https://github.com/iterait/shepherd
Provides access to computation resources on a single machine through a REST API
https://github.com/iterait/shepherd
gpu queue worker
Last synced: 5 months ago
JSON representation
Provides access to computation resources on a single machine through a REST API
- Host: GitHub
- URL: https://github.com/iterait/shepherd
- Owner: iterait
- License: other
- Created: 2018-09-13T07:59:58.000Z (almost 8 years ago)
- Default Branch: dev
- Last Pushed: 2022-01-06T09:11:27.000Z (over 4 years ago)
- Last Synced: 2025-05-08T05:42:38.903Z (about 1 year ago)
- Topics: gpu, queue, worker
- Language: Python
- Homepage: https://iterait.github.io/shepherd/
- Size: 2.23 MB
- Stars: 7
- Watchers: 8
- Forks: 2
- Open Issues: 29
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Shepherd
[](https://circleci.com/gh/iterait/shepherd/tree/master)
Provides access to computation resources on a single machine.
## Development Guide
### Prerequisites
1. Install dependencies with `pip install .`
2. Make sure you have Docker installed and that your user has permissions to use
it
3. If you intend to run computations on a GPU, install also `nvidia-docker2`
### Launching the Shepherd
First, you need to have a Docker registry and a Minio server running. The
easiest way to achieve this is to use the Docker Compose example:
```
docker-compose -f examples/docker/docker-compose-sandbox.yml up -d
```
Second, you need a configuration file. Again, examples found in the `examples/configs/`
folder are a great starting point. Feel free to pick one of those and edit it to
your needs.
Finally, you need to run the following command to start the shepherd:
```
shepherd -c examples/configs/shepherd-docker-cpu.yml
```
Be sure to adjust the command line parameters according to your needs (`-h` is
your host address, `-p` is the port number where the shepherd API server listens
and `-c` is the path to the configuration file).
After launching the shepherd, there will be an HTTP API available on the
configured port that can be used to control the shepherd.
### Processing a Request Directly
To process a request for debugging purposes, you need to:
- choose a request id
- create a bucket on your Minio server with a name same as your request id
- put the payload (input for the model) in `/inputs/input.json`
- invoke the `/start-job` API endpoint with your chosen request id
- after the job is processed, the result should be stored in Minio, in
`/outputs.json`
### Running Tests
First, install the test requirements `pip install '.[tests]'`.
The test suite can be run with `python setup.py test`.
### Running Stress Tests
First, install the test requirements `pip install '.[tests]'`.
To launch stress test, run:
```
docker-compose -f examples/docker/docker-compose-sandbox.yml up -d
shepherd -c tests/stress/shepherd-bare.yml
molotov tests/stress/loadtest.py -p 2 -w 10 -d 60 -xv
```
You can modify stress test arguments: `-p` (number of processes), `-w` (number of workers) and
`-d` (number of seconds to run the test).
You can also run stress test with time measurements:
```
molotov --use-extension tests/stress/measure_time.py --max-runs 10 tests/stress/loadtest.py
```