xianjingfeng

> Can you describe when this can cause the problem? I don't know, I add some code to print _lastCellOfPreviousBlock_ and i found it's value was modified, but i don't...

> The change here will impact performance, so we need more detailed analysis on what is the root cause, and then decide how to better fix it. [HBASE-16372](https://issues.apache.org/jira/browse/HBASE-16372) has explain...

> Mind explaining why it is incorrect? For example, if the number of files to compact is 10 and blocksize is 64k, beforeShipped was called once when every 640k was...

> I suggest to focus on the client side fallback in this PR. Skipping memory data may need careful design on the server side, which could be put in an...

> > > I suggest to focus on the client side fallback in this PR. Skipping memory data may need careful design on the server side, which could be put...

> I use the shell module of ansible to start the shuffle server without this PR, it wont block. command as: `ansible -i hosts_xxx -m shell -a "./start-shuffle-server.sh"` > >...

@zuston Blocked is because we have made a layer of encapsulation for ansible. We will check if the process start successful. Process will not start successful by using native ansible...

@jerqi @zuston 

> And i dont see any blocking in your sceenshot. Not blocked, jvm process have not start if start by ansible

> We recommend users to use StorageType MEMORY_LOCALFILE_HDFS or MEMORY_LOCALFILE. The application won't commit data. So, should we remove `LOCALFILE`?