
logging-operator
This operator is in charge of handling the setup and configuration of the logging stack at Giant Swarm.
It reconciles cluster.cluster.x-k8s.io
objects and makes sure each Cluster
is provided with logging agents:
Getting started
Get the code and build it via:
git clone https://github.com/giantswarm/logging-operator.git
cd logging-operator
make
See make help
for help.
Architecture
The operator is built around a central reconciler, that calls multiple sub-reconcilers sequentially.

- Logging-Credentials are created if they don't exist. Then, their data (credentials) is used to create the next resources.
- grafana datasource configures Grafana to read data from Loki
- loki-multi-tenant-auth secures all Loki communication (per-tenant read/write access)
- logging-secret configures write access to Loki for Promtail
- logging-config setups some Promtail settings (like which logs to collect)
- logging-wiring ensures promtail-app reads configs from previous steps
- grafana-agent config setups some grafana-agent settings like the
river
configuration to scrape Kubernetes Events
- grafana-agent secret setups logging write credentials to access loki into the
river
configuration
- Logging agents toggle enables/disables logging agents deployment on WCs
Gathering logs from WCs
When the need to gather logs from the WCs appears, the logging-operator will deploy promtail on those so that one may see the logs from the MC's grafana. It deploys also grafana-agent to be able to scrape Kubernetes Events.
In order to achieve that, one has to label the cluster(s) one wants to gather logs from thanks to the following command :
kubectl label cluster -n <wc_namespace> <wc_name> giantswarm.io/logging=true
Credits
This operator was built using kubebuilder
.