Will be useful to figure out when the NN is getting overloaded with more connections than it can handle
Reduce fair call queue backoff's impact on clients
Thanks for the work Eric Badger. I think this is a good improvement for 2.7.4; just backported to branch-2.7.
SUCCESS: Integrated in Jenkins build Hadoop-trunk-Commit #11822 (See https://builds.apache.org/job/Hadoop-trunk-Commit/11822/)
HADOOP-14440. Add metrics for connections dropped. Contributed by Eric (brahma: rev abdd609e51a80388493417126c3bc9b1badc0ac1)
Cherry-picked to branch-2.8.
Thanks, Brahma Reddy Battula!
Committed to trunk and branch-2..Eric Badger thanks for your contribution.
Yes,sure..will commit shortly
Hi Brahma Reddy Battula, looks like there are no other comments. Would you mind committing this?
Eric Badger thanks for updating the patch.. +1 on latest patch.. will commit this weekend unless there are no comments from somebody.
This message was automatically generated.
Thanks for the review, Brahma Reddy Battula!
Uploading patch with updated metrics.md, modified comments in the test case to make it more clear what is going on, and fixed the checkstyle.
Eric Badger thanks for working on this.. Approach LGTM.
i) Can you please update metrics.md also..?
ii) how about adding seperate testcase for this,testMaxConnections can be refactored Or can you comment before assertion..?
iii) Address the checkstyle issue to make jenkins happy..
Test failure is unrelated to the patch and possibly related to HADOOP-14078
Thanks, Nathan Roberts! Uploading new patch that fixes the naming nit
Thanks Eric. One minor nit, could name the getter droppedConnections() as getDroppedConnections().
Uploading patch with new metric for dropped connections