Distributed Tracing

Distributed tracing is a way to track and correlate requests across microservices. Distributed tracing must first be implemented in each application, it cannot be added by Consul. Once implemented in your applications, adding distributed tracing to Consul will add the sidecar proxies as spans in the request path.

Application Changes

Consul alone cannot implement distributed tracing for your applications. Each application must propagate the required headers. Typically this is done using a tracing library such as:

Configuration

Once your applications have been instrumented with a tracing library, you are ready to configure Consul to add sidecar proxy spans to the trace. Your eventual config will look something like:

  1. Kind = "proxy-defaults"
  2. Name = "global"
  3. Config {
  4.   protocol           = "http"
  5.   envoy_tracing_json = <<EOF
  6. {
  7.   "http":{
  8.     "name":"envoy.tracers.zipkin",
  9.     "typedConfig":{
  10.       "@type":"type.googleapis.com/envoy.config.trace.v3.ZipkinConfig",
  11.       "collector_cluster":"collector_cluster_name",
  12.       "collector_endpoint_version":"HTTP_JSON",
  13.       "collector_endpoint":"/api/v2/spans",
  14.       "shared_span_context":false
  15.     }
  16.   }
  17. }
  18. EOF
  20.   envoy_extra_static_clusters_json = <<EOF
  21. {
  22.   "connect_timeout":"3.000s",
  23.   "dns_lookup_family":"V4_ONLY",
  24.   "lb_policy":"ROUND_ROBIN",
  25.   "load_assignment":{
  26.     "cluster_name":"collector_cluster_name",
  27.     "endpoints":[
  28.       {
  29.         "lb_endpoints":[
  30.           {
  31.             "endpoint":{
  32.               "address":{
  33.                 "socket_address":{
  34.                   "address":"collector-url",
  35.                   "port_value":9411,
  36.                   "protocol":"TCP"
  37.                 }
  38.               }
  39.             }
  40.           }
  41.         ]
  42.       }
  43.     ]
  44.   },
  45.   "name":"collector_cluster_name",
  46.   "type":"STRICT_DNS"
  47. }
  48. EOF
  49. }
  1. apiVersion: consul.hashicorp.com/v1alpha1
  2. kind: ProxyDefaults
  3. metadata:
  4.   name: global
  5. spec:
  6.   config:
  7.     protocol: http
  8.     envoy_tracing_json: |
  9.       {
  10.         "http":{
  11.           "name":"envoy.tracers.zipkin",
  12.           "typedConfig":{
  13.             "@type":"type.googleapis.com/envoy.config.trace.v3.ZipkinConfig",
  14.             "collector_cluster":"collector_cluster_name",
  15.             "collector_endpoint_version":"HTTP_JSON",
  16.             "collector_endpoint":"/api/v2/spans",
  17.             "shared_span_context":false
  18.           }
  19.         }
  20.       }
  21.     envoy_extra_static_clusters_json: |
  22.       {
  23.         "connect_timeout":"3.000s",
  24.         "dns_lookup_family":"V4_ONLY",
  25.         "lb_policy":"ROUND_ROBIN",
  26.         "load_assignment":{
  27.           "cluster_name":"collector_cluster_name",
  28.           "endpoints":[
  29.             {
  30.               "lb_endpoints":[
  31.                 {
  32.                   "endpoint":{
  33.                     "address":{
  34.                       "socket_address":{
  35.                         "address":"collector-url",
  36.                         "port_value":9411,
  37.                         "protocol":"TCP"
  38.                       }
  39.                     }
  40.                   }
  41.                 }
  42.               ]
  43.             }
  44.           ]
  45.         },
  46.         "name":"collector_cluster_name",
  47.         "type":"STRICT_DNS"
  48.       }
  1. {
  2.   "Kind": "ProxyDefaults",
  3.   "Name": "global",
  4.   "Config": {
  5.     "protocol": "http",
  6.     "envoy_tracing_json": "{\"http\":{\"name\":\"envoy.tracers.zipkin\",\"typedConfig\":{\"@type\":\"type.googleapis.com/envoy.config.trace.v3.ZipkinConfig\",\"collector_cluster\":\"collector_cluster_name\",\"collector_endpoint_version\":\"HTTP_JSON\",\"collector_endpoint\":\"/api/v2/spans\",\"shared_span_context\":false}}}",
  7.     "envoy_extra_static_clusters_json": "{\"connect_timeout\":\"3.000s\",\"dns_lookup_family\":\"V4_ONLY\",\"lb_policy\":\"ROUND_ROBIN\",\"load_assignment\":{\"cluster_name\":\"collector_cluster_name\",\"endpoints\":[{\"lb_endpoints\":[{\"endpoint\":{\"address\":{\"socket_address\":{\"address\":\"collector-url\",\"port_value\":9411,\"protocol\":\"TCP\"}}}}]}]},\"name\":\"collector_cluster_name\",\"type\":\"STRICT_DNS\"}"
  8.   }
  9. }

