graphql-engine/server/src-lib/Hasura/Backends/Postgres/Instances/Execute.hs

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

425 lines
16 KiB
Haskell
Raw Normal View History

{-# LANGUAGE UndecidableInstances #-}
{-# OPTIONS_GHC -fno-warn-orphans #-}
module Hasura.Backends.Postgres.Instances.Execute
( PreparedSql (..),
)
where
import Control.Monad.Trans.Control qualified as MT
import Data.Aeson qualified as J
import Data.HashMap.Strict qualified as Map
import Data.HashMap.Strict.InsOrd qualified as OMap
import Data.IntMap qualified as IntMap
import Data.Sequence qualified as Seq
import Database.PG.Query qualified as Q
import Hasura.Backends.Postgres.Connection (runTx)
import Hasura.Backends.Postgres.Execute.Insert (convertToSQLTransaction)
import Hasura.Backends.Postgres.Execute.LiveQuery qualified as PGL
import Hasura.Backends.Postgres.Execute.Mutation qualified as PGE
import Hasura.Backends.Postgres.Execute.Prepare
( PlanningSt (..),
PrepArgMap,
initPlanningSt,
prepareWithPlan,
prepareWithoutPlan,
resolveUnpreparedValue,
withUserVars,
)
import Hasura.Backends.Postgres.Execute.Types (PGSourceConfig (..))
import Hasura.Backends.Postgres.SQL.DML qualified as S
import Hasura.Backends.Postgres.SQL.Types qualified as PG
import Hasura.Backends.Postgres.SQL.Value qualified as PG
import Hasura.Backends.Postgres.Translate.Select (PostgresAnnotatedFieldJSON)
import Hasura.Backends.Postgres.Translate.Select qualified as DS
import Hasura.Backends.Postgres.Types.Update
import Hasura.Base.Error (QErr)
import Hasura.EncJSON (EncJSON, encJFromJValue)
import Hasura.GraphQL.Execute.Backend
( BackendExecute (..),
DBStepInfo (..),
ExplainPlan (..),
convertRemoteSourceRelationship,
)
import Hasura.GraphQL.Execute.LiveQuery.Plan
( LiveQueryPlan (..),
LiveQueryPlanExplanation (..),
ParameterizedLiveQueryPlan (..),
mkCohortVariables,
newCohortId,
)
import Hasura.GraphQL.Namespace
( RootFieldAlias (..),
RootFieldMap,
)
import Hasura.GraphQL.Parser (UnpreparedValue (..))
import Hasura.Prelude
import Hasura.QueryTags
( QueryTagsComment (..),
emptyQueryTagsComment,
)
import Hasura.RQL.DML.Internal (dmlTxErrorHandler)
import Hasura.RQL.IR (MutationDB (..), QueryDB (..))
import Hasura.RQL.IR.Delete qualified as IR
import Hasura.RQL.IR.Insert qualified as IR
import Hasura.RQL.IR.Returning qualified as IR
import Hasura.RQL.IR.Select qualified as IR
import Hasura.RQL.IR.Update qualified as IR
import Hasura.RQL.Types
( Backend (..),
BackendType (Postgres),
FieldName,
JsonAggSelect (..),
SourceName,
getFieldNameTxt,
liftTx,
)
import Hasura.RQL.Types.Column (ColumnType (..), ColumnValue (..))
import Hasura.SQL.AnyBackend qualified as AB
import Hasura.Session (UserInfo (..))
import Hasura.Tracing qualified as Tracing
import Language.GraphQL.Draft.Syntax qualified as G
data PreparedSql = PreparedSql
{ _psQuery :: !Q.Query,
_psPrepArgs :: !PrepArgMap
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
}
instance
( Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind
) =>
BackendExecute ('Postgres pgKind)
where
type PreparedQuery ('Postgres pgKind) = PreparedSql
type MultiplexedQuery ('Postgres pgKind) = PGL.MultiplexedQuery
type ExecutionMonad ('Postgres pgKind) = Tracing.TraceT (Q.TxET QErr IO)
mkDBQueryPlan = pgDBQueryPlan
mkDBMutationPlan = pgDBMutationPlan
mkDBSubscriptionPlan = pgDBSubscriptionPlan
mkDBQueryExplain = pgDBQueryExplain
mkLiveQueryExplain = pgDBLiveQueryExplain
mkDBRemoteRelationshipPlan = pgDBRemoteRelationshipPlan
-- query
pgDBQueryPlan ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind,
MonadReader QueryTagsComment m
) =>
UserInfo ->
SourceName ->
SourceConfig ('Postgres pgKind) ->
QueryDB ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind)) ->
m (DBStepInfo ('Postgres pgKind))
pgDBQueryPlan userInfo sourceName sourceConfig qrf = do
(preparedQuery, PlanningSt _ _ planVals) <-
flip runStateT initPlanningSt $ traverse (prepareWithPlan userInfo) qrf
queryTagsComment <- ask
let preparedSQLWithQueryTags = appendPreparedSQLWithQueryTags (irToRootFieldPlan planVals preparedQuery) queryTagsComment
let (action, preparedSQL) = mkCurPlanTx userInfo preparedSQLWithQueryTags
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
pure $ DBStepInfo @('Postgres pgKind) sourceName sourceConfig preparedSQL action
pgDBQueryExplain ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind
) =>
RootFieldAlias ->
UserInfo ->
SourceName ->
SourceConfig ('Postgres pgKind) ->
QueryDB ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind)) ->
m (AB.AnyBackend DBStepInfo)
pgDBQueryExplain fieldName userInfo sourceName sourceConfig rootSelection = do
preparedQuery <- traverse (resolveUnpreparedValue userInfo) rootSelection
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
let PreparedSql querySQL _ = irToRootFieldPlan mempty preparedQuery
textSQL = Q.getQueryText querySQL
-- CAREFUL!: an `EXPLAIN ANALYZE` here would actually *execute* this
-- query, maybe resulting in privilege escalation:
withExplain = "EXPLAIN (FORMAT TEXT) " <> textSQL
let action =
liftTx $
Q.listQE dmlTxErrorHandler (Q.fromText withExplain) () True <&> \planList ->
encJFromJValue $ ExplainPlan fieldName (Just textSQL) (Just $ map runIdentity planList)
pure $
AB.mkAnyBackend $
DBStepInfo @('Postgres pgKind) sourceName sourceConfig Nothing action
pgDBLiveQueryExplain ::
( MonadError QErr m,
MonadIO m,
MT.MonadBaseControl IO m
) =>
LiveQueryPlan ('Postgres pgKind) (MultiplexedQuery ('Postgres pgKind)) ->
m LiveQueryPlanExplanation
pgDBLiveQueryExplain plan = do
let parameterizedPlan = _lqpParameterizedPlan plan
pgExecCtx = _pscExecCtx $ _lqpSourceConfig plan
queryText = Q.getQueryText . PGL.unMultiplexedQuery $ _plqpQuery parameterizedPlan
-- CAREFUL!: an `EXPLAIN ANALYZE` here would actually *execute* this
-- query, maybe resulting in privilege escalation:
explainQuery = Q.fromText $ "EXPLAIN (FORMAT TEXT) " <> queryText
cohortId <- newCohortId
explanationLines <-
liftEitherM $
runExceptT $
runTx pgExecCtx Q.ReadOnly $
map runIdentity <$> PGL.executeQuery explainQuery [(cohortId, _lqpVariables plan)]
pure $ LiveQueryPlanExplanation queryText explanationLines $ _lqpVariables plan
-- mutation
convertDelete ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind
) =>
UserInfo ->
IR.AnnDelG ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind)) ->
Bool ->
QueryTagsComment ->
m (Tracing.TraceT (Q.TxET QErr IO) EncJSON)
convertDelete userInfo deleteOperation stringifyNum queryTags = do
preparedDelete <- traverse (prepareWithoutPlan userInfo) deleteOperation
pure $ flip runReaderT queryTags $ PGE.execDeleteQuery stringifyNum userInfo (preparedDelete, Seq.empty)
convertUpdate ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind
) =>
UserInfo ->
IR.AnnotatedUpdateG ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind)) ->
Bool ->
QueryTagsComment ->
m (Tracing.TraceT (Q.TxET QErr IO) EncJSON)
convertUpdate userInfo updateOperation stringifyNum queryTags = do
preparedUpdate <- traverse (prepareWithoutPlan userInfo) updateOperation
if null $ updateOperations . IR._auBackend $ updateOperation
then pure $ pure $ IR.buildEmptyMutResp $ IR._auOutput preparedUpdate
else
pure $
flip runReaderT queryTags $
PGE.execUpdateQuery stringifyNum userInfo (preparedUpdate, Seq.empty)
convertInsert ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind
) =>
UserInfo ->
IR.AnnInsert ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind)) ->
Bool ->
QueryTagsComment ->
m (Tracing.TraceT (Q.TxET QErr IO) EncJSON)
convertInsert userInfo insertOperation stringifyNum queryTags = do
preparedInsert <- traverse (prepareWithoutPlan userInfo) insertOperation
pure $ flip runReaderT queryTags $ convertToSQLTransaction preparedInsert userInfo Seq.empty stringifyNum
-- | A pared-down version of 'Query.convertQuerySelSet', for use in execution of
-- special case of SQL function mutations (see 'MDBFunction').
