anonymouscodeholic commented on pull request #13393:
URL: https://github.com/apache/flink/pull/13393#issuecomment-712010812


   > **Sidenote:**
   > Checking the latest RocksDB source codes 
([write_buffer_manager.cc](https://github.com/facebook/rocksdb/blob/master/memtable/write_buffer_manager.cc#L54),
 
[write_buffer_manager.h](https://github.com/facebook/rocksdb/blob/master/include/rocksdb/write_buffer_manager.h#L51)
 and 
[column_family.cc](https://github.com/facebook/rocksdb/blob/master/db/column_family.cc#L215))
 we could make sure the calculations stay the same as our current frocksdb 
version. However, this sanity check is still implementation bound and we need 
to watch it when upgrading the base version of frocksdb.
   
   Yes, the calculations in RocksDB side are fixed. I think I eye-balled those 
through a while back and it seemed the calculations haven't changed.
   
   If the calculations changed in RocksDB and Flink was modified to use such a 
RocksDB-version that would mean that the experience of the end user of Flink 
would change too. I mean that the cases where this performance issue happens 
would not be the same anymore. Thinking aloud - maybe Flink should be more 
explicit when setting up RocksDB and not rely on the default values. Now some 
values are set in the Flink code and some RocksDB calculates.
   
   BTW, what's the reason of not migrating Flink to the latest (or at least to 
a newer) RocksDB version?


----------------------------------------------------------------
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