )

meter.create_observable_gauge(
    name="cpu.frequency",
    description="the real-time CPU clock speed",
    callback=cpu_frequency_callback,
    unit="GHz",
    value_type=float)

# Python

def cpu_frequency_callback(result):
    # Note: in the real world these would be retrieved from the operating system
    result.Observe(3.38, ("cpu", 0), ("core", 0))
    result.Observe(3.51, ("cpu", 0), ("core", 1))
    result.Observe(0.57, ("cpu", 1), ("core", 0))
    result.Observe(0.56, ("cpu", 1), ("core", 1))

meter.create_observable_gauge(
    name="cpu.frequency",
    description="the real-time CPU clock speed",
    callback=cpu_frequency_callback,
    unit="GHz",
    value_type=float)

// C#

// A simple scenario where only one value is reported

meter.CreateObservableGauge<double>("temperature", () => sensor.GetTemperature());

Asynchronous Gauge operations

Asynchronous Gauge uses an idiomatic interface for reporting measurements through a callback, which is registered during Asynchronous Gauge creation.

# Python
class Device:
    """A device with one gauge"""

    def __init__(self, meter, x):
        self.x = x
        gauge = meter.create_observable_gauge(name="pressure", description="force/area")
        self.cb = gauge.register_callback(self.gauge_callback)

    def gauge_callback(self, result):
        result.Observe(self.read_gauge(), {'x', self.x})

    def read_gauge(self):
        return 100  # ...

    def stop(self):
        self.cb.unregister()

UpDownCounter

UpDownCounter is a synchronous Instrument which supports increments and decrements.

Note: if the value is monotonically increasing, use Counter instead.

Example uses for UpDownCounter:

the number of active requests
the number of items in a queue

An UpDownCounter is intended for scenarios where the absolute values are not pre-calculated, or fetching the “current value” requires extra effort. If the pre-calculated value is already available or fetching the snapshot of the “current value” is straightforward, use Asynchronous UpDownCounter instead.

UpDownCounter supports counting the size of a collection incrementally, e.g. reporting the number of items in a concurrent bag by the “color” and “material” properties as they are added and removed.

Color	Material	Count
Red	Aluminum	1
Red	Steel	2
Blue	Aluminum	0
Blue	Steel	5
Yellow	Aluminum	0
Yellow	Steel	3

# Python

items_counter = meter.create_up_down_counter(
    name="store.inventory",
    description="the number of the items available")

def restock_item(color, material):
    inventory.add_item(color=color, material=material)
    items_counter.add(1, {"color": color, "material": material})
    return true

def sell_item(color, material):
    succeeded = inventory.take_item(color=color, material=material)
    if succeeded:
        items_counter.add(-1, {"color": color, "material": material})
    return succeeded

UpDownCounter creation

There MUST NOT be any API for creating an UpDownCounter other than with a Meter. This MAY be called CreateUpDownCounter. If strong type is desired, OpenTelemetry API authors MAY decide the language idiomatic name(s), for example CreateInt64UpDownCounter, CreateDoubleUpDownCounter, CreateUpDownCounter<Int64>, CreateUpDownCounter<double>.

See the general requirements for synchronous instruments.

Here are some examples that OpenTelemetry API authors might consider:

# Python

customers_in_store = meter.create_up_down_counter(
    name="grocery.customers",
    description="measures the current customers in the grocery store",
    value_type=int)

// C#

var customersInStore = meter.CreateUpDownCounter<int>(
    "grocery.customers",
    description: "measures the current customers in the grocery store",
    );

UpDownCounter operations

Add

Increment or decrement the UpDownCounter by a fixed amount.

This API SHOULD NOT return a value (it MAY return a dummy value if required by certain programming languages or systems, for example null, undefined).

This API MUST accept the following parameter:

A numeric value to add.

The value needs to be provided by a user. If possible, this API SHOULD be structured so a user is obligated to provide this parameter. If it is not possible to structurally enforce this obligation, this API MUST be documented in a way to communicate to users that this parameter is needed.
Attributes to associate with the value.

Users can provide attributes to associate with the value, but it is up to their discretion. Therefore, this API MUST be structured to accept a variable number of attributes, including none.

# Python
customers_in_store.add(1, {"account.type": "commercial"})
customers_in_store.add(-1, account_type="residential")

// C#
customersInStore.Add(1, ("account.type", "commercial"));
customersInStore.Add(-1, new Account { Type = "residential" });

Asynchronous UpDownCounter

Asynchronous UpDownCounter is an asynchronous Instrument which reports additive value(s) (e.g. the process heap size - it makes sense to report the heap size from multiple processes and sum them up, so we get the total heap usage) when the instrument is being observed.

Note: if the value is monotonically increasing, use Asynchronous Counter instead; if the value is non-additive, use Asynchronous Gauge instead.

