The main aim of the PR is:
1. To set up a module structure for 'remote-schemas' package.
2. Move parts by the remote schema codebase into the new module structure to validate it.
## Notes to the reviewer
Why a PR with large-ish diff?
1. We've been making progress on the MM project but we don't yet know long it is going to take us to get to the first milestone. To understand this better, we need to figure out the unknowns as soon as possible. Hence I've taken a stab at the first two items in the [end-state](https://gist.github.com/0x777/ca2bdc4284d21c3eec153b51dea255c9) document to figure out the unknowns. Unsurprisingly, there are a bunch of issues that we haven't discussed earlier. These are documented in the 'open questions' section.
1. The diff is large but that is only code moved around and I've added a section that documents how things are moved. In addition, there are fair number of PR comments to help with the review process.
## Changes in the PR
### Module structure
Sets up the module structure as follows:
```
Hasura/
RemoteSchema/
Metadata/
Types.hs
SchemaCache/
Types.hs
Permission.hs
RemoteRelationship.hs
Build.hs
MetadataAPI/
Types.hs
Execute.hs
```
### 1. Types representing metadata are moved
Types that capture metadata information (currently scattered across several RQL modules) are moved into `Hasura.RemoteSchema.Metadata.Types`.
- This new module only depends on very 'core' modules such as
`Hasura.Session` for the notion of roles and `Hasura.Incremental` for `Cacheable` typeclass.
- The requirement on database modules is avoided by generalizing the remote schemas metadata to accept an arbitrary 'r' for a remote relationship
definition.
### 2. SchemaCache related types and build logic have been moved
Types that represent remote schemas information in SchemaCache are moved into `Hasura.RemoteSchema.SchemaCache.Types`.
Similar to `H.RS.Metadata.Types`, this module depends on 'core' modules except for `Hasura.GraphQL.Parser.Variable`. It has something to do with remote relationships but I haven't spent time looking into it. The validation of 'remote relationships to remote schema' is also something that needs to be looked at.
Rips out the logic that builds remote schema's SchemaCache information from the monolithic `buildSchemaCacheRule` and moves it into `Hasura.RemoteSchema.SchemaCache.Build`. Further, the `.SchemaCache.Permission` and `.SchemaCache.RemoteRelationship` have been created from existing modules that capture schema cache building logic for those two components.
This was a fair amount of work. On main, currently remote schema's SchemaCache information is built in two phases - in the first phase, 'permissions' and 'remote relationships' are ignored and in the second phase they are filled in.
While remote relationships can only be resolved after partially resolving sources and other remote schemas, the same isn't true for permissions. Further, most of the work that is done to resolve remote relationships can be moved to the first phase so that the second phase can be a very simple traversal.
This is the approach that was taken - resolve permissions and as much as remote relationships information in the first phase.
### 3. Metadata APIs related types and build logic have been moved
The types that represent remote schema related metadata APIs and the execution logic have been moved to `Hasura.RemoteSchema.MetadataAPI.Types` and `.Execute` modules respectively.
## Open questions:
1. `Hasura.RemoteSchema.Metadata.Types` is so called because I was hoping that all of the metadata related APIs of remote schema can be brought in at `Hasura.RemoteSchema.Metadata.API`. However, as metadata APIs depended on functions from `SchemaCache` module (see [1](ceba6d6226/server/src-lib/Hasura/RQL/DDL/RemoteSchema.hs (L55)) and [2](ceba6d6226/server/src-lib/Hasura/RQL/DDL/RemoteSchema.hs (L91)), it made more sense to create a separate top-level module for `MetadataAPI`s.
Maybe we can just have `Hasura.RemoteSchema.Metadata` and get rid of the extra nesting or have `Hasura.RemoteSchema.Metadata.{Core,Permission,RemoteRelationship}` if we want to break them down further.
