virajjasani edited a comment on issue #754: HBASE-22978 : Online slow response log URL: https://github.com/apache/hbase/pull/754#issuecomment-579605235 > I think you misunderstood me here... > > "With the patch we would run at RS entrance only and for any slow running query it will go to this ring buffer running at RS. Just that this patch doesn't include storing slow logs in HDFS, which is also a good idea and probably up for a subtask given the priority is storing slow logs in ring buffer first. Storing them in HDFS would be just an option as part of another Jira(sub-task)." > > What I was saying is that your ringbuffer only keeps dfsclient ops. I was wondering if we could put actual requests into a system like this. If they are taking a long time, we could inspect the ringbuffer to see which type of queries are taking a while. It would be coarser grained than the dfs accesses only. Was just wondering if you think we could do this in a follow-on based on work h ere. Oh okk, my bad. Looks good for the follow-up task for sure, unless we are already putting these requests in ring buffer. One question: Does actual request not come to `RpcServer.logResponse()` already? If not, then definitely +1 for follow-up task. I thought all requests already do land in RpcServer. No? Edit: @saintstack you are suggesting to run ringbuffer at RS entrance and consume request soon as it arrives at RS and before triggering DFSClient ops? If so, we need to determine which one is slow for RS to send response to HBaseClient/HMaster and accordingly log entry in ring buffer right? +1 for the idea.
---------------------------------------------------------------- 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
