Sanjeet Malhotra created PHOENIX-7729:
-----------------------------------------
Summary: Identify top N slowest parallel scans for metrics
reporting
Key: PHOENIX-7729
URL: https://issues.apache.org/jira/browse/PHOENIX-7729
Project: Phoenix
Issue Type: Improvement
Reporter: Sanjeet Malhotra
Assignee: Sanjeet Malhotra
Currently in Phoenix, we create one HBase scan per region to be scanned and
submit these scans for execution in-parallel. The HBase scan metrics are
collected as sum of scan metrics across all parallel scans. But given the scans
are executing in parallel, the sum of scan metrics isn't very helpful as that
doesn't let us focus on the slowest scan which will be the reason for whole
query slowing down.
We need a way to identify the top N slowest scans so that their metrics can be
captured for reporting and debugging the reason behind query execution getting
slowed down.
Additionally, with HBASE-29233, we are capturing region hash and RS name in the
scan metrics by capturing region level scan metrics. If we can capture the
region hash and RS name for the top N slowest scans the we can exactly
pin-point the RS running the slowest scans.
Thus, enabling HBASE-29233 along with reporting metrics for top N slowest scans
can help track the slowest scan and drill down on reason for it getting slowed.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