1. `buildRemoteSchemas` in `H.RS.SchemaCache.Build` has the following type:
```haskell
buildRemoteSchemas ::
( ArrowChoice arr,
Inc.ArrowDistribute arr,
ArrowWriter (Seq CollectedInfo) arr,
Inc.ArrowCache m arr,
MonadIO m,
HasHttpManagerM m,
Inc.Cacheable remoteRelationshipDefinition,
ToJSON remoteRelationshipDefinition,
MonadError QErr m
) =>
Env.Environment ->
( (Inc.Dependency (HashMap RemoteSchemaName Inc.InvalidationKey), OrderedRoles),
[RemoteSchemaMetadataG remoteRelationshipDefinition]
)
`arr` HashMap RemoteSchemaName (PartiallyResolvedRemoteSchemaCtxG remoteRelationshipDefinition, MetadataObject)
```
Note the dependence on `CollectedInfo` which is defined as
```haskell
data CollectedInfo
= CIInconsistency InconsistentMetadata
| CIDependency
MetadataObject
-- ^ for error reporting on missing dependencies
SchemaObjId
SchemaDependency
deriving (Eq)
```
this pretty much means that remote schemas is dependent on types from databases, actions, ....
How do we fix this? Maybe introduce a typeclass such as `ArrowCollectRemoteSchemaDependencies` which is defined in `Hasura.RemoteSchema` and then implemented in graphql-engine?
1. The dependency on `buildSchemaCacheFor` in `.MetadataAPI.Execute` which has the following signature:
```haskell
buildSchemaCacheFor ::
(QErrM m, CacheRWM m, MetadataM m) =>
MetadataObjId ->
MetadataModifier ->
```
This can be easily resolved if we restrict what the metadata APIs are allowed to do. Currently, they operate in an unfettered access to modify SchemaCache (the `CacheRWM` constraint):
```haskell
runAddRemoteSchema ::
( QErrM m,
CacheRWM m,
MonadIO m,
HasHttpManagerM m,
MetadataM m,
Tracing.MonadTrace m
) =>
Env.Environment ->
AddRemoteSchemaQuery ->
m EncJSON
```
This should instead be changed to restrict remote schema APIs to only modify remote schema metadata (but has access to the remote schemas part of the schema cache), this dependency is completely removed.
```haskell
runAddRemoteSchema ::
( QErrM m,
MonadIO m,
HasHttpManagerM m,
MonadReader RemoteSchemasSchemaCache m,
MonadState RemoteSchemaMetadata m,
Tracing.MonadTrace m
) =>
Env.Environment ->
AddRemoteSchemaQuery ->
m RemoteSchemeMetadataObjId
```
The idea is that the core graphql-engine would call these functions and then call
`buildSchemaCacheFor`.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/6291
GitOrigin-RevId: 51357148c6404afe70219afa71bd1d59bdf4ffc6
This PR is the result of running the following commands:
```bash
$ git grep -l '".* : "' -- '*.hs' | xargs sed -i -E 's/(".*) : "/\1: "/'
$ scripts/dev.sh test --integration --accept
```
Also manually fixed a few tests and docs
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/6148
GitOrigin-RevId: cf8b87605d41d9ce86613a41ac5fd18691f5a641
The intent is to generalize `columnParser` to the point where it is the same across all backends, and then remove the interface in favor of a single implementation.
This extracts out `enumParser` and `possiblyNullable` as the two main areas that differ across backends. We may split `possiblyNullable` further so that we can extract some of that logic out into a common function too.
With these changes, the various `columnParser` implementations become semantically equivalent. They still do different things, and so reconciling them will require further changes.
Co-Authored-By: Antoine Leblanc <antoine@hasura.io>
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5841
GitOrigin-RevId: eec1770931eed5d72da70c97d7d0f00e33fa15d2
### Description
This PR attempts to fix several issues with source customization as it relates to remote relationships. There were several issues regarding casing: at the relationship border, we didn't properly set the target source's case, we didn't have access to the list of supported features to decide whether the feature was allowed or not, and we didn't have access to the global default.
