This is a simple, lightweight reference implementation for quick, out-of-the-box observability for simple deployments on Fly.io.
Copy the fly.toml config into a new directory, run fly launch to create a new app, create a readonly access token, and deploy:
ORG=my-org-name
fly launch --copy-config -y --org $ORG -e ORG=$ORG --no-deploy
fly secrets set ACCESS_TOKEN="$(fly tokens create readonly $ORG)" --stage
fly deploy --flycast Once the deploy finishes, you can access the Grafana service to view your collected logs+metrics over your private network at http://$FLY_APP_NAME.flycast/.
- Subscribes to logs+metrics from the Fly.io-provided NATS platform streams on
[fdaa::3]:4223. - Writes logs to local VictoriaLogs and metrics to local VictoriaMetrics for storage.
- Runs a local Grafana instance with preconfigured data sources and dashboards for visualization and alerting.
The app runs in a single monolithic instance, which you can vertically scale for small-to-medium sized orgs. Once you grow out of this setup, you can fork and modify this template to further extend it yourself for clustered storage, or ship data directly to managed services that can offer greater scale and support.
The app doesn't configure any authentication, it's a simple template for internal use on a private Flycast network. Do not deploy this app with a public IP without setting up your own authentication!
Grafana is configured for anonymous admin access, which you can reach on its default port 3000 over 6pn,
or on port 80 through Flycast. Either connect to a Wireguard peer to your org's private network
and access $FLY_APP_NAME.flycast. $FLY_APP_NAME.internal:3000, or run fly proxy 3000 and access localhost:3000.
This app supports a simple High Availability mode by scaling out to multiple Machines, in the same or different region.
Each instance subscribes to the platform streams on a shared queue group which load-balances data across each active instance, and data is replicated to all other peers through Vector's native gRPC stream. Flycast load-balances service requests to the nearest available instance. For HA alerting, Grafana is configured to use the High Availability Alertmanager implementation that suppresses duplicate alert notifications across the cluster.
Each Grafana instance uses its own local SQLite database by default. You must change this config to use a shared external database cluster (like Postgres), or for a more minimal approach, use only file-provisioned dashboards, alert rules, etc., so all state is consistently maintained across both Grafana instances.
The replication topology is resolved at boot, so you will need to restart any running instances after making changes to the cluster configuration (like scaling up/down or replacing app instances).
Required env variables (provide with fly secrets set, or modify fly.toml):
ACCESS_TOKEN(fly secrets set ACCESS_TOKEN=$(fly tokens create readonly))ORG(defaultpersonal)