[prototype] Compiled schema synchronization

[bdunogier]

Eases the synchronization of graphql schema when using multiple servers.

Publishing a schema

One server will do the schema generation + compiling, and run a command (ibexa:graphql:publish-schema) to publish the schema. Publishing will a) push the compiled schema (%kernel.cache_dir%/overblog/graphql-bundle/__definitions__/*) to a SharedSchema, and b) set the published schema timestamp using a TimestampHandler.

In the current prototype, the TimestampHandler is based on Redis, and the SharedSchema on a locally mounted shared folder.

Pulling a published schema

Other servers, when a GraphQL query is executed (UpdateSchemaIfNeeded subscriber), will compare the timestamp to theirs (mtime of __classes.map). If a newer one is available, it will be pulled from the SharedSchema, and copied locally. It will get installed in a shutdown function to prevent op cache issues.

Since the graphql schema types are compiled into the container as services, types that were added to the published schema (new content types, etc) need to be registered on runtime. This is done by the Schema\Sync\AddTypesSolutions subscriber. It checks which of the type classes do not have a solution in the current schema, and adds them to it.

TODO

• Add proper, dedicated exceptions to the SharedSchema implementation
• Add a flysystem based implementation, that would include local, but also support any remote filesystem (need to update our version of flysystem to the new stable)
• Add configuration for how many updates should be kept. When a new update is published, trim the old ones to respect that limit.

[sonarcloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
5 Code Smells

No Coverage information
0.0% Duplication