Quanlong Huang created IMPALA-15012:
---------------------------------------
Summary: Doc for client-side bottleneck
Key: IMPALA-15012
URL: https://issues.apache.org/jira/browse/IMPALA-15012
Project: IMPALA
Issue Type: Documentation
Components: Docs
Reporter: Quanlong Huang
Assignee: Sanjana Malhotra
We need document for how to identify client-side bottlenecks. We can add
content in the section of "Using the Query Profile for Performance Tuning" in
this page:
[https://impala.apache.org/docs/build/asf-site-html/topics/impala_explain_plan.html]
There is a section of ImpalaServer in the query profile which has these
counters:
* ClientFetchWaitTimer: Total time spent returning rows to the client and other
client-side processing.
* ClientFetchWaitTimeStats: Summary stats for client fetch wait time.
* ClientFetchLockWaitTimer: Cumulative time client fetch requests waiting for
locks.
* GetInFlightProfileTimeStats: Summary stats of the time dumping profiles when
the query is still in-flight.
If the ClientFetchWaitTimer is close to the total query duration, it means the
bottleneck is on the client side; the Impala server is simply waiting for the
client to fetch the query results. Note that while client is fetching results,
the query state is FINISHED. It doesn't mean the client has fetched all the
rows. Due to the nature of pipeline execution in Impala, subsequent results
cannot be produced or computed if the already generated results are not
fetched. The FINISHED state just means client can start fetching results from
Impala.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