However, all of that information is available when we build the schema cache, as we do resolve the case of some elements such as function names: we can therefore resolve source information at the same time, and simplify both the root of the schema and the remote relationship border.
To do this, this PR introduces a new type, `ResolvedSourceCustomization`, to be used in the Schema Cache, as opposed to the metadata's `SourceCustomization`, following a pattern established by a lot of other types.
### Remaining work and open questions
One major point of confusion: it seems to me that we didn't set the case at all across remote relationships, which would suggest we would use the case of the LHS source across the subset of the RHS one that is accessible through the remote relationship, which would in turn "corrupt" the parser cache and might result in the wrong case being used for that source later on. Is that assesment correct, and was I right to fix it?
Another one is that we seem not to be using the local case of the RHS to name the field in an object relationship; unless I'm mistaken we only use it for array relationships? Is that intentional?
This PR is also missing tests that would show-case the difference, and a changelog entry. To my knowledge, all the tests of this feature are in the python test suite; this could be the opportunity to move them to the hspec suite, but this might be a considerable amount of work?
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5619
GitOrigin-RevId: 51a81b713a74575e82d9f96b51633f158ce3a47b
### Description
This PR changes all the schema code to operate in a specific `SchemaT` monad, rather than in an arbitrary `m` monad. `SchemaT` is intended to be used opaquely with `runSourceSchema` and `runRemoteSchema`. The main goal of this is to allow a different reader context per part of the schema: this PR also minimizes the contexts. This means that we no longer require `SchemaOptions` when building remote schemas' schema, and this PR therefore removes a lot of dummy / placeholder values accordingly.
### Performance and stacking
This PR has been through several iterations. #5339 was the original version, that accomplished the same thing by stacking readers on top of the stack at every remote relationship boundary. This raised performance concerns, and @0x777 confirmed with an ad-hoc test that in some extreme cases we could see up to a 10% performance impact. This version, while more verbose, allows us to unstack / re-stack the readers, and avoid that problem. #5517 adds a new benchmark set to be able to automatically measure this on every PR.
### Remaining work
- [x] a comment (or perhaps even a Note?) should be added to `SchemaT`
- [x] we probably want for #5517 to be merged first so that we can confirm the lack of performance penalty
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5458
GitOrigin-RevId: e06b83d90da475f745b838f1fd8f8b4d9d3f4b10
### Description
This PR is a first step towards having a dedicated reader context per schema block. It adds the required Reader instance, and switches from a `SchemaT ReaderT` stack to a `ReaderT SchemaT` stack. Furthermore, it cleans up / harmonizes some of the top-level schema building functions.
Sources and remotes are now built each within their own run of `runReaderT`: for now, the reader context is the same in both cases, meaning no special care is required at the boundary of remote relationships.
Actions are explicitly run with the source context for now; we could envision creating a third and distinct context for them.
This PR is expected to be a no-op.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5300
GitOrigin-RevId: a014e5b3504eb4ef740c820d305d6d2695f622f7
Followup to hasura/graphql-engine-mono#4713.
The `memoizeOn` method, part of `MonadSchema`, originally had the following type:
```haskell
memoizeOn
:: (HasCallStack, Ord a, Typeable a, Typeable b, Typeable k)
=> TH.Name
-> a
-> m (Parser k n b)
-> m (Parser k n b)
```
The reason for operating on `Parser`s specifically was that the `MonadSchema` effect would additionally initialize certain `Unique` values, which appear (nested in) the type of `Parser`.
hasura/graphql-engine-mono#518 changed the type of `memoizeOn`, to additionally allow memoizing `FieldParser`s. These also contained a `Unique` value, which was similarly initialized by the `MonadSchema` effect. The new type of `memoizeOn` was as follows:
```haskell
memoizeOn
:: forall p d a b
. (HasCallStack, HasDefinition (p n b) d, Ord a, Typeable p, Typeable a, Typeable b)
=> TH.Name
-> a
-> m (p n b)
-> m (p n b)
```
Note the type `p n b` of the value being memoized: by choosing `p` to be either `Parser k` or `FieldParser`, both can be memoized. Also note the new `HasDefinition (p n b) d` constraint, which provided a `Lens` for accessing the `Unique` value to be initialized.
