Starknet validator dashboards and alerts

Grafana dashboards and alert rules for a Starknet validator, built on the metrics from eqlabs/starknet-validator-attestation and the pathfinder node behind it.

File	What it is
starknet-attestation-dashboard.json	Attestation dashboard. Seven rows, only the top one open.
starknet-node-dashboard.json	Pathfinder node health. Five rows, only Status open.
attestation-monitoring.yaml	22 alert rules, Grafana provisioning format.

Loading a dashboard

These are Grafana 13 schema v2 files. Neither carries a dashboard identifier, so the same file works whether you are creating a dashboard or replacing one.

To create one: + → New dashboard → open the toolbar on the right edge → the {} icon → select all the text in the editor and paste the file over it → Apply changes → Save. Grafana assigns the identifier itself.

To replace one you already have: open it → Edit → the same {} icon → select all → paste → Apply changes → Save. The dashboard keeps its identifier and its address.

The title comes from the file. Grafana refuses to create a second dashboard with a title that already exists in the same folder, so rename one of them if that happens.

You do not have to repoint anything afterwards. Neither dashboard hardcodes a data source: each has a Data source selector at the top and every panel reads it from there. Grafana picks a Prometheus for you on load; if you have more than one, switch it in that selector.

The Network selector

Three choices: Mainnet, Sepolia, Both, and Both is the default.

Each value is a regular expression, because the metrics name the network three different ways. The attestation tool uses the label exported_network with SN_MAIN and SN_SEPOLIA; pathfinder uses network with mainnet and testnet-sepolia; up and process_start_time_seconds carry no network label at all, and there the network is the scrape port inside instance, :9000 for mainnet and :9001 for testnet. One value covers all of them: Sepolia is SN_SEPOLIA|testnet-sepolia in the attestation dashboard and testnet-sepolia|.*9001 in the node dashboard.

To add a third network, edit the variable by hand — it is a fixed list, not a query.

Prometheus jobs

Two scrape jobs. The relabelling in the first one is what creates the exported_network label.

- job_name: "starknet-attestation"
  static_configs:
    - targets: ['localhost:9095', 'localhost:9096']
  relabel_configs:
    - source_labels: [__address__]
      regex: localhost:9095
      target_label: exported_network
      replacement: SN_MAIN
    - source_labels: [__address__]
      regex: localhost:9096
      target_label: exported_network
      replacement: SN_SEPOLIA

- job_name: "pathfinder"
  static_configs:
    - targets: ['localhost:9000', 'localhost:9001']

Run the attestation tool with --metrics-address 127.0.0.1:9095 for mainnet and :9096 for testnet, and pathfinder with --monitor-address 0.0.0.0:9000, mapped to host port 9000 for mainnet and 9001 for testnet. Reload Prometheus afterwards.

Loading the alert rules

Alerting → Import alert rules will refuse this file: that page takes Prometheus rule format, this is Grafana provisioning format. Two ways that work:

• put the file in /etc/grafana/provisioning/alerting/ and restart Grafana. The rules load but become read-only in the interface;
• or POST it to the provisioning API with the header X-Disable-Provenance: true. The rules stay editable.

The export leaves out two things, and without them the rules load but do nothing: the contact point, which the rules call by name (grafana-default-email), and the data source. Unlike the dashboards, the rules name Prometheus directly, so each one needs repointing if yours has a different identifier.

Worth knowing

Node dashboard: three panels use rpc_websocket_connections, rpc_websocket_connections_closed_total and rpc_websocket_connections_rejected_total, which exist only in pathfinder 0.24.0 and newer; on older versions they read No data. One of them, Websocket closes and rejects, last 24h, is also left out of the Network filter: those two counters appear only at the first close and the first refusal, so their labels cannot be counted on, and filtering could hide the rare event the panel is there for.

Attestation dashboard: Success rate, 30 days, Success rate, 90 days and Missed epochs, 90 days need that much history to mean what they say. Grafana Cloud on the free plan keeps 14 days — the two rates then stay correct as ratios but cover less time than their titles claim, and the missed-epoch count comes out short. Delete them there, or read them knowing that. Success rate, 14 days is in the row for exactly that case: on a store that keeps a fortnight it is the longest window that still means what it says.