convertFunction ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind
) =>
UserInfo ->
JsonAggSelect ->
-- | VOLATILE function as 'SelectExp'
IR.AnnSimpleSelectG ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind)) ->
-- | Query Tags
QueryTagsComment ->
m (Tracing.TraceT (Q.TxET QErr IO) EncJSON)
convertFunction userInfo jsonAggSelect unpreparedQuery queryTags = do
-- Transform the RQL AST into a prepared SQL query
(preparedQuery, PlanningSt _ _ planVals) <-
flip runStateT initPlanningSt $
traverse (prepareWithPlan userInfo) unpreparedQuery
let queryResultFn =
case jsonAggSelect of
JASMultipleRows -> QDBMultipleRows
JASSingleObject -> QDBSingleRow
let preparedSQLWithQueryTags = appendPreparedSQLWithQueryTags (irToRootFieldPlan planVals $ queryResultFn preparedQuery) queryTags
pure
$! fst
$ mkCurPlanTx userInfo preparedSQLWithQueryTags -- forget (Maybe PreparedSql)
pgDBMutationPlan ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind,
MonadReader QueryTagsComment m
) =>
UserInfo ->
Bool ->
SourceName ->
SourceConfig ('Postgres pgKind) ->
MutationDB ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind)) ->
m (DBStepInfo ('Postgres pgKind))
pgDBMutationPlan userInfo stringifyNum sourceName sourceConfig mrf = do
mutationQueryTagsComment <- ask
go <$> case mrf of
MDBInsert s -> convertInsert userInfo s stringifyNum mutationQueryTagsComment
MDBUpdate s -> convertUpdate userInfo s stringifyNum mutationQueryTagsComment
MDBDelete s -> convertDelete userInfo s stringifyNum mutationQueryTagsComment
MDBFunction returnsSet s -> convertFunction userInfo returnsSet s mutationQueryTagsComment
where
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
go v = DBStepInfo @('Postgres pgKind) sourceName sourceConfig Nothing v
-- subscription
pgDBSubscriptionPlan ::
forall pgKind m.
( MonadError QErr m,
MonadIO m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind,
MonadReader QueryTagsComment m
) =>
UserInfo ->
SourceName ->
SourceConfig ('Postgres pgKind) ->
Maybe G.Name ->
RootFieldMap (QueryDB ('Postgres pgKind) Void (UnpreparedValue ('Postgres pgKind))) ->
m (LiveQueryPlan ('Postgres pgKind) (MultiplexedQuery ('Postgres pgKind)))
pgDBSubscriptionPlan userInfo _sourceName sourceConfig namespace unpreparedAST = do
(preparedAST, PGL.QueryParametersInfo {..}) <-
flip runStateT mempty $
for unpreparedAST $ traverse (PGL.resolveMultiplexedValue $ _uiSession userInfo)
mutationQueryTagsComment <- ask
let multiplexedQuery = PGL.mkMultiplexedQuery $ OMap.mapKeys _rfaAlias preparedAST
multiplexedQueryWithQueryTags =
multiplexedQuery {PGL.unMultiplexedQuery = appendSQLWithQueryTags (PGL.unMultiplexedQuery multiplexedQuery) mutationQueryTagsComment}
roleName = _uiRole userInfo
parameterizedPlan = ParameterizedLiveQueryPlan roleName multiplexedQueryWithQueryTags
-- We need to ensure that the values provided for variables are correct according to Postgres.
-- Without this check an invalid value for a variable for one instance of the subscription will
-- take down the entire multiplexed query.
validatedQueryVars <- PGL.validateVariables (_pscExecCtx sourceConfig) _qpiReusableVariableValues
validatedSyntheticVars <- PGL.validateVariables (_pscExecCtx sourceConfig) $ toList _qpiSyntheticVariableValues
-- TODO validatedQueryVars validatedSyntheticVars
let cohortVariables =
mkCohortVariables
_qpiReferencedSessionVariables
(_uiSession userInfo)
validatedQueryVars
validatedSyntheticVars
pure $ LiveQueryPlan parameterizedPlan sourceConfig cohortVariables namespace
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
-- turn the current plan into a transaction
mkCurPlanTx ::
UserInfo ->
PreparedSql ->
(Tracing.TraceT (Q.TxET QErr IO) EncJSON, Maybe PreparedSql)
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
mkCurPlanTx userInfo ps@(PreparedSql q prepMap) =
-- generate the SQL and prepared vars or the bytestring
let args = withUserVars (_uiSession userInfo) prepMap
-- WARNING: this quietly assumes the intmap keys are contiguous
prepArgs = fst <$> IntMap.elems args
in (,Just ps) $
Tracing.trace "Postgres" $ liftTx $ DS.asSingleRowJsonResp q prepArgs
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
-- convert a query from an intermediate representation to... another
irToRootFieldPlan ::
( Backend ('Postgres pgKind),
DS.PostgresAnnotatedFieldJSON pgKind
) =>
PrepArgMap ->
QueryDB ('Postgres pgKind) Void S.SQLExp ->
PreparedSql
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
irToRootFieldPlan prepped = \case
QDBMultipleRows s -> mkPreparedSql (DS.selectQuerySQL JASMultipleRows) s
QDBSingleRow s -> mkPreparedSql (DS.selectQuerySQL JASSingleObject) s
QDBAggregation s -> mkPreparedSql DS.selectAggregateQuerySQL s
QDBConnection s -> mkPreparedSql DS.connectionSelectQuerySQL s
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
where
mkPreparedSql :: (t -> Q.Query) -> t -> PreparedSql
server: support remote relationships on SQL Server and BigQuery (#1497) Remote relationships are now supported on SQL Server and BigQuery. The major change though is the re-architecture of remote join execution logic. Prior to this PR, each backend is responsible for processing the remote relationships that are part of their AST. This is not ideal as there is nothing specific about a remote join's execution that ties it to a backend. The only backend specific part is whether or not the specification of the remote relationship is valid (i.e, we'll need to validate whether the scalars are compatible). The approach now changes to this: 1. Before delegating the AST to the backend, we traverse the AST, collect all the remote joins while modifying the AST to add necessary join fields where needed. 1. Once the remote joins are collected from the AST, the database call is made to fetch the response. The necessary data for the remote join(s) is collected from the database's response and one or more remote schema calls are constructed as necessary. 1. The remote schema calls are then executed and the data from the database and from the remote schemas is joined to produce the final response. ### Known issues 1. Ideally the traversal of the IR to collect remote joins should return an AST which does not include remote join fields. This operation can be type safe but isn't taken up as part of the PR. 1. There is a lot of code duplication between `Transport/HTTP.hs` and `Transport/Websocket.hs` which needs to be fixed ASAP. This too hasn't been taken up by this PR. 1. The type which represents the execution plan is only modified to handle our current remote joins and as such it will have to be changed to accommodate general remote joins. 1. Use of lenses would have reduced the boilerplate code to collect remote joins from the base AST. 1. The current remote join logic assumes that the join columns of a remote relationship appear with their names in the database response. This however is incorrect as they could be aliased. This can be taken up by anyone, I've left a comment in the code. ### Notes to the reviewers I think it is best reviewed commit by commit. 1. The first one is very straight forward. 1. The second one refactors the remote join execution logic but other than moving things around, it doesn't change the user facing functionality. This moves Postgres specific parts to `Backends/Postgres` module from `Execute`. Some IR related code to `Hasura.RQL.IR` module. Simplifies various type class function signatures as a backend doesn't have to handle remote joins anymore 1. The third one fixes partial case matches that for some weird reason weren't shown as warnings before this refactor 1. The fourth one generalizes the validation logic of remote relationships and implements `scalarTypeGraphQLName` function on SQL Server and BigQuery which is used by the validation logic. This enables remote relationships on BigQuery and SQL Server. https://github.com/hasura/graphql-engine-mono/pull/1497 GitOrigin-RevId: 77dd8eed326602b16e9a8496f52f46d22b795598
2021-06-11 06:26:50 +03:00
mkPreparedSql f simpleSel =
PreparedSql (f simpleSel) prepped
-- Append Query Tags to the Prepared SQL
appendPreparedSQLWithQueryTags :: PreparedSql -> QueryTagsComment -> PreparedSql
appendPreparedSQLWithQueryTags preparedSQL queryTags =
preparedSQL {_psQuery = appendSQLWithQueryTags query queryTags}
where
query = _psQuery preparedSQL
appendSQLWithQueryTags :: Q.Query -> QueryTagsComment -> Q.Query
appendSQLWithQueryTags query queryTags = query {Q.getQueryText = queryText <> _unQueryTagsComment queryTags}
where
queryText = Q.getQueryText query
--------------------------------------------------------------------------------
-- Remote Relationships (e.g. DB-to-DB Joins, remote schema joins, etc.)
--------------------------------------------------------------------------------
-- | Construct an action (i.e. 'DBStepInfo') which can marshal some remote
-- relationship information into a form that Postgres can query against.
pgDBRemoteRelationshipPlan ::
forall pgKind m.
( MonadError QErr m,
Backend ('Postgres pgKind),
PostgresAnnotatedFieldJSON pgKind
) =>
UserInfo ->
SourceName ->
SourceConfig ('Postgres pgKind) ->
-- | List of json objects, each of which becomes a row of the table.
NonEmpty J.Object ->
-- | The above objects have this schema
--
-- XXX: What is this for/what does this mean?
HashMap FieldName (Column ('Postgres pgKind), ScalarType ('Postgres pgKind)) ->
-- | This is a field name from the lhs that *has* to be selected in the
-- response along with the relationship.
FieldName ->
(FieldName, IR.SourceRelationshipSelection ('Postgres pgKind) Void UnpreparedValue) ->
m (DBStepInfo ('Postgres pgKind))
pgDBRemoteRelationshipPlan userInfo sourceName sourceConfig lhs lhsSchema argumentId relationship = do
-- NOTE: 'QueryTags' currently cannot support remote relationship queries.
--
-- In the future if we want to add support we'll need to add a new type of
-- metadata (e.g. 'ParameterizedQueryHash' doesn't make sense here) and find
-- a root field name that makes sense to attach to it.
flip runReaderT emptyQueryTagsComment $ pgDBQueryPlan userInfo sourceName sourceConfig rootSelection
where
coerceToColumn = PG.unsafePGCol . getFieldNameTxt
joinColumnMapping = mapKeys coerceToColumn lhsSchema
rowsArgument :: UnpreparedValue ('Postgres pgKind)
rowsArgument =
UVParameter Nothing $
ColumnValue (ColumnScalar PG.PGJSONB) $
PG.PGValJSONB $ Q.JSONB $ J.toJSON lhs
jsonToRecordSet :: IR.SelectFromG ('Postgres pgKind) (UnpreparedValue ('Postgres pgKind))
recordSetDefinitionList =
(coerceToColumn argumentId, PG.PGBigInt) : Map.toList (fmap snd joinColumnMapping)
jsonToRecordSet =
IR.FromFunction
(PG.QualifiedObject "pg_catalog" $ PG.FunctionName "jsonb_to_recordset")
(IR.FunctionArgsExp [IR.AEInput rowsArgument] mempty)
(Just recordSetDefinitionList)
rootSelection =
convertRemoteSourceRelationship
(fst <$> joinColumnMapping)
jsonToRecordSet
(PG.unsafePGCol $ getFieldNameTxt argumentId)
(ColumnScalar PG.PGBigInt)
relationship