[Enhancement] optimize variant key query #67235

dirtysalt · 2025-12-25T21:15:20Z

Why I'm doing

What I'm doing

VariantRowValue Memory Layout Optimization
- Changed from storing [metadata] and [value] in separate std::string objects to a single contiguous buffer _raw
- Reduces memory allocations from 2 to 1, improves cache locality
- Added predefined null-binary default for empty variants
- serialize()/serialize_size() now copy single block instead of two separate copies
VariantMetadata Lookup Index Caching
- Added lazy dict_size calculation and _lookup_index struct to cache parsed dictionary string views
- get_key() now returns std::string_view instead of std::string to avoid allocations
- New _build_lookup_index() builds dictionary string view cache once
- Fast _get_index() using cached views with std::lower_bound for sorted unique dictionaries
Fast Inline Little-Endian Readers
- Added inline_read_little_endian_unsigned32_fixed_size() template for 1-4 byte reads
- Uses __builtin_memcpy for zero-overhead byte swapping via arrow::bit_util::FromLittleEndian
- Replaces slower VariantUtil::read_little_endian_unsigned32() calls throughout
VariantValue Access Optimizations
- Inlined value_header() and basic_type() accessors
- Cached dictionary index scan in get_object_by_key() with hint parameter
- Template-based field ID search (SEARCH_DICT_INDEX_SZ) for different sizes (1-4 bytes)
- Added UNLIKELY hints for error path optimization
CastVariantToMap Improvements
- get_key() returns std::string_view, uses Slice(key) directly to avoid key copies
- Error messages use fmt::format for better formatting
Naming Refactoring
- Renamed ObjectExtraction → VariantObjectExtraction
- Renamed ArrayExtraction → VariantArrayExtraction

Fixes #issue

What type of PR is this:

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 pr needs auto generate documentation
This is a backport pr

Bugfix cherry-pick branch check:

Note

Focuses on performance and copy-reduction across Variant handling.

Variant storage/layout: VariantRowValue now stores [metadata][value] in a single _raw buffer with a predefined null binary; serialize/serialize_size copy one block
Dictionary lookup: VariantMetadata lazily computes dict size, caches dict string views via _build_lookup_index, and get_key returns std::string_view; fast inline LE readers replace previous helpers
Variant access: Inlines headers/type checks; accelerates get_object_by_key using cached indexes and fixed-size reads; adds bounds/overflow checks and UNLIKELY paths
Path parsing: Renames to VariantObjectExtraction/VariantArrayExtraction; VariantPath::seek uses std::visit with ASSIGN_OR_RETURN
Casting/conversion: CastVariantToMap uses string_view keys and Slice(key); improved error messages; variant_converter reduces copies and simplifies boolean/string handling
Misc: DEBUG-only warning on destroying MemTracker with unreleased children
Tests: Updated variant_path_parser_test and variant_value_test to new types/behaviors

^{Written by Cursor Bugbot for commit 4435c60. This will update automatically on new commits. Configure here.}

alvin-celerdata · 2025-12-25T21:17:33Z

@cursor review

cursor

✅ Bugbot reviewed your changes and found no bugs!

be/src/types/variant_value.h

Signed-off-by: yan zhang <[email protected]>

alvin-celerdata

@cursor review

alvin-celerdata · 2025-12-27T00:18:26Z

@cursor review

cursor

✅ Bugbot reviewed your changes and found no bugs!

Signed-off-by: yan zhang <[email protected]>

dirtysalt · 2025-12-27T09:45:58Z

be/src/exprs/cast_expr_map.cpp

                }

-                keys_builder.append(Slice(std::string(key)));
+                keys_builder.append(Slice(key));


remove memcpy

dirtysalt · 2025-12-27T09:46:21Z

be/src/exprs/variant_path_parser.cpp

    for (size_t seg_idx = 0; seg_idx < variant_path->segments.size(); ++seg_idx) {
        const auto& segment = variant_path->segments[seg_idx];
-
-        StatusOr<VariantValue> sub;


reduce ctor of StatusOr, which has memory allocation operations inside.

dirtysalt · 2025-12-27T09:47:11Z

be/src/types/variant_value.h

-              _metadata(_metadata_raw),
-              _value(_value_raw) {}
+            : _raw(), _metadata_size(metadata.size()), _metadata(), _value() {
+        if (metadata.data() != VariantMetadata::kEmptyMetadata.data() ||


fastpath for creating empty row value, to avoid memory allocation.

and cocnate metadata and value into a continous memory area.

dirtysalt · 2025-12-27T09:49:32Z

be/src/util/variant.cpp

-    return (header() & kSortedStringMask) != 0;
-}
+Status VariantMetadata::_build_lookup_index() const {
+    uint32_t dict_sz = dict_size();


build dict_strings in a single path. in this building process, we only need to call read_little_endian_unsigned N times and use speicliaize template methods.

and during get_index method, we can use linear scan or binary search in a lightweight way.

dirtysalt · 2025-12-27T09:49:50Z

be/src/util/variant.cpp

 }

-StatusOr<std::string> VariantMetadata::get_key(uint32_t index) const {
+StatusOr<std::string_view> VariantMetadata::get_key(uint32_t index) const {


no need to return string, just return string_view suffices.

dirtysalt · 2025-12-27T09:50:10Z

be/src/util/variant.cpp

+
+    if (is_sorted_unique) {
+        if (dict_sz > kBinarySearchThreshold) {
+            auto it = std::lower_bound(dict_strings.begin(), dict_strings.end(), key);


use off-the-shelf algorithm.

dirtysalt · 2025-12-27T09:50:39Z

be/src/util/variant.cpp


-std::vector<uint32_t> VariantMetadata::get_index(std::string_view key) const {
+Status VariantMetadata::_get_index(std::string_view key, void* _indexes, int hint) const {
+    KeyIndexVector& indexes = *static_cast<KeyIndexVector*>(_indexes);


Use specialized vector index to avoid mmeory allocation.

dirtysalt · 2025-12-27T09:51:43Z

be/src/util/variant.cpp

-    case VariantType::INT8:
-        json_str << std::to_string(*variant.get_int8());
+    case VariantType::INT8: {
+        ASSIGN_OR_RETURN(int8_t value, variant.get_int8());


check status right here?

dirtysalt · 2025-12-27T09:52:26Z

be/src/util/variant_converter.cpp

-            return str.status();
-        }
-
-        result.append(Slice(std::string(str.value())));


use string_view instead of string to avoid extra memory operation.

