graphql-engine/server/src-lib/Hasura/Backends/BigQuery/FromIr.hs

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

1870 lines
73 KiB
Haskell
Raw Normal View History

-- | Translate from the DML to the BigQuery dialect.
module Hasura.Backends.BigQuery.FromIr
( mkSQLSelect,
fromRootField,
fromSelectAggregate,
Error (..),
runFromIr,
FromIr,
FromIrConfig (..),
defaultFromIrConfig,
bigQuerySourceConfigToFromIrConfig,
Top (..), -- Re-export for FromIrConfig.
)
where
import Control.Monad.Validate
import Data.HashMap.Strict qualified as HM
import Data.Int qualified as Int
import Data.List.Extended (appendToNonEmpty)
import Data.List.NonEmpty qualified as NE
import Data.Map.Strict (Map)
import Data.Map.Strict qualified as M
import Data.Text qualified as T
import Hasura.Backends.BigQuery.Instances.Types ()
import Hasura.Backends.BigQuery.Source (BigQuerySourceConfig (..))
import Hasura.Backends.BigQuery.Types as BigQuery
import Hasura.Prelude
import Hasura.RQL.IR qualified as Ir
import Hasura.RQL.Types.Column qualified as Rql
import Hasura.RQL.Types.Common qualified as Rql
import Hasura.RQL.Types.Function qualified as Rql
import Hasura.RQL.Types.Relationships.Local qualified as Rql
import Hasura.SQL.Backend
--------------------------------------------------------------------------------
-- Types
-- | Most of these errors should be checked for legitimacy.
data Error
= FromTypeUnsupported (Ir.SelectFromG 'BigQuery Expression)
| NoOrderSpecifiedInOrderBy
| MalformedAgg
| FieldTypeUnsupportedForNow (Ir.AnnFieldG 'BigQuery Void Expression)
| AggTypeUnsupportedForNow (Ir.TableAggregateFieldG 'BigQuery Void Expression)
| NodesUnsupportedForNow (Ir.TableAggregateFieldG 'BigQuery Void Expression)
| NoProjectionFields
| NoAggregatesMustBeABug
| UnsupportedArraySelect (Ir.ArraySelectG 'BigQuery Void Expression)
| UnsupportedOpExpG (Ir.OpExpG 'BigQuery Expression)
| UnsupportedSQLExp Expression
| UnsupportedDistinctOn
| UnexpectedEmptyList
| InvalidIntegerishSql Expression
| ConnectionsNotSupported
| ActionsNotSupported
| -- | https://github.com/hasura/graphql-engine/issues/8526
ComputedFieldsBooleanExpressionNotSupported
| -- | https://github.com/hasura/graphql-engine/issues/8526
ComputedFieldsOrderByNotSupported
| -- | https://github.com/hasura/graphql-engine/issues/8521
ScalarComputedFieldsNotSupported
| NoParentEntityInternalError
instance Show Error where
show =
\case
FromTypeUnsupported {} -> "FromTypeUnsupported"
NoOrderSpecifiedInOrderBy {} -> "NoOrderSpecifiedInOrderBy"
MalformedAgg {} -> "MalformedAgg"
FieldTypeUnsupportedForNow {} -> "FieldTypeUnsupportedForNow"
AggTypeUnsupportedForNow {} -> "AggTypeUnsupportedForNow"
NodesUnsupportedForNow {} -> "NodesUnsupportedForNow"
NoProjectionFields {} -> "NoProjectionFields"
NoAggregatesMustBeABug {} -> "NoAggregatesMustBeABug"
UnsupportedArraySelect {} -> "UnsupportedArraySelect"
UnsupportedOpExpG {} -> "UnsupportedOpExpG"
UnsupportedSQLExp {} -> "UnsupportedSQLExp"
UnsupportedDistinctOn {} -> "UnsupportedDistinctOn"
UnexpectedEmptyList {} -> "UnexpectedEmptyList"
InvalidIntegerishSql {} -> "InvalidIntegerishSql"
ConnectionsNotSupported {} -> "ConnectionsNotSupported"
ActionsNotSupported {} -> "ActionsNotSupported"
ComputedFieldsBooleanExpressionNotSupported {} -> "ComputedFieldsBooleanExpressionNotSupported"
ComputedFieldsOrderByNotSupported {} -> "ComputedFieldsOrderByNotSupported"
ScalarComputedFieldsNotSupported {} -> "ScalarComputedFieldsNotSupported"
NoParentEntityInternalError {} -> "NoParentEntityInternalError"
-- | The base monad used throughout this module for all conversion
-- functions.
--
-- It's a Validate, so it'll continue going when it encounters errors
-- to accumulate as many as possible.
--
-- It also contains a mapping from entity prefixes to counters. So if
-- my prefix is "table" then there'll be a counter that lets me
-- generate table1, table2, etc. Same for any other prefix needed
-- (e.g. names for joins).
--
-- A ReaderT is used around this in most of the module too, for
-- setting the current entity that a given field name refers to. See
-- @fromColumn@.
newtype FromIr a = FromIr
{ unFromIr :: ReaderT FromIrReader (StateT FromIrState (Validate (NonEmpty Error))) a
}
deriving (Functor, Applicative, Monad, MonadValidate (NonEmpty Error))
data FromIrState = FromIrState
{ indices :: !(Map Text Int)
}
data FromIrReader = FromIrReader
{ config :: !FromIrConfig
}
-- | Config values for the from-IR translator.
data FromIrConfig = FromIrConfig
{ -- | Applies globally to all selects, and may be reduced to
-- something even smaller by permission/user args.
globalSelectLimit :: !Top
}
-- | A default config.
defaultFromIrConfig :: FromIrConfig
defaultFromIrConfig = FromIrConfig {globalSelectLimit = NoTop}
-- | Alias of parent SELECT FROM.
-- Functions underlying computed fields requires column values from
-- the table that is being used in FROM clause of parent SELECT.
--
-- Example SQL:
--
-- > SELECT
-- > `t_author1`.`id` AS `id`,
-- > `t_author1`.`name` AS `name`,
-- > ARRAY(
-- > SELECT
-- > AS STRUCT `id`,
-- > `title`,
-- > `content`
-- > FROM
-- > UNNEST(
-- > ARRAY(
-- > SELECT
-- > AS STRUCT *
-- > FROM `hasura_test`.`fetch_articles`(`id` => `t_author1`.`id`)
-- > )
-- > )
-- > LIMIT 1000
-- > ) AS `articles`
-- > FROM
-- > `hasura_test`.`author` AS `t_author1`
--
-- Where `t_author1` is the @'ParentSelectFromIdentity'
data ParentSelectFromEntity
= -- | There's no parent entity
NoParentEntity
| -- | Alias of the parent SELECT FROM
ParentEntityAlias EntityAlias
--------------------------------------------------------------------------------
-- Runners
runFromIr :: FromIrConfig -> FromIr a -> Validate (NonEmpty Error) a
runFromIr config fromIr =
evalStateT
(runReaderT (unFromIr fromIr) (FromIrReader {config}))
(FromIrState {indices = mempty})
bigQuerySourceConfigToFromIrConfig :: BigQuerySourceConfig -> FromIrConfig
bigQuerySourceConfigToFromIrConfig BigQuerySourceConfig {_scGlobalSelectLimit} =
FromIrConfig {globalSelectLimit = Top _scGlobalSelectLimit}
--------------------------------------------------------------------------------
-- Similar rendition of old API
-- | Here is where we apply a top-level annotation to the select to
-- indicate to the data loader that this select ought to produce a
-- single object or an array.
mkSQLSelect ::
Rql.JsonAggSelect ->
Ir.AnnSelectG 'BigQuery (Ir.AnnFieldG 'BigQuery Void) Expression ->
FromIr BigQuery.Select
mkSQLSelect jsonAggSelect annSimpleSel = do
select <- noExtraPartitionFields <$> fromSelectRows NoParentEntity annSimpleSel
pure
( select
{ selectCardinality =
case jsonAggSelect of
Rql.JASMultipleRows -> Many
Rql.JASSingleObject -> One
}
)
-- | Convert from the IR database query into a select.
