Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
Impala 2.12.0
-
None
-
ghx-label-6
Description
In a production cluster I am seeing a high rate of 'mmap' calls under a workload which starts and stops a lot of exchanges. These are causing a lot contention for the mmap semaphore, which is then causing page faults to be rather slow. By running perf record -e syscalls:*mmap I can see that most of the calls are coming from KRPC's call to GetLoggedInUser when initting a Proxy object. This was fixed by https://gerrit.cloudera.org/#/c/9895/ which needs to be cherry-picked over.
Attachments
Issue Links
- is part of
-
IMPALA-7006 Rebase KRPC onto Kudu upstream repository
- Resolved