A quick simplification is that the `HasCallStack` constraint has never been used by any code. This was realized in hasura/graphql-engine-mono#4713, by removing that constraint.
hasura/graphql-engine-mono#2980 removed the `Unique` value from our GraphQL-related types entirely, as their original purpose was never truly realized. One part of removing `Unique` consisted of dropping the `HasDefinition (p n b) d` constraint from `memoizeOn`.
What I didn't realize at the time was that this meant that the type of `memoizeOn` could be generalized and simplified much further. This PR finally implements that generalization. The new type is as follows:
```haskell
memoizeOn ::
forall a p.
(Ord a, Typeable a, Typeable p) =>
TH.Name ->
a ->
m p ->
m p
```
This change has a couple of consequences.
1. While constructing the schema, we often output `Maybe (Parser ...)`, to model that the existence of certain pieces of GraphQL schema sometimes depends on the permissions that a certain role has. The previous versions of `memoizeOn` were not able to handle this, as the only thing they could memoize was fully-defined (if not yet fully-evaluated) `(Field)Parser`s. This much more general API _would_ allow memoizing `Maybe (Parser ...)`s. However, we probably have to be continue being cautious with this: if we blindly memoize all `Maybe (Parser ...)`s, the resulting code may never be able to decide whether the value is `Just` or `Nothing` - i.e. it never commits to the existence-or-not of a GraphQL schema fragment. This would manifest as a non-well-founded knot tying, and this would get reported as an error by the implementation of `memoizeOn`.
tl;dr: This generalization _technically_ allows for memoizing `Maybe` values, but we probably still want to avoid doing so.
For this reason, the PR adds a specialized version of `memoizeOn` to `Hasura.GraphQL.Schema.Parser`.
2. There is no longer any need to connect the `MonadSchema` knot-tying effect with the `MonadParse` effect. In fact, after this PR, the `memoizeOn` method is completely GraphQL-agnostic, and so we implement hasura/graphql-engine-mono#4726, separating `memoizeOn` from `MonadParse` entirely - `memoizeOn` can be defined and implemented as a general Haskell typeclass method.
Since `MonadSchema` has been made into a single-type-parameter type class, it has been renamed to something more general, namely `MonadMemoize`. Its only task is to memoize arbitrary `Typeable p` objects under a combined key consisting of a `TH.Name` and a `Typeable a`.
Also for this reason, the new `MonadMemoize` has been moved to the more general `Control.Monad.Memoize`.
3. After this change, it's somewhat clearer what `memoizeOn` does: it memoizes an arbitrary value of a `Typeable` type. The only thing that needs to be understood in its implementation is how the manual blackholing works. There is no more semantic interaction with _any_ GraphQL code.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/4725
Co-authored-by: Daniel Harvey <4729125+danieljharvey@users.noreply.github.com>
GitOrigin-RevId: 089fa2e82c2ce29da76850e994eabb1e261f9c92
### Description
By definition, root fields are at the root of the schema: only functions that craft root fields need to know about how to customize the name of root fields. However, the presence of `Has MkRootFieldName` in `MonadBuildSchemaBase` meant that the entirety of the schema building code was implicitly aware of / capable of altering root field names.
This PR removes this constraint, and moves it to the functions that do craft root fields. This has several upsides:
- it makes it more explicit where root fields are being crafted
- it prevents functions that should not use this from mistakenly applying it to non-root fields
- it simplifies the shared schema context
### Future work
- can we maybe pass this as an argument, instead of making it a required part of the context?
- ~~AFAICT, we only ever use `mempty` for it: is this actually dead code that we should actually just remove altogether?~~
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5235
GitOrigin-RevId: 4268751f3ab87ae8e03b6fe9e1efa1b096200027
### Description
I am not 100% sure about this PR; while I think the code is better this way, I'm willing to be convinced otherwise.
