zeroshade commented on code in PR #1403: URL: https://github.com/apache/arrow-go/pull/1403#discussion_r4351823082
########## arrow/array/concat_dictionary_reuse_test.go: ########## @@ -0,0 +1,167 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package array_test + +import ( + "fmt" + "testing" + + "github.com/apache/arrow-go/v18/arrow" + "github.com/apache/arrow-go/v18/arrow/array" + "github.com/apache/arrow-go/v18/arrow/memory" + "github.com/stretchr/testify/require" +) + +// makeConcatDictionaryChunks reuses the same dictionary Data for all chunks +// selecting the same values entry, including when the indices differ. +func makeConcatDictionaryChunks(mem memory.Allocator, valueSets [][]string, sequence []int, rows int, nullable, sliced bool) []arrow.Array { + dictionaries := make([]arrow.Array, len(valueSets)) + for i, values := range valueSets { + builder := array.NewStringBuilder(mem) + builder.AppendValues(values, nil) + dictionaries[i] = builder.NewArray() + builder.Release() + } + defer func() { + for _, dict := range dictionaries { + dict.Release() + } + }() + + dtype := &arrow.DictionaryType{ + IndexType: arrow.PrimitiveTypes.Int32, + ValueType: arrow.BinaryTypes.String, + } + out := make([]arrow.Array, 0, len(sequence)) + for chunkIndex, dictIndex := range sequence { + builder := array.NewInt32Builder(mem) + for row := 0; row < rows; row++ { + if nullable && (row+chunkIndex)%11 == 0 { + builder.AppendNull() + } else { + builder.Append(int32((row + chunkIndex) % len(valueSets[dictIndex]))) + } + } + indices := builder.NewArray() + builder.Release() + dictArray := array.NewDictionaryArray(dtype, indices, dictionaries[dictIndex]) + indices.Release() + + if sliced && rows > 1 { + start := int64(chunkIndex % 2) + slice := array.NewSlice(dictArray, start, int64(rows)) + dictArray.Release() + dictArray = slice.(*array.Dictionary) + } + out = append(out, dictArray) + } + return out +} + +func releaseConcatDictionaryChunks(chunks []arrow.Array) { + for _, chunk := range chunks { + chunk.Release() + } +} + +func TestConcatenateDictionaryReuse(t *testing.T) { + base := []string{"alpha", "beta", "gamma", "delta", "epsilon"} + reversed := []string{"epsilon", "gamma", "alpha", "zeta", "beta"} + tests := []struct { + name string + valueSets [][]string + sequence []int + rows int + nullable bool + sliced bool + }{ + {"shared", [][]string{base}, []int{0, 0, 0, 0}, 29, false, false}, + {"shared_sliced_nulls", [][]string{base}, []int{0, 0, 0, 0, 0}, 29, true, true}, + {"equal_distinct_data", [][]string{base, append([]string(nil), base...)}, []int{0, 1, 0, 1}, 29, true, false}, + {"mixed_repeated", [][]string{base, reversed}, []int{0, 1, 0, 1, 1, 0, 1, 0}, 37, true, true}, + {"mixed_empty_chunks", [][]string{base, reversed}, []int{0, 1, 0, 1}, 0, false, false}, + {"single_chunk", [][]string{base}, []int{0}, 31, true, false}, + } + + for _, tc := range tests { + t.Run(tc.name, func(t *testing.T) { + mem := memory.NewCheckedAllocator(memory.NewGoAllocator()) + defer mem.AssertSize(t, 0) + + inputs := makeConcatDictionaryChunks(mem, tc.valueSets, tc.sequence, tc.rows, tc.nullable, tc.sliced) + defer releaseConcatDictionaryChunks(inputs) + expected := make([]string, 0) + for _, in := range inputs { + for i := 0; i < in.Len(); i++ { + expected = append(expected, in.ValueStr(i)) + } + } + + combined, err := array.Concatenate(inputs, mem) + require.NoError(t, err) + defer combined.Release() + require.NoError(t, array.ValidateFull(combined)) + require.Equal(t, len(expected), combined.Len()) + for i, want := range expected { + require.Equal(t, want, combined.ValueStr(i), "row %d", i) + } + }) + } +} + +// BenchmarkConcatenateDictionaryReuse measures the complete public concatenate +// call with shared dictionaries and repeated dictionary versions. Run with: +// go test ./arrow/array -run '^$' -bench '^BenchmarkConcatenateDictionaryReuse$' -benchmem +func BenchmarkConcatenateDictionaryReuse(b *testing.B) { + base := make([]string, 8192) + for i := range base { + base[i] = fmt.Sprintf("dictionary-value-%06d", i) + } + changed := append([]string(nil), base...) + for i := 0; i < len(changed); i += 16 { + changed[i] = fmt.Sprintf("new-value-%06d", i) + } + second := append([]string(nil), base...) + for i := 1; i < len(second); i += 16 { + second[i] = fmt.Sprintf("another-new-value-%06d", i) + } + + cases := []struct { + name string + valueSets [][]string + sequence []int + }{ + {"shared_dictionary", [][]string{base}, []int{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}}, + {"repeated_mixed_dictionaries", [][]string{base, changed, second}, []int{0, 1, 2, 0, 1, 2, 0, 1, 2, 0, 1, 2}}, + } + for _, tc := range cases { + b.Run(tc.name, func(b *testing.B) { + mem := memory.NewGoAllocator() + inputs := makeConcatDictionaryChunks(mem, tc.valueSets, tc.sequence, 256, true, false) + defer releaseConcatDictionaryChunks(inputs) + b.ReportAllocs() + b.ResetTimer() + for i := 0; i < b.N; i++ { Review Comment: Nit: please use `for b.Loop()`; go.mod already requires Go 1.26. ########## arrow/array/concat.go: ########## @@ -326,13 +326,29 @@ func unifyDictionaries(mem memory.Allocator, data []arrow.ArrayData, dt *arrow.D defer unifier.Release() newLookup := make([]*memory.Buffer, len(data)) + // Reusing the same dictionary data must also reuse its transposition. + // A dictionary's indices in the unified result never change as new + // dictionaries are added to the unifier. + transpositions := make(map[*Data]*memory.Buffer) for i, d := range data { - dictArr := MakeFromData(d.Dictionary()) - defer dictArr.Release() + dictData := d.Dictionary().(*Data) + if existing, ok := transpositions[dictData]; ok { Review Comment: Nit: this assertion relies on `*Data` being the only `arrow.ArrayData` implementation (true today). Keying the map by `arrow.ArrayData` would avoid it, since interface values holding the same pointer compare equal. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
