Skip to content

logging:metrics service

A module's numbers — CPU, memory, disk and network over time — in the site's metrics store (Prometheus on the logging VM), seen in Grafana's Prometheus data source and, as 5-minute means for 90 days, in Prometheus trends.

What a consumer gets

  • Proxmox nodes (pvehost): the cluster's external metric server (OpenTelemetry over HTTPS to logging.mgmt.internal:3120) carries each node's and each guest's CPU, memory, disk and network; the nodes' guest-memory sampler adds each guest's working set and pressure, the node's own pressure (PSI) and its ZFS pools' wait times. One setting for the whole cluster, with one credential for the cluster.
  • PBS (backup): its InfluxDB HTTP metric server, with the instance's own token.
  • Other modules: a push credential and an identity now; their own pipeline comes with the per-module performance levels.
  • Identity: every series carries host (from the credential) and the CI's tappaas_ci_id, service_instance_id, service_name, service_namespace and deployment_environment_name, set on the logging VM from the CMDB — never taken from the sender. A node's series about a guest carry the guest's identity.
  • The switch: site.json performance.level — off (the default on a small site) removes the metric servers and stops the pushes; basic (medium and large) collects. performance.retentionDays sets how long the store keeps them (7, 15 or 30 days by site size).
  • Network access: pinhole.json opens TCP 3120 from a consumer's VM to logging — push only. The query port 3121 stays mgmt-only and wants the admin credential.

Checking it

tappaas-metrics-target.sh all status