JulianJaffePinterest commented on pull request #12137:
URL: https://github.com/apache/druid/pull/12137#issuecomment-1017164092


   @loquisgon but how does this solution avoid the same issue? Tombstones will 
be announced immediately, before the replacement segments are loaded, and so 
you still have the scenario where old data is unavailable but new data has not 
yet been loaded. To my understanding, you are attempting to retrofit atomic 
table replacement onto atomic partition replacement either way and so the same 
data availability issues exist here. To help me understand, what happens in 
your code if the historicals are under load and so take minutes to load the new 
physical segments? How are the tombstone segments held back in this case?
   
   More broadly, it would be a useful feature for Druid to have the ability to 
atomically swap a table, not just a partition (which is just a restatement of 
what you're trying to accomplish here I think). A common way this is done in 
other systems is to fully load the new version of a table and only once that's 
done update metadata pointers from the old to the new. Was a similar design 
considered here? Why or why not? If you've already considered and rejected this 
approach, can we add a paragraph to the design alternatives discussing why?


-- 
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