zipkin.proto

//
// Copyright 2018-2019 The OpenZipkin Authors
//
// Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software distributed under the License
// is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
// or implied. See the License for the specific language governing permissions and limitations under
// the License.
//

syntax = "proto3";

package zipkin.proto3;

// In Java, the closest model type to this proto is in the "zipkin2" package
option java_package = "zipkin2.proto3";
option java_multiple_files = true;

// A span is a single-host view of an operation. A trace is a series of spans
// (often RPC calls) which nest to form a latency tree. Spans are in the same
// trace when they share the same trace ID. The parent_id field establishes the
// position of one span in the tree.
//
// The root span is where parent_id is Absent and usually has the longest
// duration in the trace. However, nested asynchronous work can materialize as
// child spans whose duration exceed the root span.
//
// Spans usually represent remote activity such as RPC calls, or messaging
// producers and consumers. However, they can also represent in-process
// activity in any position of the trace. For example, a root span could
// represent a server receiving an initial client request. A root span could
// also represent a scheduled job that has no remote context.
//
// Encoding notes:
//
// Epoch timestamp are encoded fixed64 as varint would also be 8 bytes, and more
// expensive to encode and size. Duration is stored uint64, as often the numbers
// are quite small.
//
// Default values are ok, as only natural numbers are used. For example, zero is
// an invalid timestamp and an invalid duration, false values for debug or shared
// are ignorable, and zero-length strings also coerce to null.
//
// The next id is 14.
//
// Note fields up to 15 take 1 byte to encode. Take care when adding new fields
// https://developers.google.com/protocol-buffers/docs/proto3#assigning-tags
message Span {
  // Randomly generated, unique identifier for a trace, set on all spans within
  // it.
  //
  // This field is required and encoded as 8 or 16 bytes, in big endian byte
  // order.
  bytes trace_id = 1;
  // The parent span ID or absent if this the root span in a trace.
  bytes parent_id = 2;
  // Unique identifier for this operation within the trace.
  //
  // This field is required and encoded as 8 opaque bytes.
  bytes id = 3;
  // When present, kind clarifies timestamp, duration and remote_endpoint. When
  // absent, the span is local or incomplete. Unlike client and server, there
  // is no direct critical path latency relationship between producer and
  // consumer spans.
  enum Kind {
    // Default value interpreted as absent.
    SPAN_KIND_UNSPECIFIED = 0;
    // The span represents the client side of an RPC operation, implying the
    // following:
    //
    // timestamp is the moment a request was sent to the server.
    // duration is the delay until a response or an error was received.
    // remote_endpoint is the server.
    CLIENT = 1;
    // The span represents the server side of an RPC operation, implying the
    // following:
    //
    // timestamp is the moment a client request was received.
    // duration is the delay until a response was sent or an error.
    // remote_endpoint is the client.
    SERVER = 2;
    // The span represents production of a message to a remote broker, implying
    // the following:
    //
    // timestamp is the moment a message was sent to a destination.
    // duration is the delay sending the message, such as batching.
    // remote_endpoint is the broker.
    PRODUCER = 3;
    // The span represents consumption of a message from a remote broker, not
    // time spent servicing it. For example, a message processor would be an
    // in-process child span of a consumer. Consumer spans imply the following:
    //
    // timestamp is the moment a message was received from an origin.
    // duration is the delay consuming the message, such as from backlog.
    // remote_endpoint is the broker.
    CONSUMER = 4;
  }
  // When present, used to interpret remote_endpoint
  Kind kind = 4;
  // The logical operation this span represents in lowercase (e.g. rpc method).
  // Leave absent if unknown.
  //
  // As these are lookup labels, take care to ensure names are low cardinality.
  // For example, do not embed variables into the name.
  string name = 5;
  // Epoch microseconds of the start of this span, possibly absent if
  // incomplete.
  //
  // For example, 1502787600000000 corresponds to 2017-08-15 09:00 UTC
  //
  // This value should be set directly by instrumentation, using the most
  // precise value possible. For example, gettimeofday or multiplying epoch
  // millis by 1000.
  //
  // There are three known edge-cases where this could be reported absent.
  // - A span was allocated but never started (ex not yet received a timestamp)
  // - The span's start event was lost
  // - Data about a completed span (ex tags) were sent after the fact
  fixed64 timestamp = 6;
  // Duration in microseconds of the critical path, if known. Durations of less
  // than one are rounded up. Duration of children can be longer than their
  // parents due to asynchronous operations.
  //
  // For example 150 milliseconds is 150000 microseconds.
  uint64 duration = 7;
  // The host that recorded this span, primarily for query by service name.
  //
  // Instrumentation should always record this. Usually, absent implies late
  // data. The IP address corresponding to this is usually the site local or
  // advertised service address. When present, the port indicates the listen
  // port.
  Endpoint local_endpoint = 8;
  // When an RPC (or messaging) span, indicates the other side of the
  // connection.
  //
  // By recording the remote endpoint, your trace will contain network context
  // even if the peer is not tracing. For example, you can record the IP from
  // the "X-Forwarded-For" header or the service name and socket of a remote
  // peer.
  Endpoint remote_endpoint = 9;
  // Associates events that explain latency with the time they happened.
  repeated Annotation annotations = 10;
  // Tags give your span context for search, viewing and analysis.
  //
  // For example, a key "your_app.version" would let you lookup traces by
  // version. A tag "sql.query" isn't searchable, but it can help in debugging
  // when viewing a trace.
  map<string, string> tags = 11;
  // True is a request to store this span even if it overrides sampling policy.
  //
  // This is true when the "X-B3-Flags" header has a value of 1.
  bool debug = 12;
  // True if we are contributing to a span started by another tracer (ex on a
  // different host).
  bool shared = 13;
}

