Eliaaazzz commented on issue #19137:
URL: https://github.com/apache/beam/issues/19137#issuecomment-4055866337

   For additional context, I’ve been exploring this in parallel as part of my 
GSoC preparation, and my main takeaway so far is that this problem extends 
beyond foundational API definitions.
   
   In a separate public PoC (https://github.com/Eliaaazzz/beam/tree/gsoc/poc), 
I’ve been validating an end-to-end SDF-style UnboundedSource path, including 
restriction/tracker/coder handling, checkpoint persistence and finalization, 
watermark restore/progression, deferred resumption, and a standalone reader 
transform. The UnboundedSource PoC currently has 49 passing tests.
   
   What became clear from implementing it is that the core difficulty here is 
not the class surface itself, but the execution contract and runtime semantics: 
checkpoint-based splitting, primary/residual representation, watermark 
behavior, idle resume behavior, Read(UnboundedSource) integration, and 
alignment with Beam’s existing Java/Python model all need to be pinned down 
carefully.
   
   I’m currently developing this as the main direction of my GSoC proposal, so 
I wanted to share the PoC here because I think those end-to-end 
semantic/runtime requirements are central to the overall design.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to