First, the syntax with a pseudo EBNF grammar is presented. Then, the semantic of each field is described.
A JSON schema description of the syntax is available as semconv.schema.json,
see README.md for how to use it with an editor. The documentation
here in syntax.md
should be considered more authoritative though. Please keep
semconv.schema.json
in synch when changing the "grammar" in this file!
All attributes are lower case.
groups ::= semconv
| semconv groups
semconv ::= id [convtype] brief [note] [prefix] [extends] [stability] [deprecated] attributes [constraints] [specificfields]
id ::= string
convtype ::= "span" # Default if not specified
| "resource" # see spanspecificfields
| "event" # see eventspecificfields
| "metric" # see metricfields
| "metric_group"
| "scope"
| "attribute_group" # no specific fields defined
brief ::= string
note ::= string
prefix ::= string
extends ::= string
stability ::= "experimental"
| "stable"
deprecated ::= <description>
attributes ::= (id type brief examples | ref [brief] [examples]) [tag] [stability] [deprecated] [required] [sampling_relevant] [note]
# ref MUST point to an existing attribute id
ref ::= id
type ::= simple_type
| template_type
| enum
simple_type ::= "string"
| "int"
| "double"
| "boolean"
| "string[]"
| "int[]"
| "double[]"
| "boolean[]"
template_type ::= "template[" simple_type "]" # As a single string
enum ::= [allow_custom_values] members
allow_custom_values := boolean
members ::= member {member}
member ::= id value [brief] [note] [stability] [deprecated]
requirement_level ::= "required"
| "conditionally_required" <condition>
| "recommended" [condition] # Default if not specified
| "opt_in"
sampling_relevant ::= boolean
examples ::= <example_value> {<example_value>}
constraints ::= constraint {constraint}
constraint ::= any_of
| include
any_of ::= id {id}
include ::= id
specificfields ::= spanfields
| eventfields
| metricfields
spanfields ::= [events] [span_kind]
eventfields ::= [name]
span_kind ::= "client"
| "server"
| "producer"
| "consumer"
| "internal"
events ::= id {id} # MUST point to an existing event group
name ::= string
metricfields ::= metric_name instrument unit
metric_name ::= string
instrument ::= "counter"
| "histogram"
| "gauge"
| "updowncounter"
unit ::= string
Groups contain the list of semantic conventions and it is the root node of each yaml file.
The field semconv
represents a semantic convention and it is made by:
id
, string that uniquely identifies the semantic convention.type
, optional enum, defaults tospan
(with a warning if not present).brief
, string, a brief description of the semantic convention.note
, optional string, a more elaborate description of the semantic convention. It defaults to an empty string.prefix
, optional string, prefix for the attributes for this semantic convention. It defaults to an empty string.extends
, optional string, reference another semantic conventionid
. It inherits the prefix, constraints, and all attributes defined in the specified semantic convention.deprecated
, optional, when present marks the semantic convention as deprecated. The string provided as<description>
MUST specify why it's deprecated and/or what to use instead.attributes
, list of attributes that belong to the semantic convention.constraints
, optional list, additional constraints (See later). It defaults to an empty list.
The following is only valid if type
is span
(the default):
span_kind
, optional enum, specifies the kind of the span.events
, optional list of strings that specify the ids of event semantic conventions associated with this span semantic convention.
The following is only valid if type
is event
:
name
, conditionally required string. The name of the event. If not specified, theprefix
is used. Ifprefix
is empty (or unspecified),name
is required.
Metric group inherits all from the base semantic convention, and does not add any additional fields.
The metric group semconv is a group where related metric attributes
can be defined and then referenced from other metric
groups using ref
.
The following is only valid if type
is metric
:
metric_name
, required, the metric name as described by the OpenTelemetry Specification.instrument
, required, the instrument type that should be used to record the metric. Note that the semantic conventions must be written using the names of the synchronous instrument types (counter
,gauge
,updowncounter
andhistogram
). For more details: Metrics semantic conventions - Instrument types.unit
, required, the unit in which the metric is measured, which should adhere to the guidelines.
Attribute group (attribute_group
type) defines a set of attributes that can be
declared once and referenced by semantic conventions for different signals, for example spans and logs.
Attribute groups don't have any specific fields and follow the general semconv
semantics.
An attribute is defined by:
id
, string that uniquely identifies the attribute. Required.type
, either a string literal denoting the type as a primitive or an array type, a template type or an enum definition (See later). Required. The accepted string literals are:- primitive and array types as string literals:
"string"
: String attributes."int"
: Integer attributes."double"
: Double attributes."boolean"
: Boolean attributes."string[]"
: Array of strings attributes."int[]"
: Array of integer attributes."double[]"
: Array of double attributes."boolean[]"
: Array of booleans attributes.