Signed-off-by: yan zhang <[email protected]>

alvin-celerdata · 2025-12-27T18:33:34Z

@cursor review

be/src/util/variant_converter.cpp

be/src/util/variant.cpp

Signed-off-by: yan zhang <[email protected]>

alvin-celerdata · 2025-12-28T00:11:04Z

@cursor review

github-actions · 2025-12-28T01:05:27Z

[Java-Extensions Incremental Coverage Report]

✅ pass : 0 / 0 (0%)

github-actions · 2025-12-28T01:05:30Z

[FE Incremental Coverage Report]

✅ pass : 0 / 0 (0%)

github-actions · 2025-12-28T01:22:13Z

[BE Incremental Coverage Report]

❌ fail : 2 / 4 (50.00%)

file detail

	path	covered_line	new_line	coverage	not_covered_line_detail
🔵	be/src/exprs/cast_expr_map.cpp	2	4	50.00%	[123, 124]

dirtysalt requested review from a team as code owners December 25, 2025 21:15

mergify bot assigned dirtysalt Dec 25, 2025

cursor bot reviewed Dec 25, 2025

View reviewed changes

alvin-celerdata reviewed Dec 25, 2025

View reviewed changes

be/src/types/variant_value.h Outdated Show resolved Hide resolved

dirtysalt marked this pull request as draft December 26, 2025 08:23

dirtysalt marked this pull request as ready for review December 26, 2025 14:34

dirtysalt added 9 commits December 27, 2025 07:04

optimize memory allocation of variant row value

d7ec425

Signed-off-by: yan zhang <[email protected]>

inline a few cotrs and methods

ff97b4c

Signed-off-by: yan zhang <[email protected]>

remove redudant 'validate_basic_type'

1480508

Signed-off-by: yan zhang <[email protected]>

to avoid extra copy

20f6b0b

Signed-off-by: yan zhang <[email protected]>

optimize load small endian unsigend32

5786494

Signed-off-by: yan zhang <[email protected]>

use template to optimize

3757b8f

Signed-off-by: yan zhang <[email protected]>

reduce memory allocation

44236b5

Signed-off-by: yan zhang <[email protected]>

handle empty raw

c66761c

Signed-off-by: yan zhang <[email protected]>

fix be ut

c2527fa

Signed-off-by: yan zhang <[email protected]>

dirtysalt force-pushed the optimize-variant branch from a742bfe to c2527fa Compare December 26, 2025 23:28

alvin-celerdata reviewed Dec 26, 2025

View reviewed changes

cursor bot reviewed Dec 27, 2025

View reviewed changes

dirtysalt added 4 commits December 27, 2025 12:19

fix be ut

f00a60f

Signed-off-by: yan zhang <[email protected]>

remove calling StatusOr ctor which has memory allocation

3830ddf

Signed-off-by: yan zhang <[email protected]>

refactor some code and reduce a extra copy

a884a9d

Signed-off-by: yan zhang <[email protected]>

refactor code

13e5f29

Signed-off-by: yan zhang <[email protected]>

dirtysalt commented Dec 27, 2025

View reviewed changes

dirtysalt enabled auto-merge (squash) December 27, 2025 10:49

fix ut

2281eea

Signed-off-by: yan zhang <[email protected]>

cursor bot reviewed Dec 27, 2025

View reviewed changes

be/src/util/variant_converter.cpp Outdated Show resolved Hide resolved

be/src/util/variant.cpp Show resolved Hide resolved

fix per comment

4435c60

Signed-off-by: yan zhang <[email protected]>

alvin-celerdata approved these changes Dec 28, 2025

View reviewed changes

alvin-celerdata disabled auto-merge December 28, 2025 01:39

alvin-celerdata merged commit 1b0af66 into StarRocks:main Dec 28, 2025
63 of 64 checks passed

dirtysalt deleted the optimize-variant branch December 28, 2025 01:51

[Enhancement] optimize variant key query #67235

[Enhancement] optimize variant key query #67235

Conversation

dirtysalt commented Dec 25, 2025 • edited by cursor bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why I'm doing

What I'm doing

What type of PR is this:

Checklist:

Bugfix cherry-pick branch check:

Uh oh!

alvin-celerdata commented Dec 25, 2025

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

alvin-celerdata left a comment

Choose a reason for hiding this comment

Uh oh!

alvin-celerdata commented Dec 27, 2025

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alvin-celerdata commented Dec 27, 2025

Uh oh!

Uh oh!

Uh oh!

alvin-celerdata commented Dec 28, 2025

Uh oh!

github-actions bot commented Dec 28, 2025

[Java-Extensions Incremental Coverage Report]

Uh oh!

github-actions bot commented Dec 28, 2025

[FE Incremental Coverage Report]

Uh oh!

github-actions bot commented Dec 28, 2025

[BE Incremental Coverage Report]

file detail

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dirtysalt commented Dec 25, 2025 •

edited by cursor bot

Loading