[Enhancement] Throttle scan to wait for topn filter

[satanson]

Why I'm doing:

High associative hash join is time-consuming, since it magnifies data volume many times. if there is a topn operator above it, then we can use this topn filter generated by topn operator to reduce input data volume of the hash join; however, when perform tests on this, the scan operator below hash join always transfers data to the hash join so fast that make the topn filter take effects on scan operator too late, so input data volume of the hash join is not reduced successfully, so we design a back pressure mechanism that works as follows:

1. scan operator allows rows of 10 times of limit+offset in topn operator to pass through to hash join operator, then wait for a small period of time(e.g. 100ms), we call this period the throttle period.
2. scan operator has_output return false in throttle period, so scan operator does not transfer any data, just give a chance to topn operator to generate a topn filter.
3. when current throttle period ends, scan operator use topn filter to filter its output data, if the topn filter is high selective, then scan operator can terminate this back pressure mechanism, just use this topn filter to filter incoming data.
4. otherwise, scan operator begins an another throttle period.
5. scan operator maybe begin throttle period for several times which controlled by the session variable: back_pressure_back_rounds, the throttle period equals to back_pressure_throttle_time_upper_bound/back_pressure_back_rounds.
6. topn_filter_back_pressure_mode is used to turn on/off the back pressure mechanism.

Test

when topn filter back pressure mechanism is opened,data volume of left side of hash join is reduced to 1/60.

when it is closed

data volume of left side of hash join is reduced to 1/60.

1. pipeline_dop=0, concurrency=20,back_pressure_back_rounds=3

+===================+==============+
| cases             | latency(sec) |
+===================+==============+
| disable opt       | 11.692       |
| enable opt(60ms)  | 5.920        |
| enable opt(100ms) | 5.853        |
| enable opt(300ms) | 5.959        |
| enable opt(600ms) | 6.279        |
+-------------------+--------------+

disable opt means turn off the optimization;
enable opt(60ms) means turn on the optimization; and back_pressure_throttle_time_upper_bound=60, i.e. total throttle time does not exceeds 60ms.

2. pipeline_dop=1, back_pressure_throttle_time_upper_bound=300,back_pressure_back_rounds=10

+=============+==================+=================+=========+
| concurrency | disable opt(sec) | enable opt(sec) | speedup |
+=============+==================+=================+=========+
| 1           | 0.991            | 0.953           | 1.0X    |
| 10          | 4.089            | 2.831           | 1.4X    |
| 20          | 7.735            | 5.034           | 1.5X    |
| 40          | 15.210           | 9.688           | 1.5X    |
| 60          | 22.760(OOM)      | 14.600          | 1.5X    |
+-------------+------------------+-----------------+---------+

What I'm doing:

Fixes #issue

What type of PR is this:

• BugFix
• Feature
• Enhancement
• Refactor
• UT
• Doc
• Tool

Does this PR entail a change in behavior?

• Yes, this PR will result in a change in behavior.
• No, this PR will not result in a change in behavior.

If yes, please specify the type of change:

• Interface/UI changes: syntax, type conversion, expression evaluation, display information
• Parameter changes: default values, similar parameters but with different default values
• Policy changes: use new policy to replace old one, functionality automatically enabled
• Feature removed
• Miscellaneous: upgrade & downgrade compatibility, etc.

Checklist:

• I have added test cases for my bug fix or my new feature
• This pr needs user documentation (for new or modified features or behaviors)
  • I have added documentation for my new feature or new function
• This is a backport pr

Bugfix cherry-pick branch check:

• I have checked the version labels which the pr will be auto-backported to the target branch
  • 3.4
  • 3.3
  • 3.2
  • 3.1
  • 3.0

[starrocks-cr]

The most risky bug in this code is:
Division by zero when calculating backPressureThrottleTime.

You can modify the code like this:

this.backPressureThrottleTime = 
    this.backPressureThrottleTimeUpperBound / Math.max(this.backPressureNumRows, 1);

This change prevents division by zero by using Math.max(this.backPressureNumRows, 1) instead of Math.min(this.backPressureNumRows, 1), ensuring the denominator is never zero.