- template type as string literal:
"template[<PRIMITIVE_OR_ARRAY_TYPE>]"
(See below) See the specification of Attributes for the definition of the value types.
- primitive and array types as string literals:
stability
, enum - eitherstable
orexperimental
, specifies the stability of the attribute. Required.ref
, optional string, reference an existing attribute, see below.tag
, optional string, associates a tag ("sub-group") to the attribute. It carries no particular semantic meaning but can be used e.g. for filtering in the markdown generator.requirement_level
, optional, specifies if the attribute is mandatory. Can be "required", "conditionally_required", "recommended" or "opt_in". When omitted, the attribute is "recommended". When set to "conditionally_required", the string provided as<condition>
MUST specify the conditions under which the attribute is required.sampling_relevant
, optional boolean, specifies if the attribute is (especially) relevant for sampling and thus should be set at span start. It defaults tofalse
.brief
,note
,deprecated
, same meaning as for the whole semantic convention, but per attribute.examples
, sequence of example values for the attribute or single example value. They are required only for string and string array attributes. Example values must be of the same type of the attribute. If only a single example is provided, it can directly be reported without encapsulating it into a sequence/dictionary. See below.
Examples for setting the examples
field:
A single example value for a string attribute. All the following three representations are equivalent:
examples: 'this is a single string'
or
examples: ['this is a single string']
or
examples:
- 'this is a single string'
Attention, the following will throw a type mismatch error because a string type as example value is expected and not an array of string:
examples:
- ['this is an error']
examples: [['this is an error']]
Multiple example values for a string attribute:
examples: ['this is a single string', 'this is another one']
or
examples:
- 'this is a single string'
- 'this is another one'
A single example value for an array of strings attribute:
examples: ['first element of first array', 'second element of first array']
or
examples:
- ['first element of first array', 'second element of first array']
Attention, the following will throw a type mismatch error because an array of strings as type for the example values is expected and not a string:
examples: 'this is an error'
Multiple example values for an array of string attribute:
examples: [ ['first element of first array', 'second element of first array'], ['first element of second array', 'second element of second array'] ]
or
examples:
- ['first element of first array', 'second element of first array']
- ['first element of second array', 'second element of second array']
ref
MUST have an id of an existing attribute. When it is set, id
, type
, stability
, and deprecation
MUST NOT be present.
ref
is useful for specifying that an existing attribute of another semantic convention is part of
the current semantic convention and inherit its brief
, note
, and example
values. However, if these
fields are present in the current attribute definition, they override the inherited values.
An attribute type can either be a string, int, double, boolean, array of strings, array of int, array of double, array of booleans, a template type or an enumeration.
A template type attribute represents a dictionary of attributes with a common attribute name prefix. The syntax for defining template type attributes is the following:
type: template[<PRIMITIVE_OR_ARRAY_TYPE>]
The <PRIMITIVE_OR_ARRAY_TYPE>
is one of the above-mentioned primitive or array types (not an enum) and specifies the type of the value
in the dictionary.
The following is an example for defining a template type attribute and it's resolution:
groups:
- id: trace.http.common
type: attribute_group
brief: "..."
attributes:
- id: http.request.header
type: template[string[]]
stability: stable
brief: >
HTTP request headers, the key being the normalized HTTP header name (lowercase, with `-` characters replaced by `_`), the value being the header values.
examples: ['http.request.header.content_type=["application/json"]', 'http.request.header.x_forwarded_for=["1.2.3.4", "1.2.3.5"]']
note: |
...
In this example the definition will be resolved into a dictionary of attributes http.request.header.<key>
where <key>
will be replaced by the actual HTTP header name, and the value of the attributes is of type string[]
that carries the HTTP header value.
If the type is an enumeration, additional fields are required:
allow_custom_values
, optional boolean, set to false to not accept values other than the specified members. It defaults totrue
.members
, list of enum entries.
An enum entry has the following fields:
id
, string that uniquely identifies the enum entry.value
, string, int, or boolean; value of the enum entry.brief
, optional string, brief description of the enum entry value. It defaults to the value ofid
.note
, optional string, longer description. It defaults to an empty string.stability
, required stability level. Attributes marked as experimental cannot have stable members.deprecated
, optional string, similarly to semantic convention and attribute deprecation, marks specific member as deprecated.
Allow to define additional requirements on the semantic convention.
Currently, it supports any_of
and include
.
any_of
accepts a list of sequences. Each sequence contains a list of attribute ids that are required.
any_of
enforces that all attributes of at least one of the sequences are set.
include
accepts a semantic conventions id
. It includes as part of this semantic convention all constraints
and required attributes that are not already defined in the current semantic convention.