Files
Umbraco-CMS/src/Umbraco.PublishedCache.NuCache
Bjarke Berg c06e89af64 Content Delivery API (#14051)
* Add the core parts of the headless PoC

* Add Content API project (WIP - loads of TODOs and dupes that need to be fixed!)

* Rename the content API project and namespaces

* Fixed bad merge

* Rename everything "Headless" to "ContentApi" or "Api"

* Refactor Content + Media: Key => Id, Name not nullable

* Make Content API property return value types independent of datatype configuration

* Clean up refactorings

* First stab at an expansion strategy using content picker as example implementation

* Use named JSON options for content API serialization

* Proper inclusion and registration of the content API

* Introduce API media builder

* Make MNTP return API content/media depending on configuration (instead of links) and support output expansion

* Content API: Get by controllers (#13740)

* Adding ContentApiControllerBase

* Adding get by id and url controllers

* Change route of get all test controller

* Rename to ContentApiController

* Refactoring

* Removing test controller

* Content API: Add start-node header value to deal with url collisions (#13746)

* Use start-node header value to deal with url collisions

* Cleanup

* Rename "url" param to "path"

* Adding a start node service to get the start-node header value

* Trim '/' from both beginning and end

* Content API: Support Accept-Language header (#13831)

* Move the content API JSON type resolver to an appropriate namespace

* Add localization based on Accept-Language header

* Content API: Output expansion  (#13848)

* Implement request based output expansion strategy + expansion output cache at property level

* Slighty leaner implementation for default output expansion strategy

* Clarify the code a bit

* Fix bad merge

* Encapsulate content API dependencies in the DI

* Support multi-site and multi-culture routing + a little rename/refactor (#13882)

* Support multi-site and multi-culture routing + a little rename/refactor

* Make the by route controller handle root nodes

* Rename Url to Path in API content output

* Add a few comments for magic route creation

* Rename services from "Default" to "Noop"

* Ensure that Umbraco can boot without adding "AddContentApi()" to ConfigureServices

* Moved incorrectly placed media builder

* Fix API routes (#13915)

* Fix multi URL picker value converter trying to access disposed objects in edge cases

* Delivery API: Content routing and structure (#13984)

* Introduce content route instead of content path, rename and rework start item (previously start node) handling

* Strip out start node path in generated route path

* Make the start-item header take precedence over the request domain

* Conditionally enabling the delivery API + add protection and preview support + refactor all services to be singletons + ensure no-op implementations for all required services (#13992)

* Include umbraco properties (width, height, ...) in the Media Properties collection (#14023)

* Move umbraco properties (width, height, ...) to the Properties collection of the API Media model

* Don't output the umbracoFile property of media items

* Add content type deny list (#14025)

* Create a deny list of content types and utilize it for output generation

* Add unit tests

* Dedicated property cache level for Content API (#14027)

* Support redirect tracking (#14033)

* Create a deny list of content types and utilize it for output generation

* Add unit tests

* Handle redirect tracking in the content API

* Include start item routing info for redirects

* Add cultures and their routes to the API output (#14038)

* Create a deny list of content types and utilize it for output generation

* Add unit tests

* Handle redirect tracking in the content API

* Include start item routing info for redirects

* Add culture routes to root output (for HREFLANG support)

* Rename redirect service method to better match its purpose

* Review changes

* Delivery API: Query controller (#14041)

* Initial commit

* Custom ContentAPIFieldDefinitionCollection

* Make index IUmbracoContentIndex

* Add querying for children by parent id (key)

* Add missing interface

* Adding querying endpoint

* Test code

* Compose unpublishedValueSet, so that you get the correct data in the ContentAPI index

* Renaming

* Fix ancestorKeys index values

* Adding IApiQueryExtensionService to be able to query the ContentAPI index in a generic way

* Fix IApiQueryService and clean up QueryContentApiController using it

* Support querying for path

* Fix content API indexing

* Fix default sorting

* Implement concrete QueryOption implementations

* Introduce new ExecuteQuery that uses the Core OptionHandlers

* Implement ExecuteQuery

* Change ExecuteQuery signature and implementation

* Implement demo sorting and fetching

* Add query option handlers and collection builder for them

* Cleanup

* Revert "Conditionally enabling the delivery API + add protection and preview support + refactor all services to be singletons + ensure no-op implementations for all required services (#13992)"

This reverts commit 78e1f748e55383baecd123d06457111e18f13365.

* Revert "Delivery API: Content routing and structure (#13984)"

This reverts commit a0292ae5350362dd6c1c5bc9763deda928c78a75.

* Revert "Fix multi URL picker value converter trying to access disposed objects in edge cases"

This reverts commit 6b7c37a5bf7871bee93a2b2640bbc6ef591f14db.

