Deploy and rebuild

infra/deploy.sh: copy, never install

Run on your machine from the repository root:

bash infra/deploy.sh            # uses the ssh Host alias "lake"
bash infra/deploy.sh myserver   # any other ssh host

It only copies files to the server; it installs nothing and never touches cluster.env (secrets live only on the server). It copies:

What	To
infra/scripts/*.sh	/opt/lake/scripts/
infra/lakestatus/lakestatus.py	/opt/lake/scripts/lakestatus/
apps/worker/{worker,process_csv,remove_file}.py	/opt/lake/scripts/worker/
infra/sql/*.sql	/opt/lake/sql/
infra/keycloak/realm-chdc.json	/opt/lake/keycloak/
infra/jupyterhub/*	/opt/lake/jupyterhub/

It first checks that those source files exist, and hands the target folders to the login user so they can be overwritten.

Build a server from nothing

1. An Ubuntu server, a login user with passwordless sudo, and an ssh alias for it in ~/.ssh/config.
2. On the server create /opt/lake/cluster.env from infra/cluster.env.example and replace every CHANGE_ME (openssl rand -hex 16). The versions and checksums in the example are the ones proven on staging.
3. bash infra/deploy.sh lake on your machine.
4. bash /opt/lake/scripts/install-all.sh on the server. Fix any failure and run again.
5. Check:

systemctl is-active seaweed postgresql postgrest nessie trino spark-master spark-worker nifi keycloak lake-worker jupyterhub
lakestatus

Honest status

Every configuration file a script writes was rendered with the staging cluster.env and compared with the live file: all identical. Steps 07.5, 14, 15, 18 and 19 were re-run on staging. A full run on an empty server has not been done yet, and that is the real proof. Do it on a spare machine before trusting it for production. Production also targets Ubuntu 24.04, while staging runs a newer release.

Apply a fix to one part

Change the script in the repository, then:

bash infra/deploy.sh lake                              # your machine
bash /opt/lake/scripts/install-all.sh 14               # the server: one or more step numbers

Never edit a config file on the server by hand. Put the change in the script, or it is lost at the next rebuild.

Move to several servers

Change only the four role hosts in cluster.env (INGEST_HOST, PROCESS_HOST, QUERY_HOST, STORAGE_HOST) and run only the steps for the services that belong on each server. See Architecture for the role map.