In short, this PR moves the `RoleName` field into the `SchemaContext`, instead of being a nebulous `Has RoleName` constraint on the reader monad. The major upside of this is that it makes it an explicit named field, rather than something that must be given as part of a tuple of arguments when calling `runReader`.
However, the downside is that it breaks the helper permissions functions of `Schema.Table`, which relied on `Has RoleName r`. This PR makes the choice of passing the role name explicitly to all of those functions, which in turn means first explicitly fetching the role name in a lot of places. It makes it more explicit when a schema building block relies on the role name, but is a bit verbose...
### Alternatives
Some alternatives worth considering:
- attempting something like `Has context r, Has RoleName context`, which would allow them to be independent from the context but still fetch the role name from the reader, but might require type annotations to not be ambiguous
- keeping the permission functions the same, with `Has RoleName r`, and introducing a bunch of newtypes instead of using tuples to explicitly implement all the required `Has` instances
- changing the permission functions to `Has SchemaContext r`, since they are functions used only to build the schema, and therefore may be allowed to be tied to the context.
What do y'all think?
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5073
GitOrigin-RevId: 8fd09fafb54905a4d115ef30842d35da0c3db5d2
In the process of decoupling the schema parsers from the GraphQL Engine, we need to remove dependencies on `Hasura.Base.Error`.
First of all, we have avoided using `QErr` in schema parsers code, instead returning a more appropriate data type which can be converted to a `Hasura.Base.Error.QErr` later.
Secondly, we create a new `ParseErrorCode` type to represent parse failure types, which are then converted to a `Hasura.Base.Error.Code` later.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5181
GitOrigin-RevId: 8655e26adb1e7d5e3d552c77a8a403f987b53467
This introduces an `ErrorMessage` newtype which wraps `Text` in a manner which is designed to be easy to construct, and difficult to deconstruct.
It provides functionality similar to `Data.Text.Extended`, but designed _only_ for error messages. Error messages are constructed through `fromString`, concatenation, or the `toErrorValue` function, which is designed to be overridden for all meaningful domain types that might show up in an error message. Notably, there are not and should never be instances of `ToErrorValue` for `String`, `Text`, `Int`, etc. This is so that we correctly represent the value in a way that is specific to its type. For example, all `Name` values (from the _graphql-parser-hs_ library) are single-quoted now; no exceptions.
I have mostly had to add `instance ToErrorValue` for various backend types (and also add newtypes where necessary). Some of these are not strictly necessary for this changeset, as I had bigger aspirations when I started. These aspirations have been tempered by trying and failing twice.
As such, in this changeset, I have started by introducing this type to the `parseError` and `parseErrorWith` functions. In the future, I would like to extend this to the `QErr` record and the various `throwError` functions, but this is a much larger task and should probably be done in stages.
For now, `toErrorMessage` and `fromErrorMessage` are provided for conversion to and from `Text`, but the intent is to stop exporting these once all error messages are converted to the new type.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5018
GitOrigin-RevId: 84b37e238992e4312255a87ca44f41af65e2d89a
### Description
A trivial PR, extracted out of #4936, that removes remote schema permissions from the schema context, as they are only ever used at the top level: whether or not we need to use remote schema permissions is not something that impacts _how_ we build the schema, but whether some parts of the schema should be built at all, and therefore doesn't need to be accessible throughout the build process.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/5050
GitOrigin-RevId: 734673370393d5640ad753222982baf2698f6d8f
This moves `MkTypename` and `NamingCase` into their own modules, with the intent of reducing the scope of the schema parsers code, and trying to reduce imports of large modules when small ones will do.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/4978
GitOrigin-RevId: 19541257fe010035390f6183a4eaa37bae0d3ca1
Earlier, if the `select` root field had a custom root field set, the same custom root field was then used for the streaming subscription root field as well. This leads to duplicate root fields being generated in the `subscription_root`.
