Docker Swarm: Upgrade & Uninstall

The installer is fully idempotent: secrets are preserved across runs, configs are bind-mounted from /etc/elchi and carry a per-service elchi.cfghash label, and docker stack deploy rolling-updates every changed service. So an upgrade is just install.sh re-run with new version flags, and upgrade.sh is a thin wrapper that re-execs install.sh with whatever arguments you pass.

Upgrade (version bump)

Re-run with the new --*-version flags:

deploy/docker/upgrade.sh \
  --main-address=elchi.example.com \
  --ui-version=v1.5.5 \
  --backend-version=v1.6.6-v0.14.0-envoy1.38.3

Under the hood this re-renders the configs and stack file. Each service carries an elchi.cfghash container label over its mounted files, so a changed image tag or config file changes the label, and docker stack deploy rolling-updates exactly the affected services — nothing else is bounced.

The bootstrap one-liner also supports upgrades directly:

curl -fsSL https://raw.githubusercontent.com/CloudNativeWorks/elchi-archive/main/deploy/docker/get.sh \
  | sudo bash -s -- --upgrade --main-address=elchi.example.com --ui-version=v1.5.5

Editing live configs

The bind-mounted files under /etc/elchi are the live source of truth. Edit a file, then force the service to pick it up:

sudo vi /etc/elchi/config/Corefile          # or envoy.yaml, config-prod-*.yaml, …
docker service update --force elchi_elchi-coredns

Because a bind-mount content change is invisible to Swarm on its own, a manual edit needs the --force. Re-running install.sh instead auto-applies changes via the elchi.cfghash label.

Multi-node edits

On a multi-node cluster the config tree exists on every node. For a manual live edit, edit the file on every node that runs the service (or re-run the installer, which SSH-copies /etc/elchi to every node). One exception: collector.env is an env_file: read at deploy, so editing it needs a redeploy / --force.

Uninstall

uninstall.sh removes the stack. By default it keeps your data volumes:

deploy/docker/uninstall.sh                    # remove the stack, keep data volumes
deploy/docker/uninstall.sh --purge-data       # also delete data volumes (all nodes)
deploy/docker/uninstall.sh --purge            # + configs, secrets, state, host tuning
deploy/docker/uninstall.sh --purge --leave-swarm   # + every node leaves the Swarm

Flag	Effect
(none)	docker stack rm — the stack is removed, volumes preserved.
--purge-data	Also delete the elchi_* data volumes on every node.
--purge	Everything --purge-data does, plus /etc/elchi, the state dir, legacy elchi_* config/secret objects, and the host tuning (/etc/sysctl.d/99-elchi.conf, elchi-thp.service). Leaves /etc/docker/daemon.json in place.
--leave-swarm	Every node runs docker swarm leave --force to dissolve the Swarm.

Multi-node teardown

Pass the same --nodes you installed with so M1 can SSH into the workers to remove their node-local volumes and make them leave the Swarm — a manager-only docker volume rm can't reach worker volumes. Teardown reuses the bootstrap SSH key (~/.ssh/elchi_cluster) by default.