Skip to content


For a production deployment, investigraph needs some different running services based on the deployment needs.

The use case for deploying investigraph as outlined here is a decentralized, multi-node scenario where workflows are triggered and scheduled via the server ui instead of the command line (which is often used within the investigraph documentation).

See docker deployment for documentation on how to deploy the services outlined here within a docker environment.

Prefect server

The prefect server manages workflow runs and state. It as well serves a user interface for configuration, adding blocks and scheduling workflows. After installing investigraph, you can spin up the server locally via

prefect server start

The server dashboard (prefect ui) is now running at http://localhost:4200

The investigraph docker image uses the server as the default entrypoint, so running the server as a docker container is as easy as:

docker run

More about server via documentation

Prefect agent

investigraph needs one or several agents that will execute the workflows orchestrated by the server.

prefect agent start -q "default"

where -q defines the queue to use (see documentation)

These agents don't have access to local data located in the server. That's why in a distributed deployment investigraph is using blocks to store dataset configuration.

If you still want to access local data in your environment, make sure that the agent has access to the directories, or, in a docker deployment, ensure the correct volume mounts.

The agent needs to know which server api to use which is controlled via the env var PREFECT_API_URL which defaults to

PREFECT_API_URL=http://my.prefect.server/api prefect agent start -q "other-queue"

More about agents via documentation


Per default, prefect server uses a sqlite database located in the directory specified by PREFECT_HOME. This database stores configuration, deployments, flow runs, task states. For more scalable production setups a PostgreSQL database is recomended.

Important sql adapters need to be asynchronous, like asyncpg for postgres.


Runtime cache (Redis)

Processed data is shared between tasks (and their agent workers) via a redis cache.

In local or developement mode (set via DEBUG=1), investigraph just uses fakeredis, but for scalable deployments you should use a real redis instance.

investigraph only needs the cache during runtime and mimics GETDEL for all fetching operations, so the cache doesn't have to be persistent and will be empty again after a sucessful flow run.

Specify the cache endpoint via env var REDIS_URL, e.g.: