Avoid string allocation and improve performance of `JsonProperty.WriteTo` #90074

karakasa · 2023-08-06T10:04:44Z

Description

System.Text.Json.JsonProperty.WriteTo uses get_Name, then JsonElement.GetPropertyName(), a method would always allocate a string.

This PR reuses the underlying UTF8 json ROS<byte> by adding helper methods into JsonDocument & JsonElement, to avoid alloc. Allocation is further removed by inlining JsonReaderHelper.GetUnescapedSpan into JsonElement.WritePropertyNameTo, which makes the method allocation-less when property names are shorter than JsonConstants.StackallocByteThreshold, or don't need unescaping, and the underlying buffer is long enough.

Generally the change improves the performance of JsonProperty.WriteTo by ~30%.

Fix #88767

Microbenchmark

BenchmarkDotNet v0.13.7-nightly.20230717.35, Windows 10 (10.0.19045.2546/22H2/2022Update)
Intel Core i7-8700K CPU 3.70GHz (Coffee Lake), 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.100-preview.6.23330.14
  [Host]     : .NET 8.0.0 (8.0.23.32907), X64 RyuJIT AVX2
  Job-TGMBKK : .NET 8.0.0 (42.42.42.42424), X64 RyuJIT AVX2

PowerPlanMode=00000000-0000-0000-0000-000000000000  Arguments=/p:EnableUnsafeBinaryFormatterSerialization=true  Toolchain=CoreRun  
IterationTime=250.0000 ms  MaxIterationCount=20  MinIterationCount=15  
WarmupCount=1

Type	TestCase	Mean	Error	StdDev	Median	Min	Max	Gen0	Allocated
Original	Simple	362.4 ns	4.14 ns	3.45 ns	361.4 ns	358.5 ns	370.3 ns	0.0305	200 B
Improved	Simple	230.5 ns	2.31 ns	2.04 ns	231.1 ns	227.2 ns	233.8 ns	-
Original	WithEscapedChars	662.8 ns	2.50 ns	2.34 ns	662.4 ns	659.1 ns	666.8 ns	0.0425	272 B
Improved	WithEscapedChars	458.6 ns	9.16 ns	8.57 ns	456.1 ns	450.6 ns	483.5 ns	-
Original	WithUnicodeChars	769.3 ns	5.15 ns	4.56 ns	768.0 ns	764.1 ns	779.1 ns	0.0462	304 B
Improved	WithUnicodeChars	530.3 ns	9.35 ns	10.01 ns	526.0 ns	519.8 ns	550.5 ns	-

Simple: {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0}
WithEscapedChars: {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2}
WithUnicodeChars: {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2,"测试":3}

Micro-benchmark code

using BenchmarkDotNet.Attributes;
using System;
using System.Linq;
using MicroBenchmarks;
using System.IO;

namespace System.Text.Json.Node.Tests
{
    [BenchmarkCategory(Categories.Libraries, Categories.JSON)]
    [MemoryDiagnoser]
    public class JsonPropertyWriteToTest
    {
        public enum TestCaseType
        {
            Simple,
            WithEscapedCharacters,
            WithUnescapedUnicodeCharacters,
        }

        [ParamsAllValues]
        public TestCaseType TestCase;


        const string JsonString1 = """
                {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0}
                """;
        const string JsonString2 = """
                {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2}
                """;
        const string JsonString3 = """
                {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2,"测试":3}
                """;

        private static string GetTestCase(TestCaseType type)
        {
            switch (type)
            {
                case TestCaseType.Simple:
                    return JsonString1;
                case TestCaseType.WithEscapedCharacters:
                    return JsonString2;
                case TestCaseType.WithUnescapedUnicodeCharacters:
                    return JsonString3;
            }

            return "";
        }

        const int MAX_FOLD = 10;

        private static JsonProperty[] Properties;

        private static MemoryStream Stream;
        private static Utf8JsonWriter Writer;
        [GlobalSetup]
        public void PrepareJson()
        {
            Properties = JsonDocument.Parse(Encoding.UTF8.GetBytes(GetTestCase(TestCase))).RootElement.EnumerateObject().ToArray();
            Stream = new(JsonString2.Length * MAX_FOLD);
            Writer = new Utf8JsonWriter(Stream);
        }