Example uses for Asynchronous UpDownCounter:

the process heap size
the approximate number of items in a lock-free circular buffer

Asynchronous UpDownCounter creation

There MUST NOT be any API for creating an Asynchronous UpDownCounter other than with a Meter. This MAY be called CreateObservableUpDownCounter. If strong type is desired, OpenTelemetry API authors MAY decide the language idiomatic name(s), for example CreateUInt64ObservableUpDownCounter, CreateDoubleObservableUpDownCounter, CreateObservableUpDownCounter<UInt64>, CreateObservableUpDownCounter<double>.

It is highly recommended that implementations use the name ObservableUpDownCounter (or any language idiomatic variation, e.g. observable_up_down_counter) unless there is a strong reason not to do so. Please note that the name has nothing to do with asynchronous pattern and observer pattern.

See the general requirements for asynchronous instruments.

Note: Unlike UpDownCounter.Add() which takes the increment/delta value, the callback function reports the absolute value of the Asynchronous UpDownCounter. To determine the reported rate the Asynchronous UpDownCounter is changing, the difference between successive measurements is used.

Here are some examples that OpenTelemetry API authors might consider:

# Python

def ws_callback():
    # Note: in the real world these would be retrieved from the operating system
    return (
        (8,      ("pid", 0),   ("bitness", 64)),
        (20,     ("pid", 4),   ("bitness", 64)),
        (126032, ("pid", 880), ("bitness", 32)),
    )

meter.create_observable_up_down_counter(
    name="process.workingset",
    description="process working set",
    callback=ws_callback,
    unit="kB",
    value_type=int)

# Python

def ws_callback(result):
    # Note: in the real world these would be retrieved from the operating system
    result.Observe(8,      ("pid", 0),   ("bitness", 64))
    result.Observe(20,     ("pid", 4),   ("bitness", 64))
    result.Observe(126032, ("pid", 880), ("bitness", 32))

meter.create_observable_up_down_counter(
    name="process.workingset",
    description="process working set",
    callback=ws_callback,
    unit="kB",
    value_type=int)

// C#

// A simple scenario where only one value is reported

meter.CreateObservableUpDownCounter<UInt64>("memory.physical.free", () => WMI.Query("FreePhysicalMemory"));

Asynchronous UpDownCounter operations

Asynchronous UpDownCounter uses an idiomatic interface for reporting measurements through a callback, which is registered during Asynchronous Updowncounter creation.

# Python
class Device:
    """A device with one up_down_counter"""

    def __init__(self, meter, x):
        self.x = x
        updowncounter = meter.create_observable_up_down_counter(name="queue_size", description="items in process")
        self.cb = updowncounter.register_callback(self.up_down_counter_callback)

    def up_down_counter_callback(self, result):
        result.Observe(self.read_up_down_counter(), {'x', self.x})

    def read_up_down_counter(self):
        return 100  # ...

    def stop(self):
        self.cb.unregister()

Measurement

A Measurement represents a data point reported via the metrics API to the SDK. Please refer to the Metrics Programming Model for the interaction between the API and SDK.

Measurements encapsulate:

A value
Attributes

Multiple-instrument callbacks

The Metrics API MAY support an interface allowing the use of multiple instruments from a single registered Callback. The API to register a new Callback SHOULD accept:

A callback function
A list (or tuple, etc.) of Instruments used in the callback function.

It is RECOMMENDED that the API authors use one of the following forms for the callback function:

The list (or tuple, etc.) returned by the callback function contains (Instrument, Measurement) pairs.
the Observable Result parameter receives an additional (Instrument, Measurement) pairs

This interface is typically a more performant way to report multiple measurements when they are obtained through an expensive process, such as reading /proc files or probing the garbage collection subsystem.

For example,

# Python
class Device:
    """A device with two instruments"""

    def __init__(self, meter, property):
        self.property = property
        self.usage = meter.create_observable_counter(name="usage", description="count of items used")
        self.pressure = meter.create_observable_gauge(name="pressure", description="force per unit area")

        # Note the two associated instruments are passed to the callback.
        meter.register_callback([self.usage, self.pressure], self.observe)

    def observe(self, result):
        usage, pressure = expensive_system_call()
        result.observe(self.usage, usage, {'property', self.property})
        result.observe(self.pressure, pressure, {'property', self.property})

Compatibility requirements

All the metrics components SHOULD allow new APIs to be added to existing components without introducing breaking changes.

All the metrics APIs SHOULD allow optional parameter(s) to be added to existing APIs without introducing breaking changes, if possible.

Concurrency requirements

For languages which support concurrent execution the Metrics APIs provide specific guarantees and safeties.

MeterProvider - all methods are safe to be called concurrently.

Meter - all methods are safe to be called concurrently.

Instrument - All methods of any Instrument are safe to be called concurrently.