NOTE: This example uses a proxy defaults config entry which will apply to all proxies but you can also apply this config in the proxy service registration (not supported on Kubernetes).

Within the config there are two keys you need to customize:

  1. envoy_tracing_json: Sets the tracing configuration for your specific tracing type. See the Envoy tracers documentation for your specific collector’s configuration. This configuration will reference the cluster name defined in envoy_extra_static_clusters_json.
  2. envoy_extra_static_clusters_json: Defines the address of your tracing collector where Envoy will send its spans. In this example the URL was collector-url:9411.

Applying the configuration

This configuration only applies when proxies are restarted since it changes the bootstrap config for Envoy which can only be applied on startup. This means you must restart all your proxies for changes to this config to take effect.

Note: On Kubernetes this is a matter of restarting your deployments, e.g. kubectl rollout restart deploy/deploy-name.

Considerations

  1. Distributed tracing is only supported for HTTP and gRPC services. You must specify the protocol either globally via a proxy defaults config entry:

    1. Kind      = "proxy-defaults"
    2. Name      = "global"
    3. Config {
    4.   protocol = "http"
    5. }
    1. apiVersion: consul.hashicorp.com/v1alpha1
    2. kind: ProxyDefaults
    3. metadata:
    4.   name: global
    5. spec:
    6.   config:
    7.     protocol: http
    1. {
    2.   "Kind": "proxy-defaults",
    3.   "Name": "global",
    4.   "Config": {
    5.     "protocol": "http"
    6.   }
    7. }

    Or via a service defaults config entry for each service:

    1. Kind      = "service-defaults"
    2. Name      = "service-name"
    3. Protocol  = "http"
    1. apiVersion: consul.hashicorp.com/v1alpha1
    2. kind: ProxyDefaults
    3. metadata:
    4.   name: service-name
    5. spec:
    6.   protocol: http
    1. {
    2.   "Kind": "service-defaults",
    3.   "Name": "service-name",
    4.   "Protocol": "http"
    5. }

  2. Requests through Ingress Gateways will not be traced unless the header x-client-trace-id: 1 is set (see hashicorp/consul#6645).

  3. Consul’s proxies do not currently support OpenTelemetry spans, as Envoy has not fully implemented it. Instead, you can add OpenTelemetry libraries to your application to emit spans for other tracing protocols supported by Envoy, such as Zipkin or Jaeger.

  4. Tracing is only supported with Envoy proxies, not the built-in proxy.

  5. When configuring the Zipkin tracer in envoy_tracing_json, set trace_id_128bit to true if your application is configured to generate 128-bit trace IDs. For example:

    1. {
    2.   "http": {
    3.     "name": "envoy.tracers.zipkin",
    4.     "typedConfig": {
    5.       "@type": "type.googleapis.com/envoy.config.trace.v3.ZipkinConfig",
    6.       "collector_cluster": "zipkin",
    7.       "collector_endpoint_version": "HTTP_JSON",
    8.       "collector_endpoint": "/api/v2/spans",
    9.       "shared_span_context": false,
    10.       "trace_id_128bit": true
    11.     }
    12.   }
    13. }