* Revert "Conditionally enabling the delivery API + add protection and preview support + refactor all services to be singletons + ensure no-op implementations for all required services (#13992)"

This reverts commit 78e1f748e55383baecd123d06457111e18f13365.

* Revert "Delivery API: Content routing and structure (#13984)"

This reverts commit a0292ae5350362dd6c1c5bc9763deda928c78a75.

* Revert "Fix multi URL picker value converter trying to access disposed objects in edge cases"

This reverts commit 6b7c37a5bf7871bee93a2b2640bbc6ef591f14db.

* Fix multi URL picker value converter trying to access disposed objects in edge cases

* Delivery API: Content routing and structure (#13984)

* Introduce content route instead of content path, rename and rework start item (previously start node) handling

* Strip out start node path in generated route path

* Make the start-item header take precedence over the request domain

* Conditionally enabling the delivery API + add protection and preview support + refactor all services to be singletons + ensure no-op implementations for all required services (#13992)

* Test commit

* Refactored interfaces for the query handlers and for the selectors (that will handle the value of the fetch query option)

* Implemented a base class for the query options

* Refactored the names of the selectors and made use of the base class

* Refactored the ApiQueryService

* Refactored the QueryContentApiController.cs

* Conditionally enabling the delivery API + add protection and preview support + refactor all services to be singletons + ensure no-op implementations for all required services (#13992)

* Fixing merge gone wrong

* Fix multi URL picker value converter trying to access disposed objects in edge cases

* Delivery API: Content routing and structure (#13984)

* Introduce content route instead of content path, rename and rework start item (previously start node) handling

* Strip out start node path in generated route path

* Make the start-item header take precedence over the request domain

* Conditionally enabling the delivery API + add protection and preview support + refactor all services to be singletons + ensure no-op implementations for all required services (#13992)

* Make fetching work with the new setup

* Moving files to dedicated folders

* Removing ? for array

* Rename selector query method

* Implement FilterHandler and some filters

* Implement SortHandler and sort some sorts

* Refactoring

* Adding more fields to index due to querying

* Appending filtering and sorting queries

* Implementing a new ISelectorHandler without Examine types

* Re-implementing the collection to have a dedicated one for the selectors

* Implementing a new IFilterHandler without Examine types & refactoring the filters implementing it

* Adding a new collection dedicated to filters

* Renaming the old collection

* Implementing a new ISortHandler without Examine types & refactoring the sorts implementing it

* Adding a new collection for the sorts & adding all collections to UmbracoBuilder.Collections

* Refactoring the service to use the new collections and types

* Refactoring the fields in ContentApiFieldDefinitionCollection

* Remove nullability in Handlers

* Don't return null for selector

* Add TODO for having the filters support negation

* Changing the SortType to FieldType with our custom types on the SortOption

* Fix AncestorsSelector

* Fix ApiQueryService

* Documentation

* Fix Swagger docs

* Refactor the QueryContentApiController

* Adding handling for the IApiContentResponse in the JsonTypeResolver

* Refactor the service to use a safe fallback value in Examine queries

* Adding Noop for the IApiQueryService

* Cleanup

* Remove comment

* Fix name field for indexing

* Don't inherit QueryOptionBase in filters

* Fix casing for API index constant + swap FIXME with TODO

* Add TODO for handling missing fetch with start-item header

* Rename query handler parameters to not leak source (i.e. query string)

---------

Co-authored-by: kjac <kja@umbraco.dk>
Co-authored-by: Elitsa <>
Co-authored-by: Zeegaan <nge@umbraco.dk>

* Delivery API: Adding pagination to query endpoint (#14083)

* Adding pagination to query endpoint

* Optimize the paging using Examine directly

* Fix comment

* Remove skip/take code duplication

---------

Co-authored-by: kjac <kja@umbraco.dk>

* Add missing CompatibilitySuppressions.xml

* Make Delivery API packable

* Make Api.Common packable

* Renamed extension method and namespace so it is discoverable

* Untangle ApiVersion configuration into api.common, so delivery api do not require the management api to boot.

* configure options in management api

* RTE output as JSON for Content API (#14067)

* Conditionally serve RTE output as JSON instead of HTML

* Fixed merge

* Rename to Delivery API (#14119)

* Rename ContentApi to DeliveryApi

* Rename delivery API index implementation

* Update comments from "Content API" to "Delivery API"

* Rename project from Content to Delivery

* Add dedicated controller base for content delivery API

* Rename delivery API content index to include "content" specifically

* Fix compat suppressions

---------

Co-authored-by: kjac <kja@umbraco.dk>
Co-authored-by: Elitsa Marinovska <21998037+elit0451@users.noreply.github.com>
Co-authored-by: Zeegaan <nge@umbraco.dk>
2023-04-19 11:21:31 +02:00
..
2023-04-19 11:21:31 +02:00

NuCache Documentation

HOW IT WORKS

NuCache uses a ContentStore to keep content - basically, a dictionary of int => content, and some logic to maintain it up-to-date. In order to provide immutable content to pages rendering, a ContentStore can create ContentViews. A ContentView basically is another int => content dictionary, containing entries only for things that have changed in the ContentStore - so the ContentStore changes, but it updates the views so that they

Views are chained, ie each new view is the parent of previously existing views. A view knows its parent but not the other way round, so views just disappear when they are GC.

When reading the cache, we read views up the chain until we find a value (which may be null) for the given id, and finally we read the store itself.

The PublishedSnapshotService manages a ContentStore for content, and another for media. When a PublishedSnapshot is created, the PublishedSnapshotService gets ContentView objects from the stores. Views provide an immutable snapshot of the content and media.

When the PublishedSnapshotService is notified of changes, it notifies the stores. Then it resync the current PublishedSnapshot, so that it requires new views, etc.

Whenever a content, media or member is modified or removed, the cmsContentNu table is updated with a json dictionary of alias => property value, so that a content, media or member can be loaded with one database row - this is what is used to populate the in-memory cache.

A ContentStore actually stores ContentNode instances, which contain what's common to both the published and draft version of content + the actual published and/or draft content.

LOCKS

Each ContentStore is protected by a reader/writer lock 'Locker' that is used both by the store and its views to ensure that the store remains consistent.

Each ContentStore has a _freezeLock object used to protect the 'Frozen' state of the store. It's a disposable object that releases the lock when disposed, so usage would be: using (store.Frozen) { ... }.

The PublishedSnapshotService has a _storesLock object used to guarantee atomic access to the set of content, media stores.

CACHE

For each set of views, the PublishedSnapshotService creates a SnapshotCache. So a SnapshotCache is valid until anything changes in the content or media trees. In other words, things that go in the SnapshotCache stay until a content or media is modified.

For each PublishedSnapshot, the PublishedSnapshotService creates a PublishedSnapshotCache. So a PublishedSnapshotCache is valid for the duration of the PublishedSnapshot (usually, the request). In other words, things that go in the PublishedSnapshotCache stay (and are visible to) for the duration of the request only.

The PublishedSnapshotService defines a static constant FullCacheWhenPreviewing, that defines how caches operate when previewing:

  • when true, the caches in preview mode work normally.
  • when false, everything that would go to the SnapshotCache goes to the PublishedSnapshotCache. At the moment it is true in the code, which means that eg converted values for previewed content will go in the SnapshotCache. Makes for faster preview, but uses more memory on the long term... would need some benchmarking to figure out what is best.

Members only live for the duration of the PublishedSnapshot. So, for members SnapshotCache is never used, and everything goes to the PublishedSnapshotCache.

All cache keys are computed in the CacheKeys static class.

TESTS

For testing purposes the following mechanisms exist:

The PublishedSnapshot type has a static Current property that is used to obtain the 'current' PublishedSnapshot in many places, going through the PublishedCachesServiceResolver to get the current service, and asking the current service for the current PublishedSnapshot, which by default relies on UmbracoContext. For test purposes, it is possible to override the entire mechanism by defining PublishedSnapshot.GetCurrentPublishedSnapshotFunc which should return a PublishedSnapshot.

A PublishedContent keeps only id-references to its parent and children, and needs a way to retrieve the actual objects from the cache - which depends on the current PublishedSnapshot. It is possible to override the entire mechanism by defining PublishedContent. GetContentByIdFunc or .GetMediaByIdFunc.

Setting these functions must be done before Resolution is frozen.

STATUS

"Detached" contents & properties, which need to be refactored anyway, are not supported by NuCache - throwing NotImplemented in ContentCache.

All the cached elements rely on guids for the cache key, and not ints, so it could be possible to support detached contents & properties, even those that do not have an actual int id, but again this should be refactored entirely anyway.

Not doing any row-version checks (see XmlStore) when reloading from database, though it is maintained in the database. Two FIXME in PublishedSnapshotService. Should we do it?

There is no on-disk cache at all so everything is reloaded from the cmsContentNu table when the site restarts. This is pretty fast, but we should experiment with solutions to store things locally (and deal with the sync issues, see XmlStore...).

Doing our best with PublishedMember but the whole thing should be refactored, because PublishedMember exposes properties that IPublishedContent does not, and that are going to be lost soon as the member is wrapped (content set, model...) - so we probably need some sort of IPublishedMember.

/