[
https://issues.apache.org/jira/browse/IMPALA-2939?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17058896#comment-17058896
]
Tim Armstrong commented on IMPALA-2939:
---------------------------------------
[~gaborkaszab] I guess we probably would want to re-profile these queries to
see what the bottleneck is before making targeted fixes.
> Nested Types : Address Runtime & Scoped timer overhead
> ------------------------------------------------------
>
> Key: IMPALA-2939
> URL: https://issues.apache.org/jira/browse/IMPALA-2939
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Affects Versions: Impala 2.4.0
> Reporter: Mostafa Mokhtar
> Priority: Minor
> Labels: complextype, nested_types, performance
> Attachments: Bottom-Up-HotFunctions.csv, Top-Down-HotFunctions.csv,
> nestedTypesQ1.zip
>
>
> For the following query about 45% of the time is spent in updating timers,
> RunTimeProfile and checking query state, since NestedTypes don't always
> operate on Batches the overhead of updating counters is amplified.
> {code}
> select
> l.l_shipdate, count(*) as wins
> from
> customer.c_orders o,
> o.o_lineitems l
> where
> o_orderdate = '1993-12-12'
> group by l.l_shipdate
> order by wins;
> {code}
> |Function||Effective Time by Utilization||
> |clock_gettime 29.8% 0s 0s librt.so.1| clock_gettime|
> |impala::RuntimeProfile::Counter::Add| 5.3%|
> |std::map<std::string, impala::RuntimeProfile::Counter*,
> std::less<std::string>, std::allocator<std::pair<std::string const,
> impala::RuntimeProfile::Counter*>>>::operator[]| 4.7%|
> |impala::RuntimeState::CheckQueryState| 3.2%|
> |impala::MonotonicStopWatch::Stop| 2.7%|
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
