Skip to content

Latest commit

 

History

History
46 lines (27 loc) · 1.66 KB

MAAP-DPS.md

File metadata and controls

46 lines (27 loc) · 1.66 KB

Async Task Runner

Overview:

Currently the EIS team uses the MAAP-DPS async executor (talked about below) to run their tasks. However, for reasons related to portability, data volume and simplistic idioms we've decided to leverage Dask for parallelism instead of DPS. What that means in practice is that you can run this algorithm on any cloud compute instance (EC2, GCP) or cluster (k8s, JupyterHub) taking advantage of the number of cores as long as it has Dask installed (which is a requirement in pyproject.toml). In the future we might tweaks things slightly so you can run against remote Dask clusters as well.

MAAP DPS

DPS Docs:

https://docs.maap-project.org/en/latest/technical_tutorials/dps_tutorial/dps_tutorial_demo.html#Data-Processing-System-(DPS)-Tutorial-A-to-Z

Registering the Algorithm for DPS Runs:

  1. Make any necessary code changes on your feature branch or main branch

  2. Decide on a tag name that is unique (it can be a semvar version or just text). For purposes of this documentation we'll call the new tag testing123

  3. Change the algorithm_version: testing123 line in /maap_runtime/<algo-name>/algorithm_config.yaml files to point to this new tag

  4. Git commit all your changes and push remotely

git add -u
git commit -m "updates with new tag"
git push origin <branch-name>
  1. Then tag and push your tags remotely
git tag -f testing123
git push origin -f testing123
  1. Then register your algorithm. Really the output of this step is just building a new DPS image. Either you can manually do this as talked about in the DPS docs above or automate it like in /maap_runtim/register-all.ipynb