Grafana Dashboard

Required: a cluster label

The dashboard groups every panel by a cluster label that the exporter does not emit. Prometheus attaches it, one per exporter instance:

scrape_configs:
  - job_name: openldap
    scrape_interval: 30s
    static_configs:
      - targets: ['openldap-exporter:9330']
        labels: { cluster: prod }
      - targets: ['openldap-exporter-alt:9331']
        labels: { cluster: staging }

Without it the cluster, group and target variables come up empty and most panels show no data.

Import

Download dashboards/openldap.json, then in Grafana Dashboards → New → Import. Grafana asks which Prometheus to use.

Variables

Reading a replication problem

Each step isolates the cause, so work down in order:

1. 1 Nodes up — is anything answering.
2. 2 Links healthy — did the providers observe their consumers connect.
3. 3 In sync — do the contextCSN values agree.
4. 4 Replication lag by sid — how far behind, per sid.
5. 5 Consumer links seen by providers — which node's link is dead while its own socket still looks healthy.
6. 6 Records / Change sets / Last change — did the data actually arrive.

Last change is the newest contextCSN timestamp on a node. Identical values across a cluster mean the nodes converged on the same newest change; a node that has fallen behind shows an older timestamp. It is the strongest convergence signal on the page, because it comes from the data rather than from replication's own bookkeeping.

Empty when healthy

Empty is the good outcome, not a broken panel.

One gotcha

Two providers that bootstrapped independently each hold a contextCSN for their own sid only. Until each accepts one change from the other's sid their CSN sets differ, and the sync check reports the pair as not converged — although both are healthy. One write against each node closes it.

See also: Prometheus Exporter.