        [Benchmark]
        public MemoryStream WriteProperties()
        {
            Stream.Position = 0;
            Writer.Reset();

            Writer.WriteStartObject();

            foreach (var it in Properties)
                it.WriteTo(Writer);

            Writer.WriteEndObject();
            Writer.Flush();

            return Stream;
        }
    }
}

System.Text.Json.JsonProperty.WriteTo uses get_Name, calling JsonElement.GetPropertyName() which would allocate a string. Use ReadOnlySpan<byte> from the underlying UTF8 json, when possible, by adding helper methods into JsonDocument & JsonElement. Fix dotnet#88767

Current code unescapes & escapes property names and uses ToArray. Avoid alloc by adding internal GetRaw/WriteRaw methods.

Original code doesn't handle GetRaw/WriteRaw on escaped property names correctly.

Allocations are further avoided when the property name is shorter than JsonConstants.StackallocByteThreshold, by inlining JsonReaderHelper.GetUnescapedSpan.

ghost · 2023-08-06T10:04:56Z

Tagging subscribers to this area: @dotnet/area-system-text-json, @gregsdennis
See info in area-owners.md if you want to be subscribed.

Issue Details

Description

System.Text.Json.JsonProperty.WriteTo uses get_Name, then JsonElement.GetPropertyName(), a method would always allocate a string.

This PR reuses the underlying UTF8 json ROS<byte>, by adding helper methods into JsonDocument & JsonElement. Allocation is further removed by inlining JsonReaderHelper.GetUnescapedSpan into JsonProperty.WriteTo, which makes the method allocation-less when property names are shorter than JsonConstants.StackallocByteThreshold.

The change also improves the performance of JsonProperty.WriteTo.

Fix #88767

Microbenchmark

BenchmarkDotNet v0.13.7-nightly.20230717.35, Windows 10 (10.0.19045.2546/22H2/2022Update)
Intel Core i7-8700K CPU 3.70GHz (Coffee Lake), 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.100-preview.6.23330.14
  [Host]     : .NET 8.0.0 (8.0.23.32907), X64 RyuJIT AVX2
  Job-TGMBKK : .NET 8.0.0 (42.42.42.42424), X64 RyuJIT AVX2

PowerPlanMode=00000000-0000-0000-0000-000000000000  Arguments=/p:EnableUnsafeBinaryFormatterSerialization=true  Toolchain=CoreRun  
IterationTime=250.0000 ms  MaxIterationCount=20  MinIterationCount=15  
WarmupCount=1

Original

Type	TestCase	Mean	Error	StdDev	Median	Min	Max	Gen0	Allocated
Original	Simple	362.4 ns	4.14 ns	3.45 ns	361.4 ns	358.5 ns	370.3 ns	0.0305	200 B
Improved	Simple	230.5 ns	2.31 ns	2.04 ns	231.1 ns	227.2 ns	233.8 ns	-
Original	WithEscapedChars	662.8 ns	2.50 ns	2.34 ns	662.4 ns	659.1 ns	666.8 ns	0.0425	272 B
Improved	WithEscapedChars	458.6 ns	9.16 ns	8.57 ns	456.1 ns	450.6 ns	483.5 ns	-
Original	WithUnicodeChars	769.3 ns	5.15 ns	4.56 ns	768.0 ns	764.1 ns	779.1 ns	0.0462	304 B
Improved	WithUnicodeChars	530.3 ns	9.35 ns	10.01 ns	526.0 ns	519.8 ns	550.5 ns	-

Simple: {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0}
WithEscapedChars: {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2}
WithUnicodeChars: {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2,"测试":3}

Micro-benchmark code

using BenchmarkDotNet.Attributes;
using System;
using System.Linq;
using MicroBenchmarks;
using System.IO;

namespace System.Text.Json.Node.Tests
{
    [BenchmarkCategory(Categories.Libraries, Categories.JSON)]
    [MemoryDiagnoser]
    public class JsonPropertyWriteToTest
    {
        public enum TestCaseType
        {
            Simple,
            WithEscapedCharacters,
            WithUnescapedUnicodeCharacters,
        }

        [ParamsAllValues]
        public TestCaseType TestCase;


        const string JsonString1 = """
                {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0}
                """;
        const string JsonString2 = """
                {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2}
                """;
        const string JsonString3 = """
                {"normal1":0,"normal2":0,"normal3":0,"normal4":0,"normal5":0,"\"q\t\\m\"m\t":1,"\u6d4b\u8bd5":2,"测试":3}
                """;

        private static string GetTestCase(TestCaseType type)
        {
            switch (type)
            {
                case TestCaseType.Simple:
                    return JsonString1;
                case TestCaseType.WithEscapedCharacters:
                    return JsonString2;
                case TestCaseType.WithUnescapedUnicodeCharacters:
                    return JsonString3;
            }

            return "";
        }

        const int MAX_FOLD = 10;

        private static JsonProperty[] Properties;

        private static MemoryStream Stream;
        private static Utf8JsonWriter Writer;
        [GlobalSetup]
        public void PrepareJson()
        {
            Properties = JsonDocument.Parse(Encoding.UTF8.GetBytes(GetTestCase(TestCase))).RootElement.EnumerateObject().ToArray();
            Stream = new(JsonString2.Length * MAX_FOLD);
            Writer = new Utf8JsonWriter(Stream);
        }

        [Benchmark]
        public MemoryStream WriteProperties()
        {
            Stream.Position = 0;
            Writer.Reset();

            Writer.WriteStartObject();

            foreach (var it in Properties)
                it.WriteTo(Writer);

            Writer.WriteEndObject();
            Writer.Flush();

            return Stream;
        }
    }
}

Author:	karakasa
Assignees:	-
Labels:	`area-System.Text.Json`
Milestone:	-

karakasa · 2023-08-06T10:10:50Z

@dotnet-policy-service agree

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonDocument.cs

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonProperty.cs

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonElement.cs

Shorten names of new methods; Add a test of writing out special names.

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonDocument.cs

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonElement.cs

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonDocument.cs

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonElement.cs

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonDocument.cs

ghost · 2023-09-10T05:01:12Z

Draft Pull Request was automatically closed for 30 days of inactivity. Please let us know if you'd like to reopen it.

eiriktsarpalis · 2023-09-18T14:31:41Z

Hi @karakasa, this was closed automatically because it was still marked as draft. Is there any pending work on this PR before it is marked for review?

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonDocument.cs

karakasa · 2023-09-21T13:41:24Z

@eiriktsarpalis

Could you please reopen the PR? I have fixed the code.

src/libraries/System.Text.Json/src/System/Text/Json/Document/JsonDocument.cs

eiriktsarpalis

Thank you!

karakasa · 2023-09-21T17:02:30Z

Thank you!

You are the best! 🤩

eiriktsarpalis · 2023-09-22T08:46:06Z

Test failures are known issues, merging away.

karakasa added 5 commits August 5, 2023 23:15

Avoid alloc in unescaping & escaping.

41f5c37

Current code unescapes & escapes property names and uses ToArray. Avoid alloc by adding internal GetRaw/WriteRaw methods.

Fix bugs on escaped property names

d4d84b1

Original code doesn't handle GetRaw/WriteRaw on escaped property names correctly.

Change IndexOf to Contains if possible.

c368aa3

Further avoid alloc by inlining GetUnescapedSpan

c5c5593

Allocations are further avoided when the property name is shorter than JsonConstants.StackallocByteThreshold, by inlining JsonReaderHelper.GetUnescapedSpan.

dotnet-issue-labeler bot added the area-System.Text.Json label Aug 6, 2023

ghost added the community-contribution Indicates that the PR has been added by a community member label Aug 6, 2023

karakasa marked this pull request as ready for review August 6, 2023 14:11

eiriktsarpalis added the tenet-performance Performance related issue label Aug 7, 2023