Gridworks Data

This project contains code related to working with databases in the GridWorks ecosystem.

Our platform is PostgreSQL, with the TimescaleDB (+toolkit) extensions for efficiently handling time-series data.

Not part of the production control path

This database, and everything that writes to it (the journal keeper, the web app's read paths), is an observability layer: a downstream recorder of what the production system emits. The production control system — the SCADA in each house, its LTN, the MarketMaker — never reads this database to make a control decision, and keeps running unaffected when it is down, slow, or in the middle of a migration. Each production node owns its own state locally; heavy analysis over a node's data is sent out from the node to a stateless service, never pulled from here.

Prerequisites

• Docker Engine — to run PostgreSQL+TimescaleDB locally. (If you have previously set up the RabbitMQ server from gridworks-base, you should already have Docker.)
• psql PostgreSQL command-line client (any modern version; tested with 14.x and 18.x). Install via your package manager (e.g. brew install libpq && brew link --force libpq on macOS, sudo apt install postgresql-client on Linux). Check with psql --version. (Optional: pgAdmin as a GUI for inspecting the database.)
• Python ≥ 3.12 and uv (per the rest of the GridWorks toolchain).

Database Setup

The following steps will get you set up to run with the database, either locally or on a managed Tiger Cloud instance.

Local Pre-requisite: PostgreSQL+TimescaleDB container

If running locally, you'll need to start by pulling the official TimescaleDB image and starting a container. You need the -ha variant for PostgreSQL v18 and TimescaleDB v2.25 — without -ha you'll get the "lite" version of TimescaleDB which is missing required functionality.

If you don't already have a PostgreSQL listener on port 5432, map 5432:5432. If you do (native install, or another Docker postgres), pick a free port and map <HOST_PORT>:5432 instead (e.g. 5433:5432).

Pick a POSTGRES_PASSWORD — this becomes the password for the built-in postgres superuser; record it locally (e.g. in .env), don't commit it.

Concrete docker run example (uses 5433 and a placeholder password):

docker run -d \
  --name gw-data-pg \
  -e POSTGRES_PASSWORD=changeme \
  -p 5433:5432 \
  timescale/timescaledb-ha:pg18-ts2.25

Verify the container is healthy: docker ps should show it up, and PGPASSWORD=changeme psql -h 127.0.0.1 -p 5433 -U postgres -c "SELECT version();" should return PostgreSQL 18.x.

Detailed instructions and alternative configurations: https://www.tigerdata.com/docs/self-hosted/latest/install/installation-docker

Database Setup

This repo includes a numbered sequence of scripts in the src/gw_data/db/scripts folder that will get you up and running.

0. (Local-Only) Create the Database

This step is not required (or even possible) on a Tiger Cloud managed server.

Run 0_db_create.sql as the postgres user to create the tsdb database and add the timescaledb extension.

psql -d "postgresql://127.0.0.1:<PORT>/" -U postgres -f src/gw_data/db/scripts/0_server_init.psql

• Replace <PORT> with the host port you mapped to 5432 (e.g. 5433).
• You'll be prompted for the postgres user password (i.e., the one you previously set as POSTGRES_PASSWORD in your docker run command).

1. Create the Users

Run 1_db_user_setup.psql as follows to create the users we need. This script can be run as the postgres user locally, or the tsdbadmin user in Tiger Cloud.

psql -d "postgresql://<SERVER>:<PORT>/" -U <postgres|tsdbadmin> -f src/gw_data/db/scripts/1_db_user_setup.psql

• Replace <SERVER> with the database server address (e.g., 127.0.0.1 when running locally).
• Again, replace <PORT> with your mapped host port.
• Again, you'll be prompted for the postgres user password.

This script creates five user roles, named by consumer: gw_admin (full ownership), gw_journalkeeper (insert/update/delete), and three select-only roles — gw_visualizer (the web API), gw_alerts (gwalert), gw_analyst (people and ad-hoc analysis) — with role-level statement timeouts (10 s for the two services, 2 min for the readers; long jobs use SET LOCAL statement_timeout per transaction).

