Hi runzhiwang, Thanks for highlighting this issue. Can we look at limiting the number of open RocksDB handles and also reducing cache allocation for each RocksDB?
> On Jun 12, 2020, at 1:20 AM, 617072740 <617072...@qq.com> wrote: > > Hi all, > One rocksdb for each container in datanode cost too much off-heap > memory. There are too choices to handle this: Merge Rocksdb or Remove > Rocksdb, the design doc of these two choices: > https://docs.google.com/document/d/18Ybg-NjyU602c-MYXaJHP6yrg-dVMZKGyoK5C_pp1mM/edit#. > And the jira is https://issues.apache.org/jira/browse/HDDS-3630. > I would prefre to Merge Rocksdb, because Remove Rocksdb really is a > big work and also need to do some work which could be done by rocksdb. > The shortage of Remove Rocksdb is in the design doc. > If you have any suggestions, please let me know. Thank you very much. > > > > > Thanks, > runzhiwang --------------------------------------------------------------------- To unsubscribe, e-mail: ozone-dev-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-dev-h...@hadoop.apache.org
