caiconghui opened a new issue #3696:
URL: https://github.com/apache/incubator-doris/issues/3696


   Now, there are many rpcs between be and fe, and some default config in be 
may cause poor performance for query or stream load. In our experiment, we find 
that when we execute stream load in doris, the txn can be easily publish 
timeout and query can be easily rpc timeout, http request to be can be easily 
no response. 
   
   And we read the code for server in be, and we find the num of thrift server 
threads is too small to handle so many rpc which include exec_plan_fragment and 
submit_tasks at the same time,  and http server use reactor model to process 
request, when the stream load performance is poor (server thread would be 
blocked in rpc or  other blocked function) and the num of threads in http 
server is small, it will cause http server can not handle http request in time. 
    
   
   Finally, we do some experiments after change the some config in be, and we 
find that query and stream load both get good performance and http server can 
response in time.
   
   we use thread pool in java client to make 10000 stream load requests to fe 
with 5 be instance. Every request has one row data, and we have 500 tables in 
on db, every table will be inserted 20 rows. the total time cost for java 
client is  35398ms, and some stream load result are as following:
   {
       "TxnId": 276003,
       "Label": "86292a00-511b-47de-82e2-6dfa9c2d219d",
       "Status": "Success",
       "Message": "OK",
       "NumberTotalRows": 1,
       "NumberLoadedRows": 1,
       "NumberFilteredRows": 0,
       "NumberUnselectedRows": 0,
       "LoadBytes": 12,
       "LoadTimeMs": 31,
       "BeginTxnMs": 0,
       "ReadBufferMs": 0,
       "WriteBufferMs": 12,
       "StreamLoadPutMs": 0,
       "CommitAndPublishMs": 15
   }
   {
       "TxnId": 276006,
       "Label": "7d6023a4-5fb5-4551-9327-4c758f231851",
       "Status": "Success",
       "Message": "OK",
       "NumberTotalRows": 1,
       "NumberLoadedRows": 1,
       "NumberFilteredRows": 0,
       "NumberUnselectedRows": 0,
       "LoadBytes": 12,
       "LoadTimeMs": 31,
       "BeginTxnMs": 0,
       "ReadBufferMs": 0,
       "WriteBufferMs": 12,
       "StreamLoadPutMs": 0,
       "CommitAndPublishMs": 15
   }
   {
       "TxnId": 275256,
       "Label": "3b29cd47-2aa8-4255-9d37-df6148570e93",
       "Status": "Success",
       "Message": "OK",
       "NumberTotalRows": 1,
       "NumberLoadedRows": 1,
       "NumberFilteredRows": 0,
       "NumberUnselectedRows": 0,
       "LoadBytes": 11,
       "LoadTimeMs": 34,
       "BeginTxnMs": 0,
       "ReadBufferMs": 0,
       "WriteBufferMs": 12,
       "StreamLoadPutMs": 0,
       "CommitAndPublishMs": 18
   }
   load all time 35398
    


----------------------------------------------------------------
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]



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to