[starrocks-cr]

The most risky bug in this code is:
Incorrect calculation of _current_throttle_deadline in should_throttle(), which results in the use of milliseconds instead of correctly casting steady_clock::now.

You can modify the code like this:

#include <chrono> // add this to the includes to use std::chrono_literals

if (_phase == PH_UNTHROTTLE) {
    if (_current_num_rows <= _num_rows_limiter.limit()) {
        return false;
    }
    _phase = PH_THROTTLE;
    _current_throttle_deadline = 
        (steady_clock::now() + milliseconds(_throttle_time_limiter.limit())).time_since_epoch().count();
    return true;
}

[starrocks-cr]

The most risky bug in this code is:
Potential segmentation fault due to a null pointer dereference when accessing runtime_bloom_filters().

You can modify the code like this:

void evaluate_topn_runtime_filters(Chunk* chunk) {
    if (chunk == nullptr || chunk->is_empty() || !_topn_filter_back_pressure) {
        return;
    }
    auto* topn_runtime_filters = runtime_bloom_filters();
    if (topn_runtime_filters != nullptr) {
        auto input_num_rows = chunk->num_rows();
        _init_topn_runtime_filter_counters();
        topn_runtime_filters->evaluate(chunk, _topn_filter_eval_context);
        _topn_filter_back_pressure->inc_num_rows(input_num_rows);
        if (_topn_filter_eval_context.selectivity.empty()) {
            _topn_filter_back_pressure->update_selectivity(1.0);
        } else {
            double selectivity = _topn_filter_eval_context.selectivity.begin()->first;
            if (input_num_rows > 1024) {
                _topn_filter_back_pressure->update_selectivity(selectivity);
            }
        }
    }
}

[sonarqubecloud]

Quality Gate passed

Issues
4 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[github-actions]

[Java-Extensions Incremental Coverage Report]

✅ pass : 0 / 0 (0%)

[github-actions]

[BE Incremental Coverage Report]

❌ fail : 16 / 124 (12.90%)

file detail

	path	covered_line	new_line	coverage	not_covered_line_detail
🔵	be/src/exec/olap_scan_node.cpp	0	7	00.00%	[112, 113, 114, 115, 116, 117, 118]
🔵	be/src/exec/pipeline/topn_runtime_filter_back_pressure.h	0	39	00.00%	[29, 31, 32, 34, 35, 36, 37, 39, 50, 51, 53, 54, 56, 57, 58, 59, 60, 63, 64, 67, 68, 69, 70, 72, 73, 76, 77, 78, 79, 80, 81, 82, 86, 88, 89, 90, 91, 92, 93]
🔵	be/src/exprs/runtime_filter_bank.cpp	0	16	00.00%	[394, 398, 399, 400, 417, 418, 595, 596, 597, 600, 601, 602, 603, 604, 648, 649]
🔵	be/src/exec/connector_scan_node.cpp	1	8	12.50%	[78, 79, 80, 81, 82, 83, 84]
🔵	be/src/exec/scan_node.h	1	6	16.67%	[136, 138, 140, 142, 143]
🔵	be/src/exec/pipeline/scan/scan_operator.cpp	3	12	25.00%	[94, 95, 96, 97, 98, 99, 100, 101, 102]
🔵	be/src/exec/pipeline/scan/scan_operator.h	11	36	30.56%	[180, 181, 182, 183, 184, 185, 186, 188, 189, 190, 196, 197, 198, 199, 200, 201, 202, 203, 204, 205, 206, 207, 208, 210, 220]

[github-actions]

[FE Incremental Coverage Report]

❌ fail : 22 / 49 (44.90%)

file detail

	path	covered_line	new_line	coverage	not_covered_line_detail
🔵	com/starrocks/qe/SessionVariable.java	4	12	33.33%	[4591, 4592, 4595, 4599, 4600, 4603, 4607, 4608]
🔵	com/starrocks/planner/OlapScanNode.java	15	33	45.45%	[1015, 1016, 1017, 1018, 1019, 1020, 1061, 1062, 1063, 1064, 1065, 1550, 1551, 1552, 1553, 1554, 1555, 1556]
🔵	com/starrocks/planner/RuntimeFilterDescription.java	2	3	66.67%	[159]
🔵	com/starrocks/planner/SortNode.java	1	1	100.00%	[]