Prefect Server

Please note: this repo is for Prefect Server development. If you want to run Prefect Server, the best first step is to install Prefect and run prefect server start.

If you want to install Prefect Server on Kubernetes, take a look at the Server Helm Chart.

If you would like to work on the Prefect UI or open a UI-specific issue, please visit the Prefect UI repository.

Overview

Prefect Server is an open source backend that makes it easy to monitor and execute your Prefect flows.

Prefect Server consists of a number of related services including:

postgres: the database persistence layer
hasura: a GraphQL API for Postgres (http://hasura.io)
graphql: a Python-based GraphQL server that exposes mutations (actions) representing Prefect Server's logic
apollo: an Apollo Server that serves as the main user interaction endpoint, and stitches together the hasura and graphql APIs
towel: a variety of utility services that provide maintenance routines, because a towel is just about the most massively useful thing an interstellar hitchhiker can carry
- scheduler: a service that searches for flows that need scheduling and creates new flow runs
- lazarus: a service that detects when flow runs ended abnormally and should be restarted
- zombie_killer: a service that detects when task runs ended abnormally and should be failed

These services are intended to be run within Docker and some CLI commands require docker-compose which helps orchestrate running multiple Docker containers simultaneously.

Installation

Don't Panic.
Make sure you have Python 3.7+ and Prefect installed:
```
pip install prefect
```
Clone this repo, then install Prefect Server and its dependencies by running:
```
pip install -e .
cd services/apollo && npm install
```

Note: if installing for local development, it is important to install using the -e flag with [dev] extras: pip install -e ".[dev]"

Running the system as a developer

Note: for deploying Prefect Server, please use the prefect server start CLI command in Prefect Core 0.13.0+.

If you are doing local development on Prefect Server, it is best to run most services as local processes. This allows for hot-reloading as code changes, setting debugging breakpoints, and generally speeds up the pace of iteration.

In order to run the system:

Start the database and Hasura in Docker:
```
prefect-server dev infrastructure
```
If when starting the infrastructure, you receive an error message stating infrastructure_hasura_1 exited with code 137, it is likely a memory issue with Docker. Bumping Docker Memory to 8GB should solve this.
Run the database migrations and apply Hasura metadata:
```
prefect-server database upgrade
```
In a new terminal, start the services locally:
```
prefect-server dev services
```

You can use the -i (include) or -e (exclude) flags to choose specific services:

# run only apollo and graphql
prefect-server dev services -i apollo,graphql

# run all except graphql
prefect-server dev services -e graphql

Running tests

Prefect Server has three types of tests:

unit tests: used to validate individual functions
service tests: used to verify functionality throughout Prefect Server
integration tests: used to verify functionality between Prefect Core and Server

Prefect Server uses pytest for testing. Tests are organized in a way that generally mimics the src directory. For example, in order to run all unit tests for the API and the GraphQL server, run:

pytest tests/api tests/graphql

Unit tests can be run with only prefect-server dev infrastructure running. Service and integration tests require Prefect Server's services to be running as well.

Filing an issue

Whether you'd like a feature or you're seeing a bug, we welcome users filing issues. Helpful bug issues include:

the circumstances surrounding the bug
the desired behavior
a minimum reproducible example

Helpful feature requests include:

a description of the feature
how the feature could be helpful
if applicable, initial thoughts about feature implementation

Please be aware that Prefect Server feature requests that might compete with propriety Prefect Cloud features will be rejected.

License

Prefect Server is lovingly made by the team at Prefect and licensed under the Prefect Community License. For information on how you can use, extend, and depend on Prefect Server to automate your data, take a look at our license or contact us.