t3hw commented on PR #14797:
URL: https://github.com/apache/iceberg/pull/14797#issuecomment-4797469600

   > Hey @rainerschamm . I've used recently the CDC upsert fix from @t3hw , 
which worked for me, but I found it produces equality deletes, instead of 
positional, which is incompatible with Snowflake (so unfortunately couldn't 
implement it in-house).
   > 
   > Would you know if it would be possible to implement positional deletes 
into the connector? Would be a game changer to drive Iceberg adoption further, 
as Snowflake product team does not contemplate accepting equality deletes' 
reads on their roadmap.
   > 
   > Snowflake user-base is just huge, and that would help with the adtoption 
of Iceberg usage across the industry. Not sure if it's possible to implement 
positional deletes, as an optional feature/flag (maybe by default equality 
deletes?).
   
   I think it is not possible with the current tools, as it would require the 
writer knowing the position of the record in the previous snapshot.
   The DV is used of an in-batch de-duplication when the writer holds the 
positional data of the rows it just wrote in memory, but it is lost when the 
writer commits the snapshot, so out of batch records fall back to Equality 
Deletes.
   
   see the comments by @pvary above, it seems that there is some work being 
done on deprecating Equality Deletes and adding a PK/Indexing mechanism. 
   Ill keep following that work and perhaps merge it to this branch once it 
becomes available. 
   In the meantime, the solution for kafka connect would be to fork and merge 
this branch while actively compacting the table.


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to