This file describes experimental and more advanced Relay features. It can be very unstable due to its nature so be careful. Here be dragons!
- Mock Data Generation: https://github.com/facebook/relay/commit/09d317943f6936ffb0002154c389b6d7a507c58d
renderPolicy: https://github.com/facebook/relay/commit/b1cf05de8770122b30d491c4265df01e161e67c9 (partial/full)
- New GC release buffer: https://github.com/mrtnzlml/relay/pull/126/commits/6ed264413ba8cdd586d695e5ed234951ee9eca13
- complex arguments with nested variables are now supported
- HTTP persister example: https://github.com/facebook/relay/commit/aaa9588e081d3591ad8d043e924cacfadc06ec80
- TODO: special
- https://github.com/relay-tools/relay-connection-handler-plus (use-case: https://github.com/facebook/relay/issues/3065)
- entrypoints: prepareEntryPoint, EntryPointContainer + tests
There are different tradeoffs across completeness, consistency, and performance, and there isn't one approach that is appropriate for every app. Relay focuses on cases where consistency matters: if you don't need consistency then a simpler/lighter solution can be more appropriate. (source)
Useful Links (learning resources):
Relay Config (
Relay supports configuration via CLI but also via configuration files using official NPM package
relay-config. Configuration files work only when you install this package. Relay Config relies on cosmiconfig to do its bidding. It’s configured to load from:
- or a
It accepts all the same configuration as the CLI does. Additionally, when using the
relay.config.js file, a configuration entry like the language plugin also accepts an actual function:
In the future, other entries such as
customScalars could also be configured as such and allow for projec specific setup.
GraphQL types without ID field
Ever wondered how are GraphQL types being stored inside Relay Store when the types doesn't have globally unique
ID! according to GraphQL specification? Here is an example of 2 identical stores with and without the ID: https://gist.github.com/mrtnzlml/e77315a6879ce8de26fe2a164872be09
Basically, Relay will try to use the
ID field when available (preferable). However, when it's not available, it will construct some unique key which represents the record correctly. Here is an example of the record with ID:
And here is it without the
As you can see, the ID is composed of the query itself + the path. Moreover, there are also GraphQL arguments which ensures you will always get the correct record (forementioned Relay consistency).
In general we're planning to move away from the QueryRenderer/useQuery pattern, which we're referring to as "fetch-on-render". This design makes behavior unpredictable (rendering can happen arbitrarily due to changes in parent components, suspense can cause re-renders and doesn't guarantee cleanup). The alternative is "fetch-then-render" - perform your data-fetching based on some event (user interaction, navigation, timer, app initialization) and then consume that result during render. Then "how do i refetch?" has the same answer as "how do i fetch?". Expect to see more API changes in this direction.
@defer directive is not really ready in GraphQL world (no matter what framework) but there is a different solution which you can use today. All you need is a refetch container and
@inline directive. Let's say you want to fetch "note" lazily for some reason. Simply wrap the component into
createRefetchContainer instead of
createFragmentContainer and fetch some parts of the fragment conditionally like so:
Now, the only thing you have to do is to send the refetch query on mount and you are done:
Custom Relay Compiler
Most of the people are OK with the default OSS version of Relay Compiler. However, it can be sometimes beneficial to write your own Relay Compiler in order to achieve some advanced features (custom behavior or persisting queries to your server for example). Facebook also uses internally their own Relay Compiler implementation. Here is one example of "why" (source: https://github.com/facebook/relay/commit/f1e2e79462d593d73efb80727bc5dd56b1c43cf6#commitcomment-36337550).
The default config generates the flow types inline in the generated files, so something like:
This can in some cases introduce a bunch of noise if the generated files are checked in and the schema changes frequently. For that purpose, we instead generate something like:
Doing that in OSS as well would increase the number of generated files and also add the question of where to put these files and how to import them.
Update: https://github.com/facebook/relay/commit/7a798d16abd08a930a3b249c26697cf92c269639 (
Please read this: https://github.com/facebook/relay/issues/3052
Relay is able to recover completely missing fields in the response. You can use this knowledge to optimize JSON response from the server. Let's say this is our incoming payload from the server:
Traditionally, server would return something like this in case of failure (or just missing data):
But it's not necessary to send the nullable fields at all. Afterall, server knows what fields were requested.
RelayResponseNormalizer by default recovers from this state so you can send response like this from the server (see the missing names):
Relay will show you this warning in this console (dev mode only):
Warning: RelayResponseNormalizer(): Payload did not contain a value for field
name: name. Check that you are parsing with the same query that was used to fetch the payload.
refetch is called and the
refetchQuery is executed, Relay doesn't actually use the result of the query to re-render the component. All it does is normalize the payload into the store and fire any relevant subscriptions. This means that if the fetched data is unrelated to the data that the mounted container is subscribed to (e.g. using a totally different node id that doesn't have any data overlaps), then the component won't re-render.
Refetch containers are only really meant to be used when you are changing variables in the component fragment. If you don't want or need to include variables in the fragment, you could go one level up and set new variables directly in the QueryRenderer (using props or state).
You can override default behavior of unhandled errors when using Relay Observable:
Common Relay mistakes (from user perspective)
- users are not using fragments correctly (data-masking misunderstanding, types generation misunderstanding)
- incorrect environment imports (not using the right Environment instance)
^^ this should however be unnecessary since the correct solution is to decompose
There is another plot twist into the type generation (Joe Savona on why sometimes the types are a bit weird, especially around interfaces):
The Flow types are reflecting the data that Relay will actually provide at runtime. We don't encode which types implement which interfaces (even if we encoded it at build time, it could still change on the server in the meantime), so we always read into fragment spreads on abstract types. If the object implements the interface and we've fetched those fields they'll be there, but otherwise they could be missing entirely. Generating a disjoint union wouldn't reflect the actual values RelayReader would produce.
That being said, it would be nice to have more precise output types, and it's something we're investigating.
Common Relay errors explained
Relay does not allow
__typename field on Query, Mutation or Subscription
Rather than special-case the representation of the root of the graph, Relay generates a "client" record to represent the Query and mutation/subscription objects. Like all other record instances those records have a typename, but we special-case this typename to be 'ROOT'. Querying for the typename field would overwrite this value with the actual typename (e.g. Query or whatever you call it in your schema), which messes with a few invariants. It's on our wishlist to make the root record a bit less special, but in practice we couldn't think of a reason to query __typename on the root so we just disallow it for now.
A nice workaround was mentioned here: https://github.com/facebook/relay/issues/2471#issuecomment-624238501
(instead of just
Current Relay struggles
These are things which are somehow difficult to achieve with current Relay abilities. It's almost like a wishlist:
readInlineDatadoesn't support variables (throws error "Variables are not yet supported inside @inline fragments.")
- How to figure out request is in-flight when using fetch policy
store-and-network(to show some spinner)?
- How to access fields with some complicated args (without actually knowing all the possible args combinations in advance).