fromRootField :: Ir.QueryDB 'BigQuery Void Expression -> FromIr Select
fromRootField =
\case
(Ir.QDBSingleRow s) -> mkSQLSelect Rql.JASSingleObject 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
(Ir.QDBMultipleRows s) -> mkSQLSelect Rql.JASMultipleRows s
(Ir.QDBAggregation s) -> fromSelectAggregate Nothing s
--------------------------------------------------------------------------------
-- Top-level exported functions
fromUnnestedJSON :: Expression -> [(ColumnName, ScalarType)] -> [Rql.FieldName] -> FromIr From
fromUnnestedJSON json columns _fields = do
alias <- generateEntityAlias UnnestTemplate
pure
( FromSelectJson
( Aliased
{ aliasedThing =
SelectJson
{ selectJsonBody = json,
selectJsonFields = columns
},
aliasedAlias = entityAliasText alias
}
)
)
fromSelectRows :: ParentSelectFromEntity -> Ir.AnnSelectG 'BigQuery (Ir.AnnFieldG 'BigQuery Void) Expression -> FromIr BigQuery.PartitionableSelect
fromSelectRows parentSelectFromEntity annSelectG = do
let Ir.AnnSelectG
{ _asnFields = fields,
_asnFrom = from,
_asnPerm = perm,
_asnArgs = args,
_asnStrfyNum = stringifyNumbers
} = annSelectG
Ir.TablePerm {_tpLimit = mPermLimit, _tpFilter = permFilter} = perm
permissionBasedTop =
maybe NoTop (Top . fromIntegral) mPermLimit
selectFrom <-
case from of
Ir.FromTable qualifiedObject -> fromQualifiedTable qualifiedObject
Ir.FromFunction nm (Rql.FunctionArgsExp [BigQuery.AEInput json] _) (Just columns)
| functionName nm == "unnest" -> fromUnnestedJSON json columns (map fst fields)
Ir.FromFunction functionName (Rql.FunctionArgsExp positionalArgs namedArgs) Nothing ->
fromFunction parentSelectFromEntity functionName positionalArgs namedArgs
_ -> refute (pure (FromTypeUnsupported from))
Args
{ argsOrderBy,
argsWhere,
argsJoins,
argsTop,
argsDistinct,
argsOffset,
argsExistingJoins
} <-
runReaderT (fromSelectArgsG args) (fromAlias selectFrom)
fieldSources <-
runReaderT
(traverse (fromAnnFieldsG argsExistingJoins stringifyNumbers) fields)
(fromAlias selectFrom)
filterExpression <-
runReaderT (fromAnnBoolExp permFilter) (fromAlias selectFrom)
selectProjections <- selectProjectionsFromFieldSources True fieldSources
globalTop <- getGlobalTop
let select =
Select
{ selectCardinality = Many,
selectAsStruct = NoAsStruct,
selectFinalWantedFields = pure (fieldTextNames fields),
selectGroupBy = mempty,
selectOrderBy = argsOrderBy,
-- We DO APPLY the global top here, because this pulls down all rows.
selectTop = globalTop <> permissionBasedTop <> argsTop,
selectProjections,
selectFrom,
selectJoins = argsJoins <> concat (mapMaybe fieldSourceJoins fieldSources),
selectWhere = argsWhere <> Where [filterExpression],
selectOffset = int64Expr <$> argsOffset
}
case argsDistinct of
Nothing ->
pure $ simpleSelect select
Just distinct ->
simulateDistinctOn select distinct argsOrderBy
-- | Simulates DISTINCT ON for BigQuery using ROW_NUMBER() partitioned over distinct fields
--
-- Example:
--
-- For a GraphQL query:
-- @
-- hasura_test_article(distinct_on: author_id, order_by: [{author_id: asc}, {created_at: asc}]) {
-- id
-- title
-- }
-- @
--
-- it should produce from a query without a `distinct_on` clause:
--
-- SELECT `id`, `title`
-- FROM `hasura_test`.`article`
-- ORDER BY `author_id` ASC, `created_at` ASC
--
-- a query of the following form:
--
-- SELECT `id`, `title`
-- FROM (SELECT *,
-- ROW_NUMBER() OVER (PARTITION BY `author_id` ORDER BY `created_at` ASC) as `idx1`
-- FROM `hasura_test`.`article`) as `t_article1`
-- WHERE (`t_article1`.`idx1` = 1)
-- ORDER BY `t_article1`.`author_id` ASC
--
-- Note: this method returns PartitionableSelect as it could be joined using an array relation
-- which requires extra fields added to the PARTITION BY clause to return proper results
simulateDistinctOn :: Select -> NonEmpty ColumnName -> Maybe (NonEmpty OrderBy) -> FromIr PartitionableSelect
simulateDistinctOn select distinctOnColumns orderByColumns = do
rowNumAlias <- generateEntityAlias IndexTemplate
pure
PartitionableSelect
{ pselectFrom = selectFrom select,
pselectFinalize = \mExtraPartitionField ->
let -- we use the same alias both for outer and inner selects
alias = entityAliasText (fromAlias (selectFrom select))
distinctFields = fmap (\(ColumnName name) -> FieldName name alias) distinctOnColumns
finalDistinctFields = case mExtraPartitionField of
Just extraFields
| Just neExtraFields <- nonEmpty extraFields ->
neExtraFields <> distinctFields
_ -> distinctFields
(distinctOnOrderBy, innerOrderBy) =
case orderByColumns of
Just orderBy ->
let (distincts, others) = NE.partition (\OrderBy {..} -> orderByFieldName `elem` distinctFields) orderBy
in (NE.nonEmpty distincts, NE.nonEmpty others)
Nothing ->
(Nothing, Nothing)
innerFrom =
FromSelect
Aliased
{ aliasedAlias = alias,
aliasedThing =
select
{ selectProjections =
StarProjection
:| [ WindowProjection
( Aliased
{ aliasedAlias = unEntityAlias rowNumAlias,
aliasedThing =
RowNumberOverPartitionBy
finalDistinctFields
innerOrderBy
-- Above: Having the order by
-- in here ensures that we get the proper
-- row as the first one we select
-- in the outer select WHERE condition
-- to simulate DISTINCT ON semantics
}
)
],
selectTop = mempty,
selectJoins = mempty,
selectOrderBy = mempty,
selectOffset = Nothing,
selectGroupBy = mempty,
selectFinalWantedFields = mempty
}
}
in select
{ selectFrom = innerFrom,
selectWhere =
Where
[ EqualExpression
(ColumnExpression FieldName {fieldNameEntity = alias, fieldName = unEntityAlias rowNumAlias})
(int64Expr 1)
],
selectOrderBy = distinctOnOrderBy
}
}
fromSelectAggregate ::
Maybe (EntityAlias, HashMap ColumnName ColumnName) ->
Ir.AnnSelectG 'BigQuery (Ir.TableAggregateFieldG 'BigQuery Void) Expression ->
FromIr BigQuery.Select
Bigquery/fix limit offset for array aggregates Blocked on https://github.com/hasura/graphql-engine-mono/pull/1640. While fiddling with BigQuery I noticed a severe issue with offset/limit for array-aggregates. I've fixed it now. The basic problem was that I was using a query like this: ```graphql query MyQuery { hasura_Artist(order_by: {artist_self_id: asc}) { artist_self_id albums_aggregate(order_by: {album_self_id: asc}, limit: 2) { nodes { album_self_id } aggregate { count } } } } ``` Producing this SQL: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT * FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST -- PROBLEM HERE LIMIT @param0) AS `t_Album1` GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` Note the `LIMIT @param0` -- that is incorrect because we want to limit per artist. Instead, we want: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT *, -- ADDED ROW_NUMBER() OVER(PARTITION BY artist_other_id) artist_album_index FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST ) AS `t_Album1` -- CHANGED WHERE artist_album_index <= @param GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` That serves both the LIMIT/OFFSET function in the where clause. Then, both the ARRAY_AGG and the COUNT are correct per artist. I've updated my Haskell test suite to add regression tests for this. I'll push a commit for Python tests shortly. The tests still pass there. This just fixes a case that we hadn't noticed. https://github.com/hasura/graphql-engine-mono/pull/1641 GitOrigin-RevId: 49933fa5e09a9306c89565743ecccf2cb54eaa80
2021-07-06 11:28:42 +03:00
fromSelectAggregate minnerJoinFields annSelectG = do
selectFrom <-
case from of
Ir.FromTable qualifiedObject -> fromQualifiedTable qualifiedObject
_ -> refute (pure (FromTypeUnsupported from))
args'@Args {argsWhere, argsOrderBy, argsJoins, argsTop, argsOffset, argsDistinct} <-
runReaderT (fromSelectArgsG args) (fromAlias selectFrom)
filterExpression <-
runReaderT (fromAnnBoolExp permFilter) (fromAlias selectFrom)
Bigquery/fix limit offset for array aggregates Blocked on https://github.com/hasura/graphql-engine-mono/pull/1640. While fiddling with BigQuery I noticed a severe issue with offset/limit for array-aggregates. I've fixed it now. The basic problem was that I was using a query like this: ```graphql query MyQuery { hasura_Artist(order_by: {artist_self_id: asc}) { artist_self_id albums_aggregate(order_by: {album_self_id: asc}, limit: 2) { nodes { album_self_id } aggregate { count } } } } ``` Producing this SQL: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT * FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST -- PROBLEM HERE LIMIT @param0) AS `t_Album1` GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` Note the `LIMIT @param0` -- that is incorrect because we want to limit per artist. Instead, we want: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT *, -- ADDED ROW_NUMBER() OVER(PARTITION BY artist_other_id) artist_album_index FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST ) AS `t_Album1` -- CHANGED WHERE artist_album_index <= @param GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` That serves both the LIMIT/OFFSET function in the where clause. Then, both the ARRAY_AGG and the COUNT are correct per artist. I've updated my Haskell test suite to add regression tests for this. I'll push a commit for Python tests shortly. The tests still pass there. This just fixes a case that we hadn't noticed. https://github.com/hasura/graphql-engine-mono/pull/1641 GitOrigin-RevId: 49933fa5e09a9306c89565743ecccf2cb54eaa80
2021-07-06 11:28:42 +03:00
mforeignKeyConditions <-
for minnerJoinFields $ \(entityAlias, mapping) ->
runReaderT
(fromMappingFieldNames (fromAlias selectFrom) mapping)
entityAlias
fieldSources <-
runReaderT
( traverse
( fromTableAggregateFieldG
args'
permissionBasedTop
stringifyNumbers
)
fields
)
(fromAlias selectFrom)
selectProjections <- selectProjectionsFromFieldSources True fieldSources
Bigquery/fix limit offset for array aggregates Blocked on https://github.com/hasura/graphql-engine-mono/pull/1640. While fiddling with BigQuery I noticed a severe issue with offset/limit for array-aggregates. I've fixed it now. The basic problem was that I was using a query like this: ```graphql query MyQuery { hasura_Artist(order_by: {artist_self_id: asc}) { artist_self_id albums_aggregate(order_by: {album_self_id: asc}, limit: 2) { nodes { album_self_id } aggregate { count } } } } ``` Producing this SQL: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT * FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST -- PROBLEM HERE LIMIT @param0) AS `t_Album1` GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` Note the `LIMIT @param0` -- that is incorrect because we want to limit per artist. Instead, we want: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT *, -- ADDED ROW_NUMBER() OVER(PARTITION BY artist_other_id) artist_album_index FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST ) AS `t_Album1` -- CHANGED WHERE artist_album_index <= @param GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` That serves both the LIMIT/OFFSET function in the where clause. Then, both the ARRAY_AGG and the COUNT are correct per artist. I've updated my Haskell test suite to add regression tests for this. I'll push a commit for Python tests shortly. The tests still pass there. This just fixes a case that we hadn't noticed. https://github.com/hasura/graphql-engine-mono/pull/1641 GitOrigin-RevId: 49933fa5e09a9306c89565743ecccf2cb54eaa80
2021-07-06 11:28:42 +03:00
indexAlias <- generateEntityAlias IndexTemplate
let innerSelectAlias = entityAliasText (fromAlias selectFrom)
mDistinctFields = fmap (fmap (\(ColumnName name) -> FieldName name innerSelectAlias)) argsDistinct
mPartitionFields = (mforeignKeyConditions >>= NE.nonEmpty . map fst) <> mDistinctFields
innerProjections =
case mPartitionFields of
Nothing -> pure StarProjection
Just partitionFields ->
StarProjection
:|
-- We setup an index over every row in
-- the sub select. Then if you look at
-- the outer Select, you can see we apply
-- a WHERE that uses this index for
-- LIMIT/OFFSET or DISTINCT ON.
