Move task execution architecture docs to contributing-docs - #70657
Move task execution architecture docs to contributing-docs#70657jason810496 wants to merge 2 commits into
Conversation
uranusjr
left a comment
There was a problem hiding this comment.
I don’t think we want to put coordinator first. I intentionally put the concept to the back because the user does not need to care how it works, only how to use it. The new writeup emphasis on internals too much. It should instead focus on introducing how the coordinator can be used (by configuration) instead.
Hmmm, then I will need to move the whole paragraph introduce in #69750 to somewhere else. All the diagrams and descriptions are quite verbose for the users IMO -- http://apache-airflow-docs.s3-website.eu-central-1.amazonaws.com/docs/apache-airflow/stable/core-concepts/overview.html#non-python-language-sdks-go-and-java |
|
Then would it be better to move them all (starting from "Task execution architecture" section) to |
|
I think they do not belong in user-facing documentation. It would fit if we have a section for advanced concepts or something, but we currently don’t. They do fit in dev-facing docs though ( |
The "Task execution architecture" section added in apache#69750 documents worker internals (the Supervisor, the Coordinator layer, and the language subprocess protocols). That is developer-facing material, not something end users need in the core-concepts overview. Relocate the whole section and its diagrams to a new dev-facing page under contributing-docs, next to the new-language-SDK guide, and drop it from the user-facing overview. Follow-up to review feedback on apache#70657.
8271810 to
bf8246c
Compare
|
I just moved them all to |
|
To be honest, I do not think those are "contributing" docs. Those are "deployment" docs that explain what is the interaction between different components, what processes are running and what to expect. One of the big issues it explains is what is teh resource usage connected and what processes you can expect to see running when you use the SDK. I think the deployment managers should know that in some cases they can just run a long running process that does everything (golang) , and it some other cases they will have python processes taking 100s of MBs for each task run - this is far more deployment information than contributing documentation IMHO, |
potiuk
left a comment
There was a problem hiding this comment.
I think we should first agree who is the audience of that docs.
IMHO. Deployment Managers - not contributors.
3c302ad to
8251d6e
Compare
The overview explained how components are deployed but not what happens inside a worker when a task actually runs. This adds a concise section and diagram distinguishing the supervisor process, the CoordinatorManager that routes a task to a coordinator by its queue, and the language SDK subprocess launched for non-Python tasks, so readers can reason about the process boundaries. The full developer-level detail remains in contributing-docs.
8251d6e to
3f60e8f
Compare
Why
The "Task execution architecture" section added in #69750 documents worker internals (the Supervisor, the Coordinator layer, and the language subprocess protocols). That is developer-facing material, not something end users need in the
core-concepts/overview.rstpage, so it belongs in the contributing docs instead.How
contributing-docs/31_task_execution_architecture.rst, next to the new-language-SDK guide.contributing-docs/images/and extend thegenerate-airflow-diagramsprek hook to cover that directory. Rewrite the cross-links for GitHub rendering (relative.rstlinks,.. image:: images/...).core-concepts/overview.rst; register the new page in the contributing-docs README and cross-link it from30_new_language_sdk.rst.Was generative AI tooling used to co-author this PR?