Local development

Prerequisites

Docker
ctlptl — creates and deletes the k3d cluster
Tilt — the dev loop
Helm 3+

Up and down

make up     # ctlptl apply + tilt up
make down   # tilt down + delete the k3d cluster

make up is idempotent: ctlptl apply creates the cluster only if it does not exist, and the target detects an already-running Tilt rather than starting a second one.

URL	What
`http://localhost:10350`	Tilt UI — per-resource logs and status
`http://localhost:3000`	Grafana (`admin` / `admin`)

What the dev stack changes

The Tiltfile installs the same chart with one override:

k8s_yaml(helm('chart', name='lgtm-pack', namespace='default',
              set=['nebariapp.enabled=false']))

nebariapp.enabled=false because there is no nebari-operator on a bare k3d cluster, so there is nothing to reconcile a NebariApp. Everything else — Grafana, Loki, Tempo, monolithic Mimir, Promtail, kube-state-metrics, node-exporter — runs exactly as it would in a cluster install.

Two guardrails worth knowing about:

allow_k8s_contexts('k3d-nebari-dev') means Tilt refuses to deploy anywhere else. If you renamed the cluster in ctlptl-config.yaml, update this too or Tilt will stop you.
update_settings(k8s_upsert_timeout_secs=600) raises the apply timeout, because the first run pulls a large set of upstream images.

The chart is deployed to the default namespace locally, not monitoring.

Working on the chart

Tilt watches chart/ and re-runs helm template on every save, so editing a template or a value re-applies within seconds. Watch the affected resource in the Tilt UI for the rollout.

The Mimir StatefulSet carries a checksum/config annotation over its ConfigMap, so config changes roll the pod automatically.

To render without deploying:

helm template lgtm-pack chart --set nebariapp.enabled=false | less

To check both Mimir modes render:

bash chart/tests/assert-rendering.sh

Verifying a change end to end

# Are logs flowing? Promtail should already be shipping container logs.
#   In Grafana: Explore -> Loki -> {namespace="default"}

# Is Mimir accepting writes?
kubectl exec -it sts/lgtm-pack-mimir -- wget -qO- localhost:8080/ready

# Are the dashboards provisioned?
kubectl get cm -l grafana_dashboard=1

For traces, send an OTLP payload to lgtm-pack-tempo:4318 from a pod, or port-forward the receiver and use telemetrygen:

kubectl port-forward svc/lgtm-pack-tempo 4317:4317
telemetrygen traces --otlp-insecure --otlp-endpoint localhost:4317 --traces 5

Testing distributed Mimir locally

It fits on a laptop — the chart’s mimir-distributed defaults are single replicas with replication factor 1 — but it is a lot more pods and PVCs than monolithic mode. Add the flag to the Tiltfile’s set list, or install by hand:

helm install lgtm-pack chart -n default \
  --set nebariapp.enabled=false \
  --set mimir-distributed.enabled=true

Docs site

The docs you are reading build from docs/:

make docs              # dev server with hot reload at http://localhost:4321
make docs-build        # static build into docs/dist/
make docs-preview      # serve the production build
make docs-test         # unit tests
make docs-check-links  # build, then verify every internal link resolves

Pages live in docs/src/content/docs/ — each .md or .mdx file becomes a page, and the sidebar is configured in docs/astro.config.mjs. Merges to main publish to packs.nebari.dev/lgtm-pack/; pull requests that touch docs/ get a preview URL posted as a comment.