[ WindowProjection
( Aliased
{ aliasedAlias = unEntityAlias indexAlias,
aliasedThing =
RowNumberOverPartitionBy
-- The row numbers start from 1.
partitionFields
argsOrderBy
-- Above: Having the order by
-- in here ensures that the
-- row numbers are ordered by
-- this ordering. Below, we
-- order again for the
-- general row order. Both
-- are needed!
}
)
]
indexColumn =
ColumnExpression $
FieldName
{ fieldNameEntity = innerSelectAlias,
fieldName = unEntityAlias indexAlias
}
pure
Select
{ selectCardinality = One,
selectAsStruct = NoAsStruct,
selectFinalWantedFields = Nothing,
selectGroupBy = mempty,
selectProjections,
selectTop = NoTop,
selectFrom =
FromSelect
( Aliased
{ aliasedThing =
Select
{ selectProjections = innerProjections,
selectAsStruct = NoAsStruct,
selectFrom,
selectJoins = argsJoins,
selectWhere = argsWhere <> (Where [filterExpression]),
selectOrderBy = argsOrderBy,
-- Above: This is important to have here, because
-- offset/top apply AFTER ordering is applied, so
-- you can't put an order by in afterwards in a
-- parent query. Therefore be careful about
-- putting this elsewhere.
selectFinalWantedFields = Nothing,
selectCardinality = Many,
selectTop = maybe argsTop (const NoTop) mforeignKeyConditions,
-- we apply offset only if we don't have partitions
-- when we do OFFSET/LIMIT based on ROW_NUMBER()
selectOffset = maybe (int64Expr <$> argsOffset) (const Nothing) mPartitionFields,
selectGroupBy = mempty
},
aliasedAlias = innerSelectAlias
}
),
selectJoins = concat (mapMaybe fieldSourceJoins fieldSources),
selectWhere =
case mPartitionFields of
Bigquery/fix limit offset for array aggregates Blocked on https://github.com/hasura/graphql-engine-mono/pull/1640. While fiddling with BigQuery I noticed a severe issue with offset/limit for array-aggregates. I've fixed it now. The basic problem was that I was using a query like this: ```graphql query MyQuery { hasura_Artist(order_by: {artist_self_id: asc}) { artist_self_id albums_aggregate(order_by: {album_self_id: asc}, limit: 2) { nodes { album_self_id } aggregate { count } } } } ``` Producing this SQL: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT * FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST -- PROBLEM HERE LIMIT @param0) AS `t_Album1` GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` Note the `LIMIT @param0` -- that is incorrect because we want to limit per artist. Instead, we want: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT *, -- ADDED ROW_NUMBER() OVER(PARTITION BY artist_other_id) artist_album_index FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST ) AS `t_Album1` -- CHANGED WHERE artist_album_index <= @param GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` That serves both the LIMIT/OFFSET function in the where clause. Then, both the ARRAY_AGG and the COUNT are correct per artist. I've updated my Haskell test suite to add regression tests for this. I'll push a commit for Python tests shortly. The tests still pass there. This just fixes a case that we hadn't noticed. https://github.com/hasura/graphql-engine-mono/pull/1641 GitOrigin-RevId: 49933fa5e09a9306c89565743ecccf2cb54eaa80
2021-07-06 11:28:42 +03:00
Nothing -> mempty
Just {} ->
let offset =
case argsDistinct of
Nothing ->
case argsOffset of
Nothing -> mempty
Just offset' ->
-- Apply an offset using the row_number from above.
[ OpExpression
MoreOp
indexColumn
(int64Expr offset')
]
Just {} ->
-- in case of distinct_on we need to select the row number offset+1
-- effectively skipping number of rows equal to offset
[ EqualExpression
indexColumn
(int64Expr (fromMaybe 0 argsOffset + 1))
]
Bigquery/fix limit offset for array aggregates Blocked on https://github.com/hasura/graphql-engine-mono/pull/1640. While fiddling with BigQuery I noticed a severe issue with offset/limit for array-aggregates. I've fixed it now. The basic problem was that I was using a query like this: ```graphql query MyQuery { hasura_Artist(order_by: {artist_self_id: asc}) { artist_self_id albums_aggregate(order_by: {album_self_id: asc}, limit: 2) { nodes { album_self_id } aggregate { count } } } } ``` Producing this SQL: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT * FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST -- PROBLEM HERE LIMIT @param0) AS `t_Album1` GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` Note the `LIMIT @param0` -- that is incorrect because we want to limit per artist. Instead, we want: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT *, -- ADDED ROW_NUMBER() OVER(PARTITION BY artist_other_id) artist_album_index FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST ) AS `t_Album1` -- CHANGED WHERE artist_album_index <= @param GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` That serves both the LIMIT/OFFSET function in the where clause. Then, both the ARRAY_AGG and the COUNT are correct per artist. I've updated my Haskell test suite to add regression tests for this. I'll push a commit for Python tests shortly. The tests still pass there. This just fixes a case that we hadn't noticed. https://github.com/hasura/graphql-engine-mono/pull/1641 GitOrigin-RevId: 49933fa5e09a9306c89565743ecccf2cb54eaa80
2021-07-06 11:28:42 +03:00
limit =
case argsTop of
NoTop -> mempty
Top limit' ->
-- Apply a limit using the row_number from above.
[ OpExpression
LessOp
indexColumn
( int64Expr (limit' + 1) -- Because the row_number() indexing starts at 1.
-- So idx<l+1 means idx<2 where l = 1 i.e. "limit to 1 row".
)
]
in Where (offset <> limit),
selectOrderBy = Nothing,
selectOffset = Nothing
}
where
Ir.AnnSelectG
{ _asnFields = fields,
_asnFrom = from,
_asnPerm = perm,
_asnArgs = args,
_asnStrfyNum = stringifyNumbers -- TODO: Do we ignore this for aggregates?
} = annSelectG
Ir.TablePerm {_tpLimit = mPermLimit, _tpFilter = permFilter} = perm
permissionBasedTop =
maybe NoTop (Top . fromIntegral) mPermLimit
--------------------------------------------------------------------------------
-- GraphQL Args
data Args = Args
{ argsWhere :: Where,
argsOrderBy :: Maybe (NonEmpty OrderBy),
argsJoins :: [Join],
argsTop :: Top,
argsOffset :: Maybe Int.Int64,
argsDistinct :: Maybe (NonEmpty ColumnName),
argsExistingJoins :: Map TableName EntityAlias
}
deriving (Show)
data UnfurledJoin = UnfurledJoin
{ unfurledJoin :: Join,
-- | Recorded if we joined onto an object relation.
unfurledObjectTableAlias :: Maybe (TableName, EntityAlias)
}
deriving (Show)
fromSelectArgsG :: Ir.SelectArgsG 'BigQuery Expression -> ReaderT EntityAlias FromIr Args
fromSelectArgsG selectArgsG = do
argsWhere <-
maybe (pure mempty) (fmap (Where . pure) . fromAnnBoolExp) mannBoolExp
let argsTop = maybe mempty (Top . fromIntegral) mlimit
(argsOrderBy, joins) <-
runWriterT (traverse fromAnnotatedOrderByItemG (maybe [] toList orders))
-- Any object-relation joins that we generated, we record their
-- generated names into a mapping.
let argsExistingJoins =
M.fromList (mapMaybe unfurledObjectTableAlias (toList joins))
pure
Args
{ argsJoins = toList (fmap unfurledJoin joins),
argsOrderBy = NE.nonEmpty argsOrderBy,
argsDistinct = mdistinct,
..
