This quick start is for end users of OpenTelemetry who wish to manually trace their applications. If you are a library author, please see the Library Authors Guide. If you wish to automatically instrument your application, see the automatic instrumentation documentation for the SDK you wish to use.
For a high-level overview of OpenTelemetry tracing in general and definitions of some common terms, you can refer to the OpenTelemetry Specification Overview
Trace API Specification: https://github.com/open-telemetry/opentelemetry-specification/blob/v1.6.0/specification/trace/api.md
Trace API Reference: https://open-telemetry.github.io/opentelemetry-js-api/classes/traceapi.html
In OpenTelemetry, tracing operations are performed using methods on a tracer. You can get a tracer by calling getTracer
on the global tracer provider. getTracer
takes the name and version of the application or library acquiring the tracer, and provides a tracer which can be used to trace operations.
import { trace } from '@opentelemetry/api';
const tracer = trace.getTracer("my-application", "0.1.0");
In OpenTelemetry, all traces are composed of Spans
. A span describes a single operation with a start time and and end time like a database request, outgoing remote request, or a function invocation. These spans are linked together by parent-child relationships to form a tree. The resultant tree is your trace, and the root of the tree is commonly called the root span.
You can create a span by calling Tracer#startSpan
. The only required argument to startSpan
is the span name, which should describe the operation being performed with low cardinality.
const span = tracer.startSpan("my-span-name");
// do some work
// When a span is ended, it will be exported to a tracing backend
// via the currently registered SDK.
span.end();
Most of the time, spans will be used as part of a function which responds to some event like a web request. The following example shows what it might look like to manually trace a function which responds to a get request using an imaginary http server framework.
async function onGet(request, response) {
const span = tracer.startSpan("onGet");
try {
// Do some work here
response.send();
// If we get here and nothing has thrown, the request completed successfully
span.setStatus({ code: SpanStatusCode.OK });
} catch (err) {
// When we catch an error, we want to show that an error occurred
span.setStatus({
code: SpanStatusCode.ERROR,
message: err.message,
});
} finally {
// Every span must be ended or it will not be exported
span.end();
}
}
server.on("GET", "/user/:id", onGet);
Using span relationships, attributes, kind, and the related semantic conventions, we can more accurately describe the span in a way our tracing backend will more easily understand. The following example uses these mechanisms, which are described below.
import { NetTransportValues SemanticAttributes } from '@opentelemetry/semantic-conventions';
import { trace, context, SpanKind, SpanStatusCode } from '@opentelemetry/api';
async function onGet(request, response) {
// HTTP semantic conventions determine the span name and attributes for this span
const span = tracer.startSpan(`GET /user/:id`, {
// attributes can be added when the span is started
attributes: {
// Attributes from the HTTP trace semantic conventions
// https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/semantic_conventions/http.md
[SemanticAttributes.HTTP_METHOD]: "GET",
[SemanticAttributes.HTTP_FLAVOR]: "1.1",
[SemanticAttributes.HTTP_URL]: request.url,
[SemanticAttributes.NET_PEER_IP]: "192.0.2.5",
},
// This span represents a remote incoming synchronous request
kind: SpanKind.SERVER
});
const userId = request.params.id;
// Create a new context from the current context which has the span "active"
const ctx = trace.setSpan(context.active(), span);
// Call getUser with the newly created context
//
// context.with calls a function with an associated "active" context. Within
// the function, calling context.active() returns the currently active context.
// If there is no active context, the ROOT_CONTEXT will be returned, which
// has no key-value pairs.
//
// context.with requires at least 2 arguments: a context and a function to be called.
// If a third argument is provided, it will be bound to `this` `this` inside the function.
// Any additional parameters are used as arguments when calling the function.
//
// Return value is the value returned from getUser
// | Context to be used as the "active" context
// | | Function to be called
// | | | Object assigned to this during function execution
// | | | | Passed as the first argument to getUser
// | | | | |
// V V V V V
const user = await context.with(ctx, getUser, undefined, userId);
// Attributes may also be added after the span is started.
