[V5] Database

rfcs/xxxx-v5-database.md

@@ -0,0 +1,170 @@
+- Start Date: 2023-09-04
+- RFC PR: https://github.com/strapi/rfcs/pull/52
+# Summary
+
+We need to support development of new features. Most of them are going towards a more CMS like model. to support this approach we will introduce the notion of `documents` & `links`.
+
+# Motivation
+
+Bringing new features that were not possible in v4. The most important ones:
+
+## Draft and Publish
+
+Introducing a new version of draft & publish. One which allows having one draft and one published entry at the same time.
+
+This version should support having different relations between a draft and a published version.
+
+Consider a scenario where a writer wants to work on an article (draft) while the previously published version remains unaffected. This requires a system that supports simultaneous draft and published entries, leading to the idea of `Documents`.
+
+## Synchronized relations across locales
+
+In Strapi V4, users requested the ability to share the same relation across multiple locales. This led to the concept of `Links`, allowing relations to be shared across different document locales. And also unblocks other use cases mentioned in this RFC.
+
+Instead of breaking v4 relations `Links` will be a different type, and v4 relations will be kept as legacy relations.
+
+# Detailed design
+
+Documents are introduced in v5 as a way to group multiple entries. That is implemented by adding a new `document_id` column in entries:
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/4a02817e-4004-45ee-9e4a-a755e66c2064" width="300px" />
+</p>
+
+Also, all the variations of that document will have its own column in every entry:
+
+- `locale` for i18n
+- `publication_state` for draft and publish
+
+## Document id and entry id
+
+Additionally, in V5, all identifiers (entry ids and document ids) will now be universally unique identifiers (UUIDs). This transition offers significant benefits:
+
+- `Data Transfer`: Previously, transferring data between databases while retaining the same entry IDs was not possible, due to the use of incremental ids.
+
+The switch to UUIDs, which will keep entry and document ids unique, makes this feasible.
+
+- `More secure`: The use of UUIDs also enhances security. With incremental IDs, one could predict the existence of other IDs based on a known ID. For instance, if ID `10` is known, it is reasonable to infer that IDs `9`, `8`, etc., exist in the database. UUIDs, being non-sequential and randomly generated, prevent such easy discovery.
+
+While the probability of two UUIDs colliding is not zero, it is practically negligible, making them an excellent choice for unique identifiers.
+
+## Links
+
+`Links` introduce the following key functionalities:
+
+### Relate entry to document
+
+In V4, you might have encountered this situation:
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/01061116-478b-465f-9b3b-a88ea02598c6" width="600px" />
+</p>
+
+
+When relating an article to a specific author in V4, only articles and authors of the same locale could be linked, because localizations were not really grouped in a single document.
+
+Thus, if an author did not have a locale that matched the article's, creating a relationship was impossible.
+
+`Links` will solve this by referencing a document as a whole, not a specific locale:
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/da52c5de-57c2-4ced-8986-e0f6787d8892" width="600px" />
+</p>
+
+
+### Synchronize links across locales
+
+In V4, if you want to keep the same relation across the same locales, one would have to manually create them, with the risk of missing one.
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/0e2729fe-7673-490b-9493-9ea3a8fb4470" width="600px" />
+</p>
+
+In V5, `links` will offer the possibility to be synchronized across locales.
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/0d5841dd-89b1-42f6-bed5-b463f40f515b" width="600px" />
+</p>
+
+As an additional comment, the synched relation will only affect the locales inside the same publication state. Meaning, relations will not be shared between the `draft` and `published` versions of the document.
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/4a69c2e3-628e-45b1-8c77-338272a61c32" width="600px" />
+</p>
+
+
+### Polymorphic links
+
+V5 will eventually support polymorphic links.
+
+Those will allow use cases like Website Menus , linked to many other content types
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/29628824-c735-4776-b651-4c99ad11db8f" width="500px" />
+</p>
+
+This was just not possible in V4.
+
+## Link content schema
+
+Links will be represented as so in the content type schema:
+
+```jsx
+{
+  attributes: {
+    myLink: {
+      type: "link",
+      // Target one (or multiple) content types.
+      // Target all by not defining target
+      target: "uid-a" | ["uid-a", "uid-b"] | undefined,
+      multiple: true|false
+    }
+  }
+}
+```
+
+## Links Database Schema
+
+The database schema of `links` join tables :
+
+<p align="center">
+	<img src="https://github.com/strapi/rfcs/assets/20578351/c866277d-92da-4861-891c-22cafa207a65" width="700px" />
+</p>
+
+
+The join table will now contain:
+
+- `Owner side doc id`: Reference to do the document id of the owner side (Article in the example)
+- `Inverse side doc id`: Reference ot the document id of the inverse side (Author in the example)
+- `pivot columns`: Reference the owner side variations (locale & publication state).
+- `order`: Only in `manyWays` and `morphMany` relations.
+
+## Query engine
+
+The query engine will expose the same methods as before.
+
+This layer will be used to interact with the database rows directly, and as so, methods such as `findOne` and `update` will reference entry ids and not document ids.
+
+Interactions with documents will be done in the application layer `DocumentService` , which will be documented in another RFC.
+
+# Tradeoffs
+
+Benefits of this approach:
+
+- Links schema reduces the number of queries needed to read and create relations.
+- It is conceptually easy, and similar to our approach.
+- It's easy to share the same relation between different document locales.
+
+Downsides:
+
+- Links will not allow for bidirectionality. A relation that references an entry to a document introduces many complexities that prevents bidirectionality, or makes it rather difficult. The first versions of links will not have bidirectionality in mind.
+
+# Unresolved questions
+
+- How do we manage life cycles?
+  - A database lifecycle would trigger at an entry level, not a document level. As everything will be document oriented, database life cycles might have less sense from a user perspective in V5.
+  - Document life cycles could be managed in a upper layer than the database.
+- How are we going to support v4 relation types with the new document ids.
+  - Could we connect a v4 relation with a document Id, or do we only support connecting with entry ids?
+- Do we prefix internal strapi columns in db?
+  - We need to avoid collisions
+- Which universal ID to be used. UUID, CUID, NANOID, ULID.