}
where
Ir.SelectArgs
{ _saWhere = mannBoolExp,
_saLimit = mlimit,
_saOffset = argsOffset,
_saDistinct = mdistinct,
_saOrderBy = orders
} = selectArgsG
-- | Produce a valid ORDER BY construct, telling about any joins
-- needed on the side.
fromAnnotatedOrderByItemG ::
Ir.AnnotatedOrderByItemG 'BigQuery Expression -> WriterT (Seq UnfurledJoin) (ReaderT EntityAlias FromIr) OrderBy
fromAnnotatedOrderByItemG Ir.OrderByItemG {obiType, obiColumn, obiNulls} = do
orderByFieldName <- unfurlAnnotatedOrderByElement obiColumn
let morderByOrder =
obiType
let orderByNullsOrder =
fromMaybe NullsAnyOrder obiNulls
case morderByOrder of
Just orderByOrder -> pure OrderBy {..}
Nothing -> refute (pure NoOrderSpecifiedInOrderBy)
-- | Unfurl the nested set of object relations (tell'd in the writer)
-- that are terminated by field name (Ir.AOCColumn and
-- Ir.AOCArrayAggregation).
unfurlAnnotatedOrderByElement ::
Ir.AnnotatedOrderByElement 'BigQuery Expression -> WriterT (Seq UnfurledJoin) (ReaderT EntityAlias FromIr) FieldName
unfurlAnnotatedOrderByElement =
\case
Ir.AOCColumn columnInfo -> lift (fromColumnInfo columnInfo)
Ir.AOCObjectRelation Rql.RelInfo {riMapping = mapping, riRTable = tableName} annBoolExp annOrderByElementG -> do
selectFrom <- lift (lift (fromQualifiedTable tableName))
joinAliasEntity <-
lift (lift (generateEntityAlias (ForOrderAlias (tableNameText tableName))))
joinOn <- lift (fromMappingFieldNames joinAliasEntity mapping)
whereExpression <-
lift (local (const (fromAlias selectFrom)) (fromAnnBoolExp annBoolExp))
tell
( pure
UnfurledJoin
{ unfurledJoin =
Join
{ joinSource =
JoinSelect
Select
{ selectCardinality = One,
selectAsStruct = NoAsStruct,
selectFinalWantedFields = Nothing,
selectGroupBy = mempty,
selectTop = NoTop,
selectProjections = StarProjection :| [],
selectFrom,
selectJoins = [],
selectWhere = Where ([whereExpression]),
selectOrderBy = Nothing,
selectOffset = Nothing
},
joinRightTable = fromAlias selectFrom,
joinAlias = joinAliasEntity,
joinOn,
joinProvenance = OrderByJoinProvenance,
joinFieldName = tableNameText tableName, -- TODO: not needed.
joinExtractPath = Nothing
},
unfurledObjectTableAlias = Just (tableName, joinAliasEntity)
}
)
local (const joinAliasEntity) (unfurlAnnotatedOrderByElement annOrderByElementG)
Ir.AOCArrayAggregation Rql.RelInfo {riMapping = mapping, riRTable = tableName} annBoolExp annAggregateOrderBy -> do
selectFrom <- lift (lift (fromQualifiedTable tableName))
let alias = aggFieldName
joinAlias <-
lift (lift (generateEntityAlias (ForOrderAlias (tableNameText tableName))))
joinOn <- lift (fromMappingFieldNames joinAlias mapping)
innerJoinFields <-
lift (fromMappingFieldNames (fromAlias selectFrom) mapping)
whereExpression <-
lift (local (const (fromAlias selectFrom)) (fromAnnBoolExp annBoolExp))
aggregate <-
lift
( local
(const (fromAlias selectFrom))
( case annAggregateOrderBy of
Ir.AAOCount -> pure (CountAggregate StarCountable)
Ir.AAOOp text columnInfo -> do
fieldName <- fromColumnInfo columnInfo
pure (OpAggregate text (ColumnExpression fieldName))
)
)
tell
( pure
( UnfurledJoin
{ unfurledJoin =
Join
{ joinSource =
JoinSelect
Select
{ selectCardinality = One,
selectAsStruct = NoAsStruct,
selectFinalWantedFields = Nothing,
selectTop = NoTop,
selectProjections =
AggregateProjection
Aliased
{ aliasedThing = aggregate,
aliasedAlias = alias
}
:|
-- These are group by'd below in selectGroupBy.
map
( \(fieldName', _) ->
FieldNameProjection
Aliased
{ aliasedThing = fieldName',
aliasedAlias = fieldName fieldName'
}
)
innerJoinFields,
selectFrom,
selectJoins = [],
selectWhere = Where [whereExpression],
selectOrderBy = Nothing,
selectOffset = Nothing,
-- This group by corresponds to the field name projections above.
selectGroupBy = map fst innerJoinFields
},
joinRightTable = fromAlias selectFrom,
joinProvenance = OrderByJoinProvenance,
joinAlias = joinAlias,
joinOn,
joinFieldName = tableNameText tableName, -- TODO: not needed.
joinExtractPath = Nothing
},
unfurledObjectTableAlias = Nothing
}
)
)
pure
FieldName
{ fieldNameEntity = entityAliasText joinAlias,
fieldName = alias
}
Ir.AOCComputedField {} -> refute $ pure ComputedFieldsOrderByNotSupported
--------------------------------------------------------------------------------
-- Conversion functions
tableNameText :: TableName -> Text
tableNameText (TableName {tableName = qname}) = qname
-- | This is really the start where you query the base table,
-- everything else is joins attached to it.
fromQualifiedTable :: TableName -> FromIr From
fromQualifiedTable (TableName {tableNameSchema = schemaName, tableName = qname}) = do
alias <- generateEntityAlias (TableTemplate qname)
pure
( FromQualifiedTable
( Aliased
{ aliasedThing =
TableName {tableName = qname, tableNameSchema = schemaName},
aliasedAlias = entityAliasText alias
}
)
)
-- | Build a @'From' expression out of a function that returns a set of rows.
fromFunction ::
-- | The parent's entity alias from which the column values for computed fields are referred
ParentSelectFromEntity ->
-- | The function
FunctionName ->
-- | List of positional Arguments
[ArgumentExp Expression] ->
-- | List of named arguments
HM.HashMap Text (ArgumentExp Expression) ->
FromIr From
fromFunction parentEntityAlias functionName positionalArgs namedArgs = do
alias <- generateEntityAlias (FunctionTemplate functionName)
positionalArgExps <- mapM fromArgumentExp positionalArgs
namedArgExps <- for (HM.toList namedArgs) $ \(argName, argValue) -> FunctionNamedArgument argName <$> fromArgumentExp argValue
pure
( FromFunction
( Aliased
{ aliasedThing = SelectFromFunction functionName (positionalArgExps <> namedArgExps),
aliasedAlias = entityAliasText alias
}
)
)
where
fromArgumentExp :: ArgumentExp Expression -> FromIr Expression
fromArgumentExp = \case
AEInput e -> pure e
AETableColumn (ColumnName columnName) -> do
case parentEntityAlias of
NoParentEntity -> refute $ pure NoParentEntityInternalError
ParentEntityAlias entityAlias ->
pure $
ColumnExpression $
FieldName columnName (entityAliasText entityAlias)
fromAnnBoolExp ::
Ir.GBoolExp 'BigQuery (Ir.AnnBoolExpFld 'BigQuery Expression) ->
ReaderT EntityAlias FromIr Expression
fromAnnBoolExp = traverse fromAnnBoolExpFld >=> fromGBoolExp
fromAnnBoolExpFld ::
Ir.AnnBoolExpFld 'BigQuery Expression -> ReaderT EntityAlias FromIr Expression
fromAnnBoolExpFld =
\case
Ir.AVColumn columnInfo opExpGs -> do
expression <- fmap ColumnExpression (fromColumnInfo columnInfo)
expressions <- traverse (lift . fromOpExpG expression) opExpGs
pure (AndExpression expressions)
Ir.AVRelationship Rql.RelInfo {riMapping = mapping, riRTable = table} annBoolExp -> do
selectFrom <- lift (fromQualifiedTable table)
foreignKeyConditions <- fromMapping selectFrom mapping
whereExpression <-
local (const (fromAlias selectFrom)) (fromAnnBoolExp annBoolExp)
pure
( ExistsExpression
Select
{ selectCardinality = One,
selectAsStruct = NoAsStruct,
selectFinalWantedFields = Nothing,
selectGroupBy = mempty,
selectOrderBy = Nothing,
selectProjections =
ExpressionProjection
( Aliased
{ aliasedThing = trueExpression,
aliasedAlias = existsFieldName
}
)
:| [],
selectFrom,
selectJoins = mempty,
selectWhere = Where (foreignKeyConditions <> [whereExpression]),
selectTop = NoTop,
selectOffset = Nothing
}
)
Ir.AVComputedField {} -> refute $ pure ComputedFieldsBooleanExpressionNotSupported
fromColumnInfo :: Rql.ColumnInfo 'BigQuery -> ReaderT EntityAlias FromIr FieldName
fromColumnInfo Rql.ColumnInfo {ciColumn = ColumnName column} = do
EntityAlias {entityAliasText} <- ask
pure
( FieldName
{ fieldName = column,
fieldNameEntity = entityAliasText
}
)
fromGExists :: Ir.GExists 'BigQuery Expression -> ReaderT EntityAlias FromIr Select
fromGExists Ir.GExists {_geTable, _geWhere} = do
selectFrom <- lift (fromQualifiedTable _geTable)
whereExpression <-
local (const (fromAlias selectFrom)) (fromGBoolExp _geWhere)
pure
Select
{ selectCardinality = One,
selectAsStruct = NoAsStruct,
selectFinalWantedFields = Nothing,
selectGroupBy = mempty,
selectOrderBy = Nothing,
selectProjections =
ExpressionProjection
( Aliased
{ aliasedThing = trueExpression,
aliasedAlias = existsFieldName
}
)
:| [],
selectFrom,
selectJoins = mempty,
selectWhere = Where [whereExpression],
selectTop = NoTop,
selectOffset = Nothing
}
--------------------------------------------------------------------------------
-- Sources of projected fields
--
-- Because in the IR, a field projected can be a foreign object, we
-- have to both generate a projection AND on the side generate a join.
--
-- So a @FieldSource@ couples the idea of the projected thing and the
-- source of it (via 'Aliased').
data FieldSource
= ExpressionFieldSource (Aliased Expression)
| JoinFieldSource (Aliased Join)
| AggregateFieldSource Text (NonEmpty (Aliased Aggregate))
| ArrayAggFieldSource (Aliased ArrayAgg) (Maybe [FieldSource])
deriving (Eq, Show)
-- Example:
--
-- @
-- Track_aggregate {
-- aggregate {
-- count(columns: AlbumId)
-- foo: count(columns: AlbumId)
-- max {
-- AlbumId
-- TrackId
-- }
-- }
-- }
-- @
--
-- field =
-- @
-- TAFAgg
-- [ ( FieldName {getFieldNameTxt = "count"}
-- , AFCount (NonNullFieldCountable [ColumnName {columnName = "AlbumId"}]))
-- , ( FieldName {getFieldNameTxt = "foo"}
-- , AFCount (NonNullFieldCountable [ColumnName {columnName = "AlbumId"}]))
-- , ( FieldName {getFieldNameTxt = "max"}
-- , AFOp
-- (AggregateOp
-- { _aoOp = "max"
-- , _aoFields =
-- [ ( FieldName {getFieldNameTxt = "AlbumId"}
-- , CFCol (ColumnName {columnName = "AlbumId"} (ColumnScalar IntegerScalarType)))
-- , ( FieldName {getFieldNameTxt = "TrackId"}
-- , CFCol (ColumnName {columnName = "TrackId"} (ColumnScalar IntegerScalarType)))
-- ]
-- }))
-- ]
-- @
--
-- should produce:
--
-- SELECT COUNT(`t_Track1`.`AlbumId`) AS `count`,
-- COUNT(`t_Track1`.`AlbumId`) AS `foo`,
-- struct(max(`t_Track1`.`AlbumId`) AS `AlbumId`, max(`t_Track1`.`TrackId`) as TrackId) as `max`
-- FROM chinook.`Track` AS `t_Track1`
--
fromTableAggregateFieldG ::
Args ->
Top ->
Rql.StringifyNumbers ->
(Rql.FieldName, Ir.TableAggregateFieldG 'BigQuery Void Expression) ->
ReaderT EntityAlias FromIr FieldSource
fromTableAggregateFieldG args permissionBasedTop stringifyNumbers (Rql.FieldName name, field) =
case field of
Ir.TAFAgg (aggregateFields :: [(Rql.FieldName, Ir.AggregateField 'BigQuery)]) ->
case NE.nonEmpty aggregateFields of
Nothing -> refute (pure NoAggregatesMustBeABug)
Just fields -> do
aggregates <-
traverse
( \(fieldName, aggregateField) -> do
fmap
( \aliasedThing ->
Aliased {aliasedAlias = Rql.getFieldNameTxt fieldName, ..}
)
(fromAggregateField aggregateField)
)
fields
pure (AggregateFieldSource name aggregates)
Ir.TAFExp text ->
pure
( ExpressionFieldSource
Aliased
{ aliasedThing = BigQuery.ValueExpression (StringValue text),
aliasedAlias = name
}
)
Ir.TAFNodes _ (fields :: [(Rql.FieldName, Ir.AnnFieldG 'BigQuery Void Expression)]) -> do
fieldSources <-
traverse
(fromAnnFieldsG (argsExistingJoins args) stringifyNumbers)
fields
arrayAggProjections <- lift (selectProjectionsFromFieldSources False fieldSources)
globalTop <- lift getGlobalTop
let arrayAgg =
Aliased
{ aliasedThing =
ArrayAgg
{ arrayAggProjections,
arrayAggOrderBy = argsOrderBy args,
arrayAggTop = globalTop <> argsTop args <> permissionBasedTop
},
aliasedAlias = name
}
pure (ArrayAggFieldSource arrayAgg (Just fieldSources))
fromAggregateField :: Ir.AggregateField 'BigQuery -> ReaderT EntityAlias FromIr Aggregate
fromAggregateField aggregateField =
case aggregateField of
Ir.AFExp text -> pure (TextAggregate text)
Ir.AFCount countType ->
CountAggregate <$> case countType of
StarCountable -> pure StarCountable
NonNullFieldCountable names -> NonNullFieldCountable <$> traverse fromColumn names
DistinctCountable names -> DistinctCountable <$> traverse fromColumn names
Ir.AFOp Ir.AggregateOp {_aoOp = op, _aoFields = fields} -> do
fs <- NE.nonEmpty fields `onNothing` refute (pure MalformedAgg)
args <-
traverse
( \(Rql.FieldName fieldName, columnField) -> do
expression' <-
case columnField of
Ir.CFCol column _columnType -> fmap ColumnExpression (fromColumn column)
Ir.CFExp text -> pure (ValueExpression (StringValue text))
pure (fieldName, expression')
)
fs
pure (OpAggregates op args)
-- | The main sources of fields, either constants, fields or via joins.
fromAnnFieldsG ::
Map TableName EntityAlias ->
Rql.StringifyNumbers ->
(Rql.FieldName, Ir.AnnFieldG 'BigQuery Void Expression) ->
ReaderT EntityAlias FromIr FieldSource
fromAnnFieldsG existingJoins stringifyNumbers (Rql.FieldName name, field) =
case field of
Ir.AFColumn annColumnField -> do
expression <- fromAnnColumnField stringifyNumbers annColumnField
pure
( ExpressionFieldSource
Aliased {aliasedThing = expression, aliasedAlias = name}
)
Ir.AFExpression text ->
pure
( ExpressionFieldSource
Aliased
{ aliasedThing = BigQuery.ValueExpression (StringValue text),
aliasedAlias = name
}
)
Ir.AFObjectRelation objectRelationSelectG ->
fmap
( \aliasedThing ->
JoinFieldSource (Aliased {aliasedThing, aliasedAlias = name})
)
(fromObjectRelationSelectG existingJoins objectRelationSelectG)
Ir.AFArrayRelation arraySelectG ->
fmap
( \aliasedThing ->
JoinFieldSource (Aliased {aliasedThing, aliasedAlias = name})
)
(fromArraySelectG arraySelectG)
Ir.AFComputedField _ _ computedFieldSelect -> do
expression <- fromComputedFieldSelect computedFieldSelect
pure
( ExpressionFieldSource
Aliased {aliasedThing = expression, aliasedAlias = name}
)
-- | Here is where we project a field as a column expression. If
-- number stringification is on, then we wrap it in a
-- 'ToStringExpression' so that it's casted when being projected.
fromAnnColumnField ::
Rql.StringifyNumbers ->
Ir.AnnColumnField 'BigQuery Expression ->
ReaderT EntityAlias FromIr Expression
fromAnnColumnField _stringifyNumbers annColumnField = do
fieldName <- fromColumn column
if asText || False -- TODO: (Rql.isScalarColumnWhere Psql.isBigNum typ && stringifyNumbers == Rql.StringifyNumbers)
then pure (ToStringExpression (ColumnExpression fieldName))
else case caseBoolExpMaybe of
Nothing -> pure (ColumnExpression fieldName)
Just ex -> do
ex' <- (traverse fromAnnBoolExpFld >=> fromGBoolExp) (coerce ex)
pure (ConditionalProjection ex' fieldName)
where
Ir.AnnColumnField
{ _acfColumn = column,
_acfAsText = asText :: Bool,
_acfArguments = _ :: Maybe Void,
_acfCaseBoolExpression = caseBoolExpMaybe :: Maybe (Ir.AnnColumnCaseBoolExp 'BigQuery Expression)
} = annColumnField
-- | This is where a field name "foo" is resolved to a fully qualified
-- field name [table].[foo]. The table name comes from EntityAlias in
-- the ReaderT.
fromColumn :: ColumnName -> ReaderT EntityAlias FromIr FieldName
fromColumn (ColumnName txt) = do
EntityAlias {entityAliasText} <- ask
pure (FieldName {fieldName = txt, fieldNameEntity = entityAliasText})
fieldSourceProjections :: Bool -> FieldSource -> FromIr (NonEmpty Projection)
fieldSourceProjections keepJoinField =
\case
ExpressionFieldSource aliasedExpression ->
pure (ExpressionProjection aliasedExpression :| [])
JoinFieldSource aliasedJoin ->
toNonEmpty
-- Here we're producing all join fields needed later for
-- Haskell-native joining. They will be removed by upstream
-- code if keepJoinField is True
( [ FieldNameProjection
( Aliased
{ aliasedThing = right,
aliasedAlias = fieldNameText right
}
)
| keepJoinField,
(_left, right) <- joinOn join'
]
<>
-- Below:
-- When we're doing an array-aggregate, e.g.
--
-- query MyQuery {
-- hasura_Artist {
-- albums_aggregate {
-- aggregate {
-- count
-- }
-- }
-- }
-- }
--
-- we're going to do a join on the albums table, and that
-- join query will produce a single-row result. Therefore we
-- can grab the whole entity as a STRUCT-typed object. See
-- also the docs for 'fromArrayRelationSelectG' and for
-- 'fromArrayAggregateSelectG'.
case joinProvenance join' of
ArrayJoinProvenance fields ->
pure
( ArrayEntityProjection
(joinAlias join')
aliasedJoin
{ aliasedThing =
fmap
( \name ->
FieldName
{ fieldName = name,
fieldNameEntity =
entityAliasText (joinAlias join')
}
)
fields,
aliasedAlias = aliasedAlias aliasedJoin
}
)
ObjectJoinProvenance fields ->
pure
( EntityProjection
aliasedJoin
{ aliasedThing =
fmap
( \name ->
( FieldName
{ fieldName = name,
fieldNameEntity =
entityAliasText (joinAlias join')
},
NoOrigin
)
)
fields,
aliasedAlias = aliasedAlias aliasedJoin
}
)
ArrayAggregateJoinProvenance fields ->
pure
( EntityProjection
aliasedJoin
{ aliasedThing =
fmap
( \(name, fieldOrigin) ->
( FieldName
{ fieldName = name,
fieldNameEntity =
entityAliasText (joinAlias join')
},
fieldOrigin
)
)
fields,
aliasedAlias = aliasedAlias aliasedJoin
}
)
_ -> []
)
where
join' = aliasedThing aliasedJoin
AggregateFieldSource name aggregates ->
pure
( AggregateProjections
(Aliased {aliasedThing = aggregates, aliasedAlias = name})
:| []
)
ArrayAggFieldSource arrayAgg _ -> pure (ArrayAggProjection arrayAgg :| [])
where
fieldNameText FieldName {fieldName} = fieldName
fieldSourceJoins :: FieldSource -> Maybe [Join]
fieldSourceJoins =
\case
JoinFieldSource aliasedJoin -> pure [aliasedThing aliasedJoin]
ExpressionFieldSource {} -> Nothing
AggregateFieldSource {} -> Nothing
ArrayAggFieldSource _ sources -> fmap (concat . mapMaybe fieldSourceJoins) sources
--------------------------------------------------------------------------------
-- Joins
-- | Produce the join for an object relation. We produce a normal
-- select, but then include join fields. Then downstream, the
-- DataLoader will execute the lhs select and rhs join in separate
-- server queries, then do a Haskell-native join on the join fields.
--
-- See also 'fromArrayRelationSelectG' for similar example.
fromObjectRelationSelectG ::
Map TableName EntityAlias ->
Ir.ObjectRelationSelectG 'BigQuery Void Expression ->
ReaderT EntityAlias FromIr Join
-- We're not using existingJoins at the moment, which was used to
-- avoid re-joining on the same table twice.
fromObjectRelationSelectG _existingJoins annRelationSelectG = do
selectFrom <- lift (fromQualifiedTable tableFrom)
let entityAlias :: EntityAlias = fromAlias selectFrom
fieldSources <-
local
(const entityAlias)
(traverse (fromAnnFieldsG mempty Rql.LeaveNumbersAlone) fields)
selectProjections <- lift (selectProjectionsFromFieldSources True fieldSources)
joinFieldName <- lift (fromRelName _aarRelationshipName)
joinAlias <-
lift (generateEntityAlias (ObjectRelationTemplate joinFieldName))
filterExpression <- local (const entityAlias) (fromAnnBoolExp tableFilter)
-- @mapping@ here describes the pairs of columns that form foreign key
-- relationships between tables. For example, when querying an "article"
-- table for article titles and joining on an "authors" table for author
-- names, we could end up with something like the following:
--
-- @
-- [ ( ColumnName { columnName = "author_id" }
-- , ColumnName { columnName = "id" }
-- )
-- ]
-- @
--
-- Note that the "local" table is on the left, and the "remote" table is on
-- the right.
joinFields <- fromMappingFieldNames (fromAlias selectFrom) mapping
joinOn <- fromMappingFieldNames joinAlias mapping
joinFieldProjections <- toNonEmpty (map prepareJoinFieldProjection joinFields)
let selectFinalWantedFields = pure (fieldTextNames fields)
pure
Join
{ joinAlias,
joinSource =
JoinSelect
Select
{ selectCardinality = One,
selectAsStruct = NoAsStruct,
selectFinalWantedFields,
selectGroupBy = mempty,
selectOrderBy = Nothing,
selectTop = NoTop,
selectProjections = joinFieldProjections <> selectProjections,
selectFrom,
selectJoins = concat (mapMaybe fieldSourceJoins fieldSources),
selectWhere = Where [filterExpression],
selectOffset = Nothing
},
joinOn,
joinRightTable = fromAlias selectFrom,
joinProvenance =
ObjectJoinProvenance
(fromMaybe [] selectFinalWantedFields), -- TODO: OK?
-- Above: Needed by DataLoader to determine the type of
-- Haskell-native join to perform.
joinFieldName,
joinExtractPath = Nothing
}
where
Ir.AnnObjectSelectG
{ _aosFields = fields :: Ir.AnnFieldsG 'BigQuery Void Expression,
_aosTableFrom = tableFrom :: TableName,
_aosTableFilter = tableFilter :: Ir.AnnBoolExp 'BigQuery Expression
} = annObjectSelectG
Ir.AnnRelationSelectG
{ _aarRelationshipName,
_aarColumnMapping = mapping :: HashMap ColumnName ColumnName,
_aarAnnSelect = annObjectSelectG :: Ir.AnnObjectSelectG 'BigQuery Void Expression
} = annRelationSelectG
-- We're not using existingJoins at the moment, which was used to
-- avoid re-joining on the same table twice.
_lookupTableFrom ::
Map TableName EntityAlias ->
TableName ->
FromIr (Either EntityAlias From)
_lookupTableFrom existingJoins tableFrom = do
case M.lookup tableFrom existingJoins of
Just entityAlias -> pure (Left entityAlias)
Nothing -> fmap Right (fromQualifiedTable tableFrom)
fromArraySelectG :: Ir.ArraySelectG 'BigQuery Void Expression -> ReaderT EntityAlias FromIr Join
fromArraySelectG =
\case
Ir.ASSimple arrayRelationSelectG ->
fromArrayRelationSelectG arrayRelationSelectG
Ir.ASAggregate arrayAggregateSelectG ->
fromArrayAggregateSelectG arrayAggregateSelectG
-- | Generate a select field @'Expression' for a computed field
--
-- > ARRAY(
-- > SELECT
-- > AS STRUCT
-- > `column_1`,
-- > `column_2`,
-- > `column_3`
-- > FROM
-- > UNNEST(
-- > ARRAY(
-- > SELECT AS STRUCT *
-- > FROM `dataset`.`function_name`(`argument_name` => `parent_entity`.`column`)
-- > )
-- > )
-- > LIMIT 1000 -- global limit
-- > ) AS `field_name`
--
-- Using 'LIMIT' right after 'FROM <function>' expression raises query exception.
-- To avoid this problem, we are packing and unpacking the rows returned from the function
-- using 'ARRAY' and 'UNNEST', then applying LIMIT. Somehow this is working with exact reason
-- being unknown. See https://github.com/hasura/graphql-engine/issues/8562 for more details.
fromComputedFieldSelect ::
Ir.ComputedFieldSelect 'BigQuery Void Expression ->
ReaderT EntityAlias FromIr Expression
fromComputedFieldSelect = \case
Ir.CFSScalar {} ->
-- As of now, we don't have support for computed fields returning a scalar value.
-- See https://github.com/hasura/graphql-engine/issues/8521
refute $ pure ScalarComputedFieldsNotSupported
Ir.CFSTable jsonAggSelect annSimpleSelect -> do
entityAlias <- ask
select <- lift $ noExtraPartitionFields <$> fromSelectRows (ParentEntityAlias entityAlias) annSimpleSelect
let selectWithCardinality =
select
{ selectCardinality =
case jsonAggSelect of
Rql.JASMultipleRows -> Many
Rql.JASSingleObject -> One,
selectAsStruct = AsStruct,
selectFrom = wrapUnnest (selectFrom select)
}
pure $ applyArrayOnSelect selectWithCardinality
where
applyArrayOnSelect :: Select -> Expression
applyArrayOnSelect select =
FunctionExpression (FunctionName "ARRAY" Nothing) [SelectExpression select]
wrapUnnest :: From -> From
wrapUnnest from =
let starSelect =
Select
{ selectTop = NoTop,
selectAsStruct = AsStruct,
selectProjections = pure StarProjection,
selectFrom = from,
selectJoins = [],
selectWhere = Where [],
selectOrderBy = Nothing,
selectOffset = Nothing,
selectGroupBy = [],
selectFinalWantedFields = Nothing,
selectCardinality = Many
}
arraySelect = applyArrayOnSelect starSelect
in FromFunction
Aliased
{ aliasedThing = SelectFromFunction (FunctionName "UNNEST" Nothing) [arraySelect],
aliasedAlias = entityAliasText (fromAlias from)
}
-- | Produce the join for an array aggregate relation. We produce a
-- normal select, but then include join fields. Then downstream, the
-- DataLoader will execute the lhs select and rhs join in separate
-- server queries, then do a Haskell-native join on the join fields.
--
-- See also 'fromArrayRelationSelectG' for similar example.
fromArrayAggregateSelectG ::
Ir.AnnRelationSelectG 'BigQuery (Ir.AnnAggregateSelectG 'BigQuery Void Expression) ->
ReaderT EntityAlias FromIr Join
fromArrayAggregateSelectG annRelationSelectG = do
joinFieldName <- lift (fromRelName _aarRelationshipName)
Bigquery/fix limit offset for array aggregates Blocked on https://github.com/hasura/graphql-engine-mono/pull/1640. While fiddling with BigQuery I noticed a severe issue with offset/limit for array-aggregates. I've fixed it now. The basic problem was that I was using a query like this: ```graphql query MyQuery { hasura_Artist(order_by: {artist_self_id: asc}) { artist_self_id albums_aggregate(order_by: {album_self_id: asc}, limit: 2) { nodes { album_self_id } aggregate { count } } } } ``` Producing this SQL: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT * FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST -- PROBLEM HERE LIMIT @param0) AS `t_Album1` GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` Note the `LIMIT @param0` -- that is incorrect because we want to limit per artist. Instead, we want: ```sql SELECT `t_Artist1`.`artist_self_id` AS `artist_self_id`, STRUCT(IFNULL(`aa_albums1`.`nodes`, NULL) AS `nodes`, IFNULL(`aa_albums1`.`aggregate`, STRUCT(0 AS `count`)) AS `aggregate`) AS `albums_aggregate` FROM `hasura`.`Artist` AS `t_Artist1` LEFT OUTER JOIN (SELECT ARRAY_AGG(STRUCT(`t_Album1`.`album_self_id` AS `album_self_id`) ORDER BY (`t_Album1`.`album_self_id`) ASC) AS `nodes`, STRUCT(COUNT(*) AS `count`) AS `aggregate`, `t_Album1`.`artist_other_id` AS `artist_other_id` FROM (SELECT *, -- ADDED ROW_NUMBER() OVER(PARTITION BY artist_other_id) artist_album_index FROM `hasura`.`Album` AS `t_Album1` ORDER BY (`t_Album1`.`album_self_id`) ASC NULLS FIRST ) AS `t_Album1` -- CHANGED WHERE artist_album_index <= @param GROUP BY `t_Album1`.`artist_other_id`) AS `aa_albums1` ON (`aa_albums1`.`artist_other_id` = `t_Artist1`.`artist_self_id`) ORDER BY (`t_Artist1`.`artist_self_id`) ASC NULLS FIRST ``` That serves both the LIMIT/OFFSET function in the where clause. Then, both the ARRAY_AGG and the COUNT are correct per artist. I've updated my Haskell test suite to add regression tests for this. I'll push a commit for Python tests shortly. The tests still pass there. This just fixes a case that we hadn't noticed. https://github.com/hasura/graphql-engine-mono/pull/1641 GitOrigin-RevId: 49933fa5e09a9306c89565743ecccf2cb54eaa80
2021-07-06 11:28:42 +03:00
select <- do
lhsEntityAlias <- ask
lift (fromSelectAggregate (pure (lhsEntityAlias, mapping)) annSelectG)
alias <- lift (generateEntityAlias (ArrayAggregateTemplate joinFieldName))
joinOn <- fromMappingFieldNames alias mapping
joinFields <- fromMappingFieldNames (fromAlias (selectFrom select)) mapping
joinFieldProjections <- toNonEmpty (map prepareJoinFieldProjection joinFields)
let projections = selectProjections select <> joinFieldProjections
joinSelect =
select
{ selectWhere = selectWhere select,
selectGroupBy = map fst joinFields,
selectProjections = projections
}
pure
Join
{ joinAlias = alias,
joinSource = JoinSelect joinSelect,
joinRightTable = fromAlias (selectFrom select),
joinOn,
joinProvenance =
ArrayAggregateJoinProvenance $
mapMaybe (\p -> (,aggregateProjectionsFieldOrigin p) <$> projectionAlias p) . toList . selectProjections $ select,
-- Above: Needed by DataLoader to determine the type of
-- Haskell-native join to perform.
joinFieldName,
joinExtractPath = Nothing
}
where
Ir.AnnRelationSelectG
{ _aarRelationshipName,
_aarColumnMapping = mapping :: HashMap ColumnName ColumnName,
_aarAnnSelect = annSelectG
} = annRelationSelectG
-- | Produce a join for an array relation.
--
-- Array relations in PG/MSSQL are expressed using LEFT OUTER JOIN
-- LATERAL or OUTER APPLY, which are essentially producing for each
-- row on the left an array of the result from the right. Which is
-- absolutely what you want for the array relationship.
--
-- BigQuery doesn't support that. Therefore we are instead performing
-- one big array aggregation, for ALL rows in the table - there is no
-- join occurring on the left-hand-side table, grouped by join
-- fields. The data-loader will perform the LHS query and the RHS query
-- separately.
--
-- What we do have is a GROUP BY and make sure that the join fields
-- are included in the output. Finally, in the
-- DataLoader.Plan/DataLoader.Execute, we implement a Haskell-native
-- join of the left-hand-side table and the right-hand-side table.
--
-- Data looks like:
--
-- join_field_a | join_field_b | aggFieldName (array type)
-- 1 | 1 | [ { x: 1, y: 2 }, ... ]
-- 1 | 2 | [ { x: 1, y: 2 }, ... ]
--
-- etc.
--
-- We want to produce a query that looks like:
--
-- SELECT artist_other_id, -- For joining.
--
-- array_agg(struct(album_self_id, title)) as aggFieldName
--
-- -- ^ Aggregating the actual data.
--
-- FROM (SELECT *, -- Get everything, plus the row number:
--
-- ROW_NUMBER() OVER(PARTITION BY artist_other_id) artist_album_index
--
-- FROM hasura.Album
-- ORDER BY album_self_id ASC
--
-- -- ^ Order by here is important for stable results. Any
-- order by clauses for the album should appear here, NOT IN
-- THE ARRAY_AGG.
--
-- )
--
-- AS indexed_album
--
-- WHERE artist_album_index > 1
-- -- ^ Here is where offsetting occurs.
--
-- GROUP BY artist_other_id
-- -- ^ Group by for joining.
--
-- ORDER BY artist_other_id;
-- ^ Ordering for the artist table should appear here.
--
-- Note: if original select already uses a PARTITION BY internally (for distinct_on)
-- join fields are added to partition expressions to give proper semantics of distinct_on
-- combined with an array relation
fromArrayRelationSelectG ::
Ir.ArrayRelationSelectG 'BigQuery Void Expression ->
ReaderT EntityAlias FromIr Join
fromArrayRelationSelectG annRelationSelectG = do
pselect <- (lift . flip fromSelectRows annSelectG . ParentEntityAlias) =<< ask -- Take the original select.
joinFieldName <- lift (fromRelName _aarRelationshipName)
alias <- lift (generateEntityAlias (ArrayRelationTemplate joinFieldName))
indexAlias <- lift (generateEntityAlias IndexTemplate)
joinOn <- fromMappingFieldNames alias mapping
joinFields <- fromMappingFieldNames (fromAlias (pselectFrom pselect)) mapping >>= toNonEmpty
let select = withExtraPartitionFields pselect $ NE.toList (fmap fst joinFields)
joinFieldProjections = fmap prepareJoinFieldProjection joinFields
let joinSelect =
Select
{ selectCardinality = One,
selectAsStruct = NoAsStruct,
selectFinalWantedFields = selectFinalWantedFields select,
selectTop = NoTop,
selectProjections =
joinFieldProjections
<> pure
( ArrayAggProjection
Aliased
{ aliasedThing =
ArrayAgg
{ arrayAggProjections =
fmap
(aliasToFieldProjection (fromAlias (selectFrom select)))
(selectProjections select),
arrayAggOrderBy = selectOrderBy select,
arrayAggTop = selectTop select
-- The sub-select takes care of caring about global top.
--
-- This handles the LIMIT need.
},
aliasedAlias = aggFieldName
}
),
selectFrom =
FromSelect
( Aliased
{ aliasedAlias = coerce (fromAlias (selectFrom select)),
aliasedThing =
Select
{ selectProjections =
selectProjections select
<> joinFieldProjections
`appendToNonEmpty` foldMap @Maybe
( map \OrderBy {orderByFieldName} ->
FieldNameProjection
Aliased
{ aliasedThing = orderByFieldName,
aliasedAlias = fieldName orderByFieldName
}
)
(toList <$> selectOrderBy select)
-- Above: Select "order by" fields as they're being used
-- inside `ARRAY_AGG` function (as ORDER BY clause)
<> pure
( WindowProjection
( Aliased
{ aliasedAlias = unEntityAlias indexAlias,
aliasedThing =
RowNumberOverPartitionBy
-- The row numbers start from 1.
( fmap fst joinFields
)
(selectOrderBy select)
-- Above: Having the order by
-- in here ensures that the
-- row numbers are ordered by
-- this ordering. Below, we
-- order again for the
-- general row order. Both
-- are needed!
