Step 2: Deploy Vector

Vector runs as a DaemonSet, tails every pod’s logs on its node, filters down to your CNPG instance pods, and reshapes CNPG’s JSON log envelope into the OTLP log record format using a VRL transform maintained in the pganalyze collector repository.

This page covers the quickest path to get it running. For a full breakdown of what each part of the Vector configuration does (the Kubernetes log source, the CNPG pod filter, the JSON remap, and the OTLP sink), see the Vector configuration reference.

Fetch the example values.yaml

Saved as vector-values.yaml rather than the generic values.yaml, so it doesn’t collide with collector-values.yaml from Initial Setup if you keep both in the same directory:

wget -O vector-values.yaml https://github.com/pganalyze/collector/raw/refs/heads/main/contrib/vector/values.yaml

This file already implements the full pipeline: a kubernetes_logs source, a cnpg_route filter (matching pods with the cnpg.io/podRole: instance label), the cnpg_remap JSON transform, and an opentelemetry sink.

Customize it for your cluster

Three things need to change before this will work for you:

1. The cluster filter. In the cnpg_route transform, update the list to your Cluster resource’s name, e.g. my-cluster, not a per-instance pod name like my-cluster-1:

   route:
     cnpg: |
       exists(.kubernetes.pod_labels."cnpg.io/cluster") &&
       exists(.kubernetes.pod_labels."cnpg.io/podRole") &&
       .kubernetes.pod_labels."cnpg.io/podRole" == "instance" &&
       includes(
         [
           "my-cluster",
         ],
         .kubernetes.pod_labels."cnpg.io/cluster"
       )
   

2. The sink URI. Point it at the pganalyze-collector-otel-service Service you created in Step 1, using its namespace-qualified form. Vector runs in its own vector namespace (per the helm install command below), while the collector’s Service is wherever you deployed the collector, and a bare Service name only resolves via Kubernetes DNS within the same namespace as the caller. Replace <collector-namespace> with the actual namespace (default if you didn’t set one):

   sinks:
     pganalyze:
       type: opentelemetry
       protocol:
         type: http
         uri: "http://pganalyze-collector-otel-service.<collector-namespace>.svc.cluster.local:4318/v1/logs"
         # ...
   

3. The sink’s batch.max_events, which must be 1. The downloaded file sets it higher, but Vector’s opentelemetry sink does not currently support batching more than one event per request with the json codec, anything higher silently drops logs. See the Vector configuration reference for details:

   sinks:
     pganalyze:
       protocol:
         batch:
           max_events: 1
           # ...
   

Deploy

Deploy Vector with its Helm chart, into its own vector namespace:

helm repo add vector https://helm.vector.dev
helm repo update
helm install vector vector/vector \
  --namespace vector \
  --create-namespace \
  --values vector-values.yaml

Check that the Vector pods came up on every node:

$ kubectl get pods -n vector
NAME       READY   STATUS    RESTARTS   AGE
vector-0   1/1     Running   0          1m