vinothchandar edited a comment on issue #2330:
URL: https://github.com/apache/hudi/issues/2330#issuecomment-743277688


   @borislitvak https://twitter.com/byte_array/status/1337319056433528832?s=21 
good discussion. Thanks for starting it
   
   On performance hit, no i don't think iceberg incurs any additional 
performance hit due to this in the normal path. The trade off is coarser, table 
level locking (in contrast e.g rdbms-es do row level page level locking). So 
even if writes don't conflict with each other, you end up locking and failing 
one of the writes. Imagine that one of them is a large backfill job, then it 
wastes a lot of compute. 
   
   
   With conflict checks and relying on hdfs atomic renames for the update, 
at-least offers some mechanism to let all non-conflicting writes commit. Best 
approach IMO for such systems that deal with large amounts of data is to try to 
embrace a more log based design 
   
   I know this is a lot more info than what you were probably looking for. :)


----------------------------------------------------------------
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.

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


Reply via email to