aweisberg commented on PR #4508:
URL: https://github.com/apache/cassandra/pull/4508#issuecomment-3643177948

   RE InteropTokenRange test and testing splitting of range reads
   The take away from InteropTokenRangeTest is that you test queries with those 
token bounds and GT/GTE/LT/LTE to make sure that your migration code correctly 
splits range reads across tracked/untracked boundaries. There are just a ton of 
edge cases there. The thing you are testing here is the splitting and that the 
commands it outputs are valid and actually return the data that is supposed to 
be in the read.
   
   InteropTokenRangeTest doesn't demonstrate different permutations of 
migration, but there a small number of tokens so you can enumerate and test all 
range migration combinations.
   
   > I added a "wait for epoch" in the prepare phase handler for repair which 
should serve that purpose. A repair starts, it waits for the epoch, and then no 
more untracked writes will be coordinated by it by the time it replies
   That tells you that the epoch has been received but it doesn't tell you 
anything about in flight writes that didn't complete yet? For that you would do 
the op order thing where you have the node update to the epoch and then create 
an op order to make sure all in flight writes that don't use that epoch are 
retired.
   
   There is still the separate issue of this can cause writes to fail and the 
errors get surfaced to the client which is not ideal.
   
   RE Hints
   Right now hints will get stuck for all keyspaces if you enable mutation 
tracking and there are hints for a keyspace with mutation tracking. That seems 
problematic. Hints are host oriented and contain a log of hints for multiple 
keyspaces. There doesn't seem to be graceful cleanup of hints for keyspaces 
that have mutation tracking enabled so hint delivery for other keyspaces can 
continue.
   


-- 
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