// The network context of a node in the service graph.
//
// The next id is 5.
message Endpoint {
  // Lower-case label of this node in the service graph, such as "favstar".
  // Leave absent if unknown.
  //
  // This is a primary label for trace lookup and aggregation, so it should be
  // intuitive and consistent. Many use a name from service discovery.
  string service_name = 1;
  // 4 byte representation of the primary IPv4 address associated with this
  // connection. Absent if unknown.
  bytes ipv4 = 2;
  // 16 byte representation of the primary IPv6 address associated with this
  // connection. Absent if unknown.
  //
  // Prefer using the ipv4 field for mapped addresses.
  bytes ipv6 = 3;
  // Depending on context, this could be a listen port or the client-side of a
  // socket. Absent if unknown.
  int32 port = 4;
}

// Associates an event that explains latency with a timestamp.
// Unlike log statements, annotations are often codes. Ex. "ws" for WireSend
//
// The next id is 3.
message Annotation {
  // Epoch microseconds of this event.
  //
  // For example, 1502787600000000 corresponds to 2017-08-15 09:00 UTC
  //
  // This value should be set directly by instrumentation, using the most
  // precise value possible. For example, gettimeofday or multiplying epoch
  // millis by 1000.
  fixed64 timestamp = 1;
  // Usually a short tag indicating an event, like "error"
  //
  // While possible to add larger data, such as garbage collection details, low
  // cardinality event names both keep the size of spans down and also are easy
  // to search against.
  string value = 2;
}

// A list of spans with possibly different trace ids, in no particular order.
//
// This is used for all transports: POST, Kafka messages etc. No other fields
// are expected, This message facilitates the mechanics of encoding a list, as
// a field number is required. The name of this type is the same in the OpenApi
// aka Swagger specification. https://zipkin.io/zipkin-api/#/default/post_spans
message ListOfSpans {
  repeated Span spans = 1;
}

// Response for SpanService/Report RPC. This response currently does not return
// any information beyond indicating that the request has finished. That said,
// it may be extended in the future.
message ReportResponse {
}

// SpanService allows reporting spans using gRPC, as opposed to HTTP POST 
// reporting. Implementations are asynchronous and may drop spans for reasons 
// of sampling or storage availability. While this is primarily used to store 
// spans, other operations may take place such as aggregation of service 
// dependencies or data cleaning.
service SpanService {

  // Report the provided spans to the collector. Analogous to the HTTP POST 
  // /api/v2/spans endpoint. Spans are not required to be complete or belonging 
  // to the same trace.
  rpc Report(ListOfSpans) returns (ReportResponse) {}
}