Less clock_gettime calls

[filimonov]

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

Changelog category (leave one):

• Performance Improvement

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
Improve the performance of fast queries when max_execution_time=0 by reducing the number of clock_gettime system calls.

Detailed description / Documentation draft:

-- before 
SELECT count() FROM zeros(100000000000)

┌──────count()─┐
│ 100000000000 │
└──────────────┘

1 rows in set. Elapsed: 4.933 sec. Processed 100.00 billion rows, 100.00 GB (20.27 billion rows/s., 20.27 GB/s.)

-- after 
select count() from zeros(100000000000);

┌──────count()─┐
│ 100000000000 │
└──────────────┘

1 rows in set. Elapsed: 3.964 sec. Processed 100.00 billion rows, 100.00 GB (25.23 billion rows/s., 25.23 GB/s.)

[alexey-milovidov]

BTW if only reference/pointer is used, forward declaration is enough.

[filimonov]

It's quite a light header, but sure: #27569

[filimonov]

Performance tests don't show any difference because of

ClickHouse/docker/test/performance-comparison/config/users.d/perf-comparison-tweaks-users.xml

Line 16 in e192711

<max_execution_time>300</max_execution_time>

[filimonov]

Yandex synchronization - maybe i should overload the function instead of changing its interface?

[alexey-milovidov]

Yandex synchronization - maybe i should overload the function instead of changing its interface?

Don't worry about "Yandex synchronization" at all.

[filimonov]

BTW: it looks like part of the original issue is build-related, on official builds difference is smaller, while still clearly visible:

SET max_execution_time = 0; SELECT count() FROM zeros_mt(1000000000000);

1 rows in set. Elapsed: 10.922 sec. Processed 1.00 trillion rows, 1.00 TB (91.56 billion rows/s., 91.56 GB/s.)

SET max_execution_time = 300; SELECT count() FROM zeros_mt(1000000000000);

1 rows in set. Elapsed: 13.124 sec. Processed 1.00 trillion rows, 1.00 TB (76.20 billion rows/s., 76.20 GB/s.)

I've updated also the numbers in the original message.

[alexey-milovidov]

Why do you even try to use unofficial builds?
Only official builds are supported. And we give no guarantee about any third-party builds.
Moreover we are confident that third-party builds will work terribly in multiple ways, because they are not covered by our continuous integration, including performance and fuzz testing.

[filimonov]

Why do you even try to use unofficial builds?

It was my local dev build of master using clang-11. I was debugging a different thing.

But the issue exists, just locally i had it more obvious.