A runtime system for NMDC data management and orchestration.
http://nmdcstatus.polyneme.xyz/
-
issues
tracks issues related to NMDC, which may necessitate work across multiple repos. -
nmdc-schema houses the LinkML schema specification, as well as generated artifacts (e.g. JSON Schema).
-
nmdc-server houses code specific to the data portal -- its database, back-end API, and front-end application.
-
workflow_documentation references workflow code spread across several repositories, that take source data and produce computed data.
-
This repo (nmdc-runtime)
- houses code that takes source data and computed data, and transforms it to broadly accommodate downstream applications such as the data portal
- manages execution of the above (i.e., lightweight data transformations) and also of computationally- and data-intensive workflows performed at other sites, ensuring that claimed jobs have access to needed configuration and data resources.
The NMDC metadata as of 2021-10 is available here:
https://drs.microbiomedata.org/ga4gh/drs/v1/objects/sys086d541
The link returns a GA4GH DRS API bundle object record, with the NMDC metadata collections (study_set, biosample_set, etc.) as contents, each a DRS API blob object.
For example the blob for the study_set collection export, named "study_set.jsonl.gz", is listed with DRS API ID "sys0xsry70". Thus, it is retrievable via
https://drs.microbiomedata.org/ga4gh/drs/v1/objects/sys0xsry70
The returned blob object record lists https://nmdc-runtime.files.polyneme.xyz/nmdcdb-mongoexport/2021-10-14/study_set.jsonl.gz as the url for an access method.
The 2021-10 exports are currently all accessible at https://nmdc-runtime.files.polyneme.xyz/nmdcdb-mongoexport/2021-10-14/${COLLECTION_NAME}.jsonl.gz, but the DRS API indirection allows these links to change in the future, for mirroring via other URLs, etc. So, the DRS API links should be the links you share.
The runtime features:
-
Dagster orchestration:
- dagit - a web UI to monitor and manage the running system.
- dagster-daemon - a service that triggers pipeline runs based on time or external state.
- PostgresSQL database - for storing run history, event logs, and scheduler state.
- workspace code
- Code to run is loaded into a Dagster
workspace. This code is loaded from one or more dagsterrepositories. Each Dagsterrepositorymay be run with a different Python virtual environment if need be, and may be loaded from a local Python file orpip installed from an external source. In our case, each Dagsterrepositoryis simply loaded from a Python file local to the nmdc-runtime GitHub repository, and all code is run in the same Python environment. - A Dagster repository consists of
solidsandpipelines, and optionallyschedulesandsensors.solidsrepresent individual units of computationpipelinesare built up from solidsschedulestrigger recurring pipeline runs based on timesensorstrigger pipeline runs based on external state
- Each
pipelinecan declare dependencies on any runtimeresourcesor additional configuration. There are TerminusDB and MongoDBresourcesdefined, as well aspresetconfiguration definitions for both "dev" and "prod"modes. Thepresets tell Dagster to look to a set of known environment variables to load resources configurations, depending on themode.
- Code to run is loaded into a Dagster
-
A TerminusDB database supporting revision control of schema-validated data.
-
A MongoDB database supporting write-once, high-throughput internal data storage by the nmdc-runtime FastAPI instance.
-
A FastAPI service to interface with the orchestrator and database, as a hub for data management and workflow automation.
Ensure Docker (and Docker Compose) are installed; and the Docker engine is running.
docker --version
docker compose version
docker infoEnsure the permissions of ./mongoKeyFile are such that only the file's owner can read or write the file.
chmod 600 ./mongoKeyFileEnsure you have a .env file for the Docker services to source from. You may copy .env.example to
.env (which is gitignore'd) to get started.
cp .env.example .envCreate environment variables in your shell session, based upon the contents of the .env file.
export $(grep -v '^#' .env | xargs)If you are connecting to resources that require an SSH tunnel—for example, a MongoDB server that is only accessible on the NERSC network—set up the SSH tunnel.
The following command could be useful to you, either directly or as a template (see Makefile).
make nersc-mongo-tunnelsFinally, spin up the Docker Compose stack.
make up-devDocker Compose is used to start local MongoDB and PostgresSQL (used by Dagster) instances, as well as a Dagster web server (dagit) and daemon (dagster-daemon).
The Dagit web server is viewable at http://127.0.0.1:3000/.
The FastAPI service is viewable at http://127.0.0.1:8000/ -- e.g., rendered documentation at http://127.0.0.1:8000/redoc/.
Tests can be found in tests and are run with the following commands:
On an M1 Mac? May need to export DOCKER_DEFAULT_PLATFORM=linux/amd64.
make up-test
make testAs you create Dagster solids and pipelines, add tests in tests/ to check that your code behaves as
desired and does not break over time.
This repository contains a GitHub Actions workflow that publishes a Python package to PyPI.
You can also manually publish the Python package to PyPI by issuing the following commands in the root directory of the repository:
rm -rf dist
python -m build
twine upload dist/*
Here are links related to this repository:
- Production API server: https://api.microbiomedata.org
- PyPI package: https://pypi.org/project/nmdc-runtime
- DockerHub image (API server): https://hub.docker.com/r/microbiomedata/nmdc-runtime-fastapi
- DockerHub image (Dagster): https://hub.docker.com/r/microbiomedata/nmdc-runtime-dagster
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent