V15: Hybrid Caching (#16938)

* Update to dotnet 9 and update nuget packages

* Update umbraco code version

* Update Directory.Build.props

Co-authored-by: Elitsa Marinovska <21998037+elit0451@users.noreply.github.com>

* Include preview version in pipeline

* update template projects

* update global json with specific version

* Update version.json to v15

* Rename TrimStart and TrimEnd to string specific

* Rename to Exact

* Update global.json

Co-authored-by: Ronald Barendse <ronald@barend.se>

* Remove includePreviewVersion

* Rename to trim exact

* Add new Hybridcache project

* Add tests

* Start implementing PublishedContent.cs

* Implement repository for content

* Refactor to use async everywhere

* Add cache refresher

* make public as needed for serialization

* Use content type cache to get content type out

* Refactor to use ContentCacheNode model, that goes in the memory cache

* Remove content node kit as its not needed

* Implement tests for ensuring caching

* Implement better asserts

* Implement published property

* Refactor to use mapping

* Rename to document tests

* Update to test properties

* Create more tests

* Refactor mock tests into own file

* Update property test

* Fix published version of content

* Change default cache level to elements

* Refactor to always have draft

* Refactor to not use PublishedModelFactory

* Added tests

* Added and updated tests

* Fixed tests

* Don't return empty object with id

* More tests

* Added key

* Another key

* Refactor CacheService to be responsible for using the hybrid cache

* Use notification handler to remove deleted content from cache

* Add more tests for missing functions

* Implement missing methods

* Remove HasContent as it pertains to routing

* Fik up test

* formatting

* refactor variable names

* Implement variant tests

* Map all the published content properties

* Get item out of cache first, to assert updated

* Implement member cache

* Add member test

* Implement media cache

* Implement property tests for media tests

* Refactor tests to use extension method

* Add more media tests

* Refactor properties to no longer have element caching

* Don't use property cache level

* Start implementing seeding

* Only seed when main

* Add Immutable for performance

* Implement permanent seeding of content

* Implement cache settings

* Implement tests for seeding

* Update package version

* start refactoring nurepo

* Refactor so draft & published nodes are cached individually

* Refactor RefreshContent to take node instead of IContent

* Refactor media to also use cache nodes

* Remove member from repo as it isn't cached

* Refactor media to not include preview, as media has no draft

* create new benchmark project

* POC Integration benchmarks with custom api controllers

* Start implementing content picker tests

* Implement domain cache

* Rework content cache to implement interface

* Start implementing elements cache

* Implement published snapshot service

* Publish snapshot tests

* Use snapshot for elements cache

* Create test proving we don't clear cache when updating content picker

* Clear entire elements cache

* Remove properties from element cache, when content gets updated.

* Rename methods to async

* Refactor to use old cache interfaces instead of new ones

* Remove snapshot, as it is no longer needed

* Fix tests building

* Refactor domaincache to not have snapshots

* Delete benchmarks

* Delete benchmarks

* Add HybridCacheProject to Umbraco

* Add comment to route value transformer

* Implement is draft

* remove snapshot from property

* V15 updated the hybrid caching integration tests to use ContentEditingService (#16947)

* Added builder extension withParentKey

* Created builder with ContentEditingService

* Added usage of the ContentEditingService to SETUP

* Started using ContentEditingService builder in tests

* Updated builder extensions

* Fixed builder

* Clean up

* Clean up, not done

* Added Ids

* Remove entries from cache on delete

* Fix up seeding logic

* Don't register hybrid cache twice

* Change seeded entry options

* Update hybrid cache package

* Fix up published property to work with delivery api again

* Fix dependency injection to work with tests

* Fix naming

* Dont make caches nullable

* Make content node sealed

* Remove path and other unused from content node

* Remove hacky 2 phase ctor

* Refactor to actually set content templates

* Remove umbraco context

* Remove "HasBy" methods

* rename property data

* Delete obsolete legacy stuff

* Add todo for making expiration configurable

* Add todo in UmbracoContext

* Add clarifying comment in content factory

* Remove xml stuff from published property

* Fix according to review

* Make content type cache injectible

* Make content type cache injectible

* Rename to database cache repository

* Rename to document cache

* Add TODO

* Refactor to async

* Rename to async

* Make everything async

* Remove duplicate line from json schema

* Move Hybrid cache project

* Remove leftover file

* Refactor to use keys

* Refactor published content to no longer have content data, as it is on the node itself

* Refactor to member to use proper content node ctor

* Move tests to own folder

* Add immutable objects to property and content data for performance

* Make property data public

* Fix member caching to be singleton

* Obsolete GetContentType

* Remove todo

* Fix naming

* Fix lots of exposed errors due to scope test

* Add final scope tests

* Rename to document cache service

* Rename test files

* Create new doc type tests

* Add ignore to tests

* Start implementing refresh for content type save

* Clear contenttype cache when contenttype is updated

* Fix test

Teh contenttype is not upated unless the property is dirty

* Use init for ContentSourceDto

* Fix get by key in PublishedContentTypeCache

* Remove ContentType from PublishedContentTypeCache when contenttype is deleted

* Update to preview 7

* Fix versions

* Increase timeout for sqlite integration tests

* Undo timeout increase

* Try and undo init change to ContentSourceDto

* That wasn't it chief

* Try and make DomainAndUrlsTests non NonParallelizable

* Update versions

* Only run cache tests on linux for now

---------

Co-authored-by: Elitsa Marinovska <21998037+elit0451@users.noreply.github.com>
Co-authored-by: Ronald Barendse <ronald@barend.se>
Co-authored-by: Andreas Zerbst <andr317c@live.dk>
Co-authored-by: Sven Geusens <sge@umbraco.dk>
Co-authored-by: Andreas Zerbst <73799582+andr317c@users.noreply.github.com>
Co-authored-by: nikolajlauridsen <nikolajlauridsen@protonmail.ch>
This commit is contained in:
Nikolaj Geisle
2024-09-10 00:49:18 +09:00
committed by GitHub
parent dcd6f1fbf4
commit 2704d4a34a
102 changed files with 5881 additions and 132 deletions

View File

@@ -0,0 +1,147 @@
using System.Text;
using K4os.Compression.LZ4;
using MessagePack;
using MessagePack.Resolvers;
using Umbraco.Cms.Core.Models;
using Umbraco.Cms.Core.PropertyEditors;
namespace Umbraco.Cms.Infrastructure.HybridCache.Serialization;
/// <summary>
/// Serializes/Deserializes <see cref="ContentCacheDataModel" /> document to the SQL Database as bytes using
/// MessagePack
/// </summary>
internal sealed class MsgPackContentNestedDataSerializer : IContentCacheDataSerializer
{
private readonly MessagePackSerializerOptions _options;
private readonly IPropertyCacheCompression _propertyOptions;
public MsgPackContentNestedDataSerializer(IPropertyCacheCompression propertyOptions)
{
_propertyOptions = propertyOptions ?? throw new ArgumentNullException(nameof(propertyOptions));
MessagePackSerializerOptions? defaultOptions = ContractlessStandardResolver.Options;
IFormatterResolver? resolver = CompositeResolver.Create(
// TODO: We want to be able to intern the strings for aliases when deserializing like we do for Newtonsoft but I'm unsure exactly how
// to do that but it would seem to be with a custom message pack resolver but I haven't quite figured out based on the docs how
// to do that since that is part of the int key -> string mapping operation, might have to see the source code to figure that one out.
// There are docs here on how to build one of these: https://github.com/neuecc/MessagePack-CSharp/blob/master/README.md#low-level-api-imessagepackformattert
// and there are a couple examples if you search on google for them but this will need to be a separate project.
// NOTE: resolver custom types first
// new ContentNestedDataResolver(),
// finally use standard resolver
defaultOptions.Resolver);
_options = defaultOptions
.WithResolver(resolver)
.WithCompression(MessagePackCompression.Lz4BlockArray)
.WithSecurity(MessagePackSecurity.UntrustedData);
}
public ContentCacheDataModel? Deserialize(IReadOnlyContentBase content, string? stringData, byte[]? byteData, bool published)
{
if (byteData != null)
{
ContentCacheDataModel? cacheModel =
MessagePackSerializer.Deserialize<ContentCacheDataModel>(byteData, _options);
Expand(content, cacheModel, published);
return cacheModel;
}
if (stringData != null)
{
// NOTE: We don't really support strings but it's possible if manually used (i.e. tests)
var bin = Convert.FromBase64String(stringData);
ContentCacheDataModel? cacheModel = MessagePackSerializer.Deserialize<ContentCacheDataModel>(bin, _options);
Expand(content, cacheModel, published);
return cacheModel;
}
return null;
}
public ContentCacheDataSerializationResult Serialize(IReadOnlyContentBase content, ContentCacheDataModel model, bool published)
{
Compress(content, model, published);
var bytes = MessagePackSerializer.Serialize(model, _options);
return new ContentCacheDataSerializationResult(null, bytes);
}
public string ToJson(byte[] bin)
{
var json = MessagePackSerializer.ConvertToJson(bin, _options);
return json;
}
/// <summary>
/// Used during serialization to compress properties
/// </summary>
/// <param name="content"></param>
/// <param name="model"></param>
/// <param name="published"></param>
/// <remarks>
/// This will essentially 'double compress' property data. The MsgPack data as a whole will already be compressed
/// but this will go a step further and double compress property data so that it is stored in the nucache file
/// as compressed bytes and therefore will exist in memory as compressed bytes. That is, until the bytes are
/// read/decompressed as a string to be displayed on the front-end. This allows for potentially a significant
/// memory savings but could also affect performance of first rendering pages while decompression occurs.
/// </remarks>
private void Compress(IReadOnlyContentBase content, ContentCacheDataModel model, bool published)
{
if (model.PropertyData is null)
{
return;
}
foreach (KeyValuePair<string, PropertyData[]> propertyAliasToData in model.PropertyData)
{
if (_propertyOptions.IsCompressed(content, propertyAliasToData.Key, published))
{
foreach (PropertyData property in propertyAliasToData.Value.Where(x =>
x.Value != null && x.Value is string))
{
if (property.Value is string propertyValue)
{
property.Value = LZ4Pickler.Pickle(Encoding.UTF8.GetBytes(propertyValue));
}
}
foreach (PropertyData property in propertyAliasToData.Value.Where(x =>
x.Value != null && x.Value is int intVal))
{
property.Value = Convert.ToBoolean((int?)property.Value);
}
}
}
}
/// <summary>
/// Used during deserialization to map the property data as lazy or expand the value
/// </summary>
/// <param name="content"></param>
/// <param name="nestedData"></param>
/// <param name="published"></param>
private void Expand(IReadOnlyContentBase content, ContentCacheDataModel nestedData, bool published)
{
if (nestedData.PropertyData is null)
{
return;
}
foreach (KeyValuePair<string, PropertyData[]> propertyAliasToData in nestedData.PropertyData)
{
if (_propertyOptions.IsCompressed(content, propertyAliasToData.Key, published))
{
foreach (PropertyData property in propertyAliasToData.Value.Where(x => x.Value != null))
{
if (property.Value is byte[] byteArrayValue)
{
property.Value = new LazyCompressedString(byteArrayValue);
}
}
}
}
}
}