Navigation :

Tracing Setup

Tracing

Gloo Edge makes it easy to implement tracing on your system through Envoy’s tracing capabilities.

Usage

If you have not yet enabled tracing, please see the configuration details below.

Produce a trace by passing the header: x-client-trace-id
- This id provides a means of associating the spans produced during a trace. The value must be unique, a uuid4 is recommended.
Optionally annotate your trace with the x-envoy-decorator-operation header.
- This will be emitted with the resulting trace and can be a means of identifying the origin of a given trace. Note that it will override any pre-specified route decorator. Additional details can be found here.

Configuration

There are a few steps to make tracing available through Gloo Edge:

Configure a tracing cluster
Configure a tracing provider
(Optional) Annotate routes with descriptors

1. Configure a tracing cluster

Tracing requires a cluster that will collect the traces. For example, Zipkin requires a collector_cluster to be specified in the bootstrap config. If your provider requires a cluster to be specified, you can provide it in the config, as shown below.

The bootstrap config is the portion of Envoy’s config that is applied when an Envoy process is initialized. That means that you must either apply this configuration through Helm values during installation or that you must edit the proxy’s config map and restart the pod. We describe both methods below.

Option 1: Set the tracing cluster through helm values:

gatewayProxies:
  gatewayProxy:
    tracing:
      cluster:
        - name: zipkin
          connect_timeout: 1s
          type: STRICT_DNS
          load_assignment:
            cluster_name: zipkin
            endpoints:
            - lb_endpoints:
              - endpoint:
                  address:
                    socket_address:
                      address: zipkin
                      port_value: 1234

When you install Gloo Edge using these Helm values, Envoy will be configured with the tracing cluster you specified.

Option 2: Set the tracing cluster by editing the config map:

First, edit the config map pertaining to your proxy. This should be gateway-proxy-envoy-config in the gloo-system namespace.

kubectl edit configmap -n gloo-system gateway-proxy-envoy-config

Apply the tracing cluster changes. A sample Zipkin configuration is shown below.

apiVersion: v1
kind: ConfigMap
data:
  envoy.yaml:
    node:
      cluster: gateway
      id: "{{.PodName}}{{.PodNamespace}}"
      metadata:
        role: "{{.PodNamespace}}~gateway-proxy"
    static_resources:
      clusters:
        - name: xds_cluster
          connect_timeout: 5.000s
          load_assignment:
            cluster_name: xds_cluster
            endpoints:
            - lb_endpoints:
              - endpoint:
                  address:
                    socket_address:
                      address: gloo
                      port_value: 9977
          http2_protocol_options: {}
          type: STRICT_DNS
        - name: zipkin
          connect_timeout: 1s
          type: STRICT_DNS
          load_assignment:
            cluster_name: zipkin
            endpoints:
            - lb_endpoints:
              - endpoint:
                  address:
                    socket_address:
                      address: zipkin
                      port_value: 1234

To apply the bootstrap config to Envoy we need to restart the process. An easy way to do this is with kubectl rollout restart.

kubectl rollout restart deployment [deployment_name]

When the gateway-proxy pod restarts it should have the new trace cluster config.

2. Configure a tracing provider

For a list of supported tracing providers, and the configuration that they expect, please see Envoy’s documentation on trace provider configuration. For demonstration purposes, we show how to configure a zipkin trace provider below.

Option 1 (Preferred): Set the tracing provider on a dynamic listener:

You can enable tracing on a listener-by-listener basis. Please see the tracing listener docs for details on how to enable tracing on a listener.

Option 2: Set the tracing provider by editing the config map:

First, edit the config map pertaining to your proxy. This should be gateway-proxy-envoy-config in the gloo-system namespace.

kubectl edit configmap -n gloo-system gateway-proxy-envoy-config

Apply the tracing provider changes. A sample Zipkin configuration is shown below.

apiVersion: v1
kind: ConfigMap
data:
  envoy.yaml:
    node:
      cluster: gateway
      id: "{{.PodName}}{{.PodNamespace}}"
      metadata:
        role: "{{.PodNamespace}}~gateway-proxy"
    static_resources:
      listeners:
        - name: prometheus_listener
          address:
            socket_address:
              address: 0.0.0.0
              port_value: 8081
          filter_chains:
            - filters:
                - name: envoy.filters.network.http_connection_manager
                  typed_config:
                    "@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
                    codec_type: AUTO
                    stat_prefix: prometheus
                    route_config: # collapsed for brevity
                    http_filters:
                      - name: envoy.filters.http.router
                    tracing:
                      provider:
                        name: envoy.tracers.zipkin
                        typed_config:
                          "@type": "type.googleapis.com/envoy.config.trace.v2.ZipkinConfig"
                          collector_cluster: zipkin
                          collector_endpoint: "/api/v2/spans"
                          collector_endpoint_version: HTTP_JSON

To apply the bootstrap config to Envoy we need to restart the process. An easy way to do this is with kubectl rollout restart.

kubectl rollout restart deployment [deployment_name]

When the gateway-proxy pod restarts it should have the new trace provider config.

Note: This provider configuration will only be applied to the static listeners that are defined in the bootstrap config. If you need to support tracing on dynamically created listeners, follow the steps outlined in Option 1.

3. (Optional) Annotate routes with descriptors

In order to associate a trace with a route, it can be helpful to annotate your routes with a descriptive name. This can be applied to the route, via a route plugin, or provided through a header x-envoy-decorator-operation. If both means are used, the header’s value will override the routes’s value.

You can set a route descriptor with kubectl edit virtualservice -n gloo-system [name-of-vs]. Edit your virtual service as shown below.

apiVersion: gateway.solo.io/v1
kind: VirtualService
metadata: # collapsed for brevity
spec:
  virtualHost:
    domains:
    - '*'
    routes:
    - matchers:
      - exact: /abc
      routeAction:
        single:
          upstream:
            name: my-upstream
            namespace: gloo-system
      options:
        tracing:
          routeDescriptor: my-route-from-abc-jan-01
        prefixRewrite: /
status: # collapsed for brevity