Note: the script uses psql's interactive \password meta-command five times (once for each user). It must be run interactively — it will prompt for each password as it runs and cannot be piped or run via CI:

You'll be prompted for new passwords for each of the users. Pick whatever you like and record them somewhere.

Local dev convention: password = role name. For the local container, run 1_db_user_setup_dev.psql instead — it creates the same five roles non-interactively, each with password = role name (gw_admin/gw_admin, …):

psql -d "postgresql://127.0.0.1:<PORT>/" -U postgres -f src/gw_data/db/scripts/1_db_user_setup_dev.psql

Consumers of this database (journalkeeper, gwalert, …) ship default db_urls that assume this convention, so a fresh checkout runs against the local container with no .env. Never use it on a shared or production server; the interactive script above is the production path.

Reset shortcut: if your local DB ever gets into a weird state, you can wipe it and start over with:

psql -d "postgresql://127.0.0.1:<PORT>/" -U postgres -f src/gw_data/db/scripts/_XX_drop_all.sql

2. Create the gridworks Schema

Run 2_db_schema_setup.sql as follows to create our private gridworks schema and apply appropriate permissions. This should be run as the gw_admin user.

psql -d "postgresql://<SERVER>:<PORT>/" -U gw_admin -f src/gw_data/db/scripts/2_db_schema_setup.psql

• Again, replace <SERVER> with the database server address.
• Again, replace <PORT> with your mapped host port.
• This time you'll be prompted for the gw_admin user password.

3. Run the Alembic Migrations

Next we need to run the database migrations we've defined with Alembic to create the tables, etc. that we need.

But first we need to update our .env file. Copy template.env (at the repo root) to .env:

cp template.env .env

Then, edit .env as follows:

• Replace <SERVER> with the database server address (e.g., 127.0.0.1 when running locally).
• Replace <%PASSWORD%> in GW_DATA_DB_URL with the gw_admin password you just set.
• Replace <%PORT%> with the host port you mapped to 5432 (e.g. 5433).

Now we can run 3_db_alembic_upgrade.sh as follows:

`sh src/gw_data/db/scripts/3_db_alembic_upgrade.sh`

This should create 12 tables in the gridworks schema: alembic_version, connectivity_edges, customers, g_nodes, installations, installers, messages, position_points, reading_channels, readings, user_installation_roles, users.

Verify with psql -d "postgresql://<SERVER>:<PORT>/tsdb" -U gw_admin -c "\dt gridworks.*".

4. Seed the Database

With the database created, gw_admin ready, and .env filled in, you can seed some initial data:

uv run python ./src/gw_data/db/scripts/1_db_seed.py

Note: the seed script is also interactive — it uses Python's getpass to prompt for passwords for two seeded users (admin and beech-user). It must be run from a real terminal (no piping). The seed populates a small set of dev users, a customer, an installation, and one g_node.

In the future we will have a more comprehensive seeding process that will ingest some actual message data.

Best Practices for Databases

The following are some best practices that we should follow when at all possible:

• Primary Keys should be UUIDs, stored as the DB-native UUID type
• Dates/Times should be stored as TIMESTAMPTZ
• Migrations should be encouraged and done frequently to provide new functionality. Database schema is always temporary.
• Each application that uses that database should have its own dedicated user, with the minimal set of permissions. (In particular, postgres should never be used at all, and gw_admin should only be used for migrations and other tasks that require full ownership of the database.)

TimescaleDB Performance

TimescaleDB does two main things to improve performance:

1. Separates time-series tables (which it calls "hypertables") into "chunks", each of which cover a certain timeframe.
2. Allows time-series data to be stored in column order with compression, which vastly improves performance for data that changes very little over time. This compression happens in a scheduled job, and only for data older than a configured threshold.

We have column-store compression enabled on the readings table for data older than 2 weeks, with compression segmented by the channel ID.

Useful Queries

There are some useful queries listed in useful-queries.sql in the project root.