Repository navigation
Fixes the upgrade from v2.2.0 over an existing Grafana volume, which left three dashboards unapplied (listed as known in v2.3.1).
Fixed
- Upgrading from v2.2.0 over an existing Grafana volume now applies the
nginx,postgresqlandredisdashboards. Their files kept the same path but got new uids in v2.3.0. Grafana 13.2.3 does not save a file in that case (failed to save dashboardin its log) whiledocker compose up -dexits 0. The dashboards are now provisioned from/etc/grafana/provisioning/dashboards/monitoring-stack. - The Ansible playbook restarts Grafana when the dashboard provider file changes. Grafana reads providers only at start, so without a restart an upgraded server kept the old provider and the old dashboards. The old
grafana/provisioning/dashboards/jsonfolder on the server is no longer used; the playbook leaves it, delete it if you like.
Changed
disableDeletionis nowfalse: Grafana deletes provisioned dashboards whose file is gone, so old ones no longer pile up after an upgrade. Dashboards made in the UI are not touched. While the dashboards folder is empty or not mounted, Grafana removes the provisioned ones; they return when the files are back.- CI: the upgrade check starts from v2.3.1, v2.3.0 and v2.2.0 on Docker Compose, and a new job runs the same upgrades with the Ansible playbook. Both fail on any dashboard provisioning error in the Grafana log.
Upgrade notes
- If you already upgraded to v2.3.0 or v2.3.1: this upgrade applies the three dashboards and the errors stop. Four old dashboards (Node Exporter Full, Prometheus 2.0 Overview, Prometheus Blackbox Exporter, RabbitMQ-Overview) stay in Grafana; delete them in the UI. Details in
docs/TROUBLESHOOTING.md. - With Ansible: check out the tag and run the playbook again.
Still open from v2.3.1
- The Prometheus healthcheck runs inside the container and cannot detect a port taken by another process. The Ansible playbook does not check targets after it starts.
Full list: CHANGELOG.md.