je-ik commented on issue #27635:
URL: https://github.com/apache/beam/issues/27635#issuecomment-1687765270

   > Since Beam doesn't support "hot" upgrade on the fly while running a 
pipeline, and Kryo extension is used only as a `KryoCoder` (correct me if I'm 
wrong) then serialised data should not be stored somewhere after pipeline is 
stopped. Also, I hope we don't expose any Kryo API to users as well.
   
   @aromanenko-dev Is this an official policy? Upgrading pipeline in general 
can be tricky, but I believe we should do our best to enable users to upgrade 
their pipelines, if possible. This change forces users to do a bootstrap of 
streaming pipelines which might be expensive and might discourage users from 
upgrading.
   
   I think we could have tried introducing KryoV5Coder and deprecate original 
KryoCoder for at least several releases, we might have even kept old coder 
until we drop support for JDK 11 (Kryo 4 does not support JDK 17). Also some 
more context on this: 
https://lists.apache.org/thread/5glkdp68b024t8pnwq7ncz5mt9gcr27r
   
   I came across this when validating 2.50.0 RC (this is even a code breaking 
change, beacuse the KryoCoder 'leaks' kryo API). Wondering if we should try to 
fix it before the release.


-- 
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