// http.status_code is required by the HTTP trace semantic conventions
span.setAttribute("http.status_code", 200);
response.send(user.toJson());
span.setStatus({
code: SpanStatusCode.OK,
});
span.end();
// Attributes MAY NOT be added after the span ends
span.setAttribute("my.attribute", false); // this does nothing
}
async function getUser(userId) {
// when this span is created, it will automatically use the span from the context as its parent
const span = tracer.startSpan("SELECT ShopDb.Users", {
attributes: {
// Attributes from the database trace semantic conventions
// https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/semantic_conventions/database.md
[SemanticAttributes.DB_SYSTEM]: "mysql",
[SemanticAttributes.DB_CONNECTION_STRING]: "Server=shopdb.example.com;Database=ShopDb;Uid=billing_user;TableCache=true;UseCompression=True;MinimumPoolSize=10;MaximumPoolSize=50;",
[SemanticAttributes.DB_USER]: "app_user",
[SemanticAttributes.NET_PEER_NAME]: "shopdb.example.com",
[SemanticAttributes.NET_PEER_IP]: "192.0.2.12",
[SemanticAttributes.NET_PEER_PORT]: 3306,
[SemanticAttributes.NET_TRANSPORT]: NetTransportValues.IP_TCP,
[SemanticAttributes.DB_NAME]: "ShopDb",
[SemanticAttributes.DB_STATEMENT]: `Select * from Users WHERE user_id = ${userId}`,
[SemanticAttributes.DB_OPERATION]: "SELECT",
[SemanticAttributes.DB_SQL_TABLE]: "Users",
},
kind: SpanKind.CLIENT,
});
const user = await db.select("Users", { id: userId });
span.setStatus({
code: SpanStatusCode.OK,
});
span.end();
return user;
}
server.on("GET", "/user/:id", onGet);
One of the most important aspects of spans is their relationships to each other. For instance, if one span describes an incoming request which makes a database call, it is recommended to trace the database call as a separate span which is a child of the original request span. In order to do this, when we create a span we can tell OpenTelemetry which span to use as its parent using a mechanism called Context.
Context is a very important part of the OpenTelemetry API which cannot be adequately explained in a single paragraph. To read more about context, see the context documentation.
While name, start time, end time, and status are the minimum information required to trace an operation, most of the time they will not be enough information on their own to effectively observe an application. To solve this, OpenTelemetry uses Span Attributes. Span attributes are an object with string keys and string, number, or boolean values which describe the span. For example, we can use the span attributes to add route and http response code information to the example above.
When a span is created, it is one of Client
, Server
, Internal
, Producer
, or Consumer
. This span kind provides a hint to the tracing backend as to how the trace should be assembled. According to the OpenTelemetry specification, the parent of a server span is always a client span, and the child of a client span is always a server span. Similarly, the parent of a consumer span is always a producer and the child of a producer span is always a consumer. If not provided, the span kind is assumed to be internal.
For more information regarding SpanKind, see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/api.md#spankind.
Client spans represent a synchronous outgoing remote call such as an outgoing HTTP request or database call. Note that in this context, "synchronous" does not refer to async/await
, but to the fact that it is not queued for later processing.
Server spans represent a synchronous incoming remote call such as an incoming HTTP request or remote procedure call.
Internal spans represent operations which do not cross a process boundary. Things like instrumenting a function call or an express middleware may use internal spans.
Producer spans represent the creation of a job which may be asynchronously processed later. It may be a remote job such as one inserted into a job queue or a local job handled by an event listener.
Consumer spans represent the processing of a job created by a producer and may start long after the producer span has already ended.
One problem with span names and attributes is recognizing, categorizing, and analyzing them in your tracing backend. Between different applications, libraries, and tracing backends there might be different names and expected values for various attributes. For example, your application may use http.status
to describe the HTTP status code, but a library you use may use http.status_code
. In order to solve this problem, OpenTelemetry uses a library of semantic conventions which describe the name and attributes which should be used for specific types of spans. The use of semantic conventions is always recommended where applicable, but they are merely conventions. For example, you may find that some name other than the name suggested by the semantic conventions more accurately describes your span, you may decide not to include a span attribute which is suggested by semantic conventions for privacy reasons, or you may wish to add a custom attribute which isn't covered by semantic conventions. All of these cases are fine, but please keep in mind that if you stray from the semantic conventions, the categorization of spans in your tracing backend may be affected.
See the current trace semantic conventions in the OpenTelemetry Specification repository: https://github.com/open-telemetry/opentelemetry-specification/tree/main/specification/trace/semantic_conventions