mengxr edited a comment on issue #24734: [SPARK-27870][SQL][PySpark] Flush 
batch timely for pandas UDF (for improving pandas UDFs pipeline)
URL: https://github.com/apache/spark/pull/24734#issuecomment-499951173
 
 
   Had offline discussion with @WeichenXu123 and I removed `flush_timely` flag. 
There are several scenarios:
   
   1. high volume, large batch
   2. high volume, small batch
   3. low volume, large batch
   4. low volume, small batch
   
   Additional flush doesn't hurt 1) and 3) because writing a large batch 
(bigger than buffer size) would trigger flush already. It would greatly help 4) 
by reducing the latency.
   
   It does add more flush calls to 2), at the rate no higher than 10/second. 
This is significant only if without it the flush frequency is much less than 
10/second. Given the default buffer size 65536, the data volume is less than 
1MB/s to keep the flush frequency under 10/second. Additional flush to local 
tcp socket at this rate doesn't hurt much.

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


With regards,
Apache Git Services

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

Reply via email to