[GitHub] [hbase] ndimiduk commented on issue #754: HBASE-22978 : Online slow response log

[GitBox <gi...@apache.org>]

ndimiduk commented on issue #754: HBASE-22978 : Online slow response log
URL: https://github.com/apache/hbase/pull/754#issuecomment-584229446
 
 
   >> 1. Since the detailed info is going through this ring buffer subsystem, why do we continue to log at all? Why not replace the logging with a metric?
   > 1. Ring Buffer is not a replacement of slow logs that we write to file, it's just a complete representation of slow logs, whereas in log files we might have trimmed logs. We also have plan to write all slowlogs to HDFS (optionally) and provide HDFS file path to user optionally, at that time we can think of removing logging in log files. However, I am not sure about metric. Which metric are you suggesting to expose? You mean slow logs with some parameters?
   
   I'm suggesting we do away with file logging and instead expose the presence of slow activity as a (new?) metric.
   
   By the time this patch lands, we'll have the logged information in two places, neither of which, on their own, allow an operator to easily see how prevalent the problem is. I'm suggesting that we could instead provide a metric that says "this region server is experiencing N slow request per time interval T", as well as providing detailed incident fidelity via this new mechanism.

----------------------------------------------------------------
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:
users@infra.apache.org


With regards,
Apache Git Services