This PR fixes that. It provides a way to customize the streaming subscription root field and not use the `select` root field's custom root field name for the streaming subscription root field.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/4967
Co-authored-by: Anon Ray <616387+ecthiender@users.noreply.github.com>
GitOrigin-RevId: 54e74ce97561b0e5cfdfc60d1ca340aaebecf7d4
This reduces the usage of "utils" modules in the parsers code, especially those that are simply re-exported from elsewhere, to facilitate extracting the parsers code into its own library.
It mostly inlines the imports that are re-exported from `Hasura.Prelude` and `Data.Parser.JSONPath`. It also removes references to `Data.*.Extended` modules. When necessary, it re-implements the functionality (which is typically trivial).
It does not tackle all external dependencies. I observed the following that will take more work:
- `Data.GADT.Compare.Extended`
- `Data.Text.Extended`
- `Hasura.Base.Error`
- `Hasura.RQL.Types.Common`
- `Hasura.Server.Utils`
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/4964
GitOrigin-RevId: 54ad3c1b7a31f13e34340ebe9fcc36d0ad57b8bd
The code that builds the GraphQL schema, and `buildGQLContext` in particular, is partial: not every value of `(ServerConfigCtx, GraphQLQueryType, SourceCache, HashMap RemoteSchemaName (RemoteSchemaCtx, MetadataObject), ActionCache, AnnotatedCustomTypes)` results in a valid GraphQL schema. When it fails, we want to be able to return better error messages than we currently do.
The key thing that is missing is a way to trace back GraphQL type information to their origin from the Hasura metadata. Currently, we have a number of correctness checks of our GraphQL schema. But these correctness checks only have access to pure GraphQL type information, and hence can only report errors in terms of that. Possibly the worst is the "conflicting definitions" error, which, in practice, can only be debugged by Hasura engineers. This is terrible DX for customers.
This PR allows us to print better error messages, by adding a field to the `Definition` type that traces the GraphQL type to its origin in the metadata. So the idea is simple: just add `MetadataObjId`, or `Maybe` that, or some other sum type of that, to `Definition`.
However, we want to avoid having to import a `Hasura.RQL` module from `Hasura.GraphQL.Parser`. So we instead define this additional field of `Definition` through a new type parameter, which is threaded through in `Hasura.GraphQL.Parser`. We then define type synonyms in `Hasura.GraphQL.Schema.Parser` that fill in this type parameter, so that it is not visible for the majority of the codebase.
The idea of associating metadata information to `Definition`s really comes to fruition when combined with hasura/graphql-engine-mono#4517. Their combination would allow us to use the API of fatal errors (just like the current `MonadError QErr`) to report _inconsistencies_ in the metadata. Such inconsistencies are then _automatically_ ignored. So no ad-hoc decisions need to be made on how to cut out inconsistent metadata from the GraphQL schema. This will allow us to report much better errors, as well as improve the likelihood of a successful HGE startup.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/4770
Co-authored-by: Samir Talwar <47582+SamirTalwar@users.noreply.github.com>
GitOrigin-RevId: 728402b0cae83ae8e83463a826ceeb609001acae
### Description
This PR removes the need for the `SourceCache` when building the schema for the actions. To do so, it changes the way we represent custom types in the source cache. Instead of trying to reuse the same `ObjectTypeDefinition` and `TypeRelationship`. we now have separate `AnnotatedObjectType` and `AnnotatedRelationship`. When building them, at schema cache building time, we persist all the relevant source information, so that it's all available at schema building time.
This PR makes no attempt at re-using `RemoteRelationship` primitives, to avoid having to change the way async action queries are executed, and to avoid having to make complicated changes to how we parse and represent those relationships.
PR-URL: https://github.com/hasura/graphql-engine-mono/pull/4813
GitOrigin-RevId: 3cc65c5a043c8d3da5f7214eed40c558c4349327