}
)
),
selectFrom = selectFrom select,
selectJoins = selectJoins select,
selectWhere = selectWhere select,
selectOrderBy = selectOrderBy select,
-- Above: This orders the rows themselves. In
-- the RowNumberOverPartitionBy, we also set
-- a row order for the calculation of the
-- indices. Both are needed!
selectOffset = Nothing,
selectFinalWantedFields =
selectFinalWantedFields select,
selectCardinality = Many,
selectAsStruct = NoAsStruct,
selectTop = NoTop,
selectGroupBy = mempty
}
}
),
selectWhere =
case selectOffset select of
Nothing -> mempty
Just offset ->
Where
[ OpExpression
MoreOp
(ColumnExpression FieldName {fieldNameEntity = coerce (fromAlias (selectFrom select)), fieldName = unEntityAlias indexAlias})
offset
],
selectOrderBy = Nothing, -- Not needed.
selectJoins = mempty,
selectOffset = Nothing,
-- This group by corresponds to the field name projections above. E.g. artist_other_id
selectGroupBy = map fst (NE.toList joinFields)
}
pure
Join
{ joinAlias = alias,
joinSource = JoinSelect joinSelect,
joinRightTable = fromAlias (selectFrom select),
joinOn,
joinProvenance =
ArrayJoinProvenance
( if True
then (fromMaybe [] (selectFinalWantedFields select))
else
( mapMaybe
projectionAlias
(toList (selectProjections select))
)
),
-- Above: Needed by DataLoader to determine the type of
-- Haskell-native join to perform.
joinFieldName,
joinExtractPath = Just aggFieldName
}
where
Ir.AnnRelationSelectG
{ _aarRelationshipName,
_aarColumnMapping = mapping :: HashMap ColumnName ColumnName,
_aarAnnSelect = annSelectG
} = annRelationSelectG
-- | For entity projections, convert any entity aliases to their field
-- names. ArrayEntityProjection and ExpressionProjection get converted
-- to aliases to fields with the same names as all the expressions
-- have already aliases applied in select from ArrayAgg
-- (created in Hasura.Backends.BigQuery.ToQuery.fromArrayAgg)
aliasToFieldProjection :: EntityAlias -> Projection -> Projection
aliasToFieldProjection (EntityAlias selectAlias) =
\case
EntityProjection Aliased {aliasedAlias = name, aliasedThing = fields} ->
EntityProjection
Aliased
{ aliasedAlias = name,
aliasedThing =
fmap
(\(FieldName {..}, origin) -> (FieldName {fieldNameEntity = name, ..}, origin))
fields
}
ArrayEntityProjection _ aliased ->
aliasColumn aliased
ExpressionProjection aliased ->
aliasColumn aliased
p -> p
where
aliasColumn :: Aliased a -> Projection
aliasColumn aliased =
ExpressionProjection
aliased
{ aliasedThing = ColumnExpression (FieldName {fieldName = aliasedAlias aliased, fieldNameEntity = selectAlias})
}
fromRelName :: Rql.RelName -> FromIr Text
fromRelName relName =
pure (Rql.relNameToTxt relName)
-- | The context given by the reader is of the previous/parent
-- "remote" table. The WHERE that we're generating goes in the child,
-- "local" query. The @From@ passed in as argument is the local table.
--
-- We should hope to see e.g. "post.category = category.id" for a
-- local table of post and a remote table of category.
--
-- The left/right columns in @HashMap ColumnName ColumnName@ corresponds
-- to the left/right of @select ... join ...@. Therefore left=remote,
-- right=local in this context.
fromMapping ::
From ->
HashMap ColumnName ColumnName ->
ReaderT EntityAlias FromIr [Expression]
fromMapping localFrom =
traverse
( \(remoteColumn, localColumn) -> do
localFieldName <- local (const (fromAlias localFrom)) (fromColumn localColumn)
remoteFieldName <- fromColumn remoteColumn
pure
( EqualExpression
(ColumnExpression localFieldName)
(ColumnExpression remoteFieldName)
)
)
. HM.toList
-- | Given an alias for the remote table, and a map of local-to-remote column
-- name pairings, produce 'FieldName' pairings (column names paired with their
-- associated table names).
--
-- For example, we might convert the following:
--
-- @
-- [ ( ColumnName { columnName = "author_id" }
-- , ColumnName { columnName = "id" }
-- )
-- ]
-- @
--
-- ... into something like this:
--
-- @
-- ( FieldName
-- { fieldName = "id"
-- , fieldNameEntity = "t_author1"
-- }
-- , FieldName
-- { fieldName = "author_id"
-- , fieldNameEntity = "t_article1"
-- }
-- )
-- @
--
-- Note that the columns __flip around__ for the output. The input map is
-- @(local, remote)@.
fromMappingFieldNames ::
EntityAlias ->
HashMap ColumnName ColumnName ->
ReaderT EntityAlias FromIr [(FieldName, FieldName)]
fromMappingFieldNames remoteFrom = traverse go . HM.toList
where
go (localColumn, remoteColumn) = do
remoteFieldName <- local (const remoteFrom) (fromColumn remoteColumn)
localFieldName <- fromColumn localColumn
pure (remoteFieldName, localFieldName)
--------------------------------------------------------------------------------
-- Basic SQL expression types
fromOpExpG :: Expression -> Ir.OpExpG 'BigQuery Expression -> FromIr Expression
fromOpExpG expression op =
case op of
Ir.ANISNULL -> pure (IsNullExpression expression)
Ir.ANISNOTNULL -> pure (IsNotNullExpression expression)
Ir.AEQ False val -> pure (nullableBoolEquality expression val)
Ir.AEQ True val -> pure (EqualExpression expression val)
Ir.ANE False val -> pure (nullableBoolInequality expression val)
Ir.ANE True val -> pure (NotEqualExpression expression val)
Ir.AIN val -> pure (OpExpression InOp expression val)
Ir.ANIN val -> pure (OpExpression NotInOp expression val)
Ir.AGT val -> pure (OpExpression MoreOp expression val)
Ir.ALT val -> pure (OpExpression LessOp expression val)
Ir.AGTE val -> pure (OpExpression MoreOrEqualOp expression val)
Ir.ALTE val -> pure (OpExpression LessOrEqualOp expression val)
Ir.ACast _casts -> refute (pure (UnsupportedOpExpG op)) -- mkCastsExp casts
Ir.ALIKE val -> pure (OpExpression LikeOp expression val)
Ir.ANLIKE val -> pure (OpExpression NotLikeOp expression val)
Ir.ABackendSpecific op' -> pure (fromBackendSpecificOpExpG expression op')
Ir.CEQ _rhsCol -> refute (pure (UnsupportedOpExpG op)) -- S.BECompare S.SEQ lhs $ mkQCol rhsCol
Ir.CNE _rhsCol -> refute (pure (UnsupportedOpExpG op)) -- S.BECompare S.SNE lhs $ mkQCol rhsCol
Ir.CGT _rhsCol -> refute (pure (UnsupportedOpExpG op)) -- S.BECompare S.SGT lhs $ mkQCol rhsCol
Ir.CLT _rhsCol -> refute (pure (UnsupportedOpExpG op)) -- S.BECompare S.SLT lhs $ mkQCol rhsCol
Ir.CGTE _rhsCol -> refute (pure (UnsupportedOpExpG op)) -- S.BECompare S.SGTE lhs $ mkQCol rhsCol
Ir.CLTE _rhsCol -> refute (pure (UnsupportedOpExpG op)) -- S.BECompare S.SLTE lhs $ mkQCol rhsCol
-- These are new as of 2021-02-18 to this API. Not sure what to do with them at present, marking as unsupported.
fromBackendSpecificOpExpG :: Expression -> BigQuery.BooleanOperators Expression -> Expression
fromBackendSpecificOpExpG expression op =
let func name val = FunctionExpression (FunctionName name Nothing) [expression, val]
in case op of
BigQuery.ASTContains v -> func "ST_CONTAINS" v
BigQuery.ASTEquals v -> func "ST_EQUALS" v
BigQuery.ASTTouches v -> func "ST_TOUCHES" v
BigQuery.ASTWithin v -> func "ST_WITHIN" v
BigQuery.ASTIntersects v -> func "ST_INTERSECTS" v
BigQuery.ASTDWithin (Ir.DWithinGeogOp r v sph) ->
FunctionExpression (FunctionName "ST_DWITHIN" Nothing) [expression, v, r, sph]
nullableBoolEquality :: Expression -> Expression -> Expression
nullableBoolEquality x y =
OrExpression
[ EqualExpression x y,
AndExpression [IsNullExpression x, IsNullExpression y]
]
nullableBoolInequality :: Expression -> Expression -> Expression
nullableBoolInequality x y =
OrExpression
[ NotEqualExpression x y,
AndExpression [IsNotNullExpression x, IsNullExpression y]
]
fromGBoolExp :: Ir.GBoolExp 'BigQuery Expression -> ReaderT EntityAlias FromIr Expression
fromGBoolExp =
\case
Ir.BoolAnd expressions ->
fmap AndExpression (traverse fromGBoolExp expressions)
Ir.BoolOr expressions ->
fmap OrExpression (traverse fromGBoolExp expressions)
Ir.BoolNot expression -> fmap NotExpression (fromGBoolExp expression)
Ir.BoolExists gExists -> fmap ExistsExpression (fromGExists gExists)
Ir.BoolFld expression -> pure expression
--------------------------------------------------------------------------------
-- Misc combinators
-- | Attempt to refine a list into a 'NonEmpty'. If the given list is empty,
-- this will 'refute' the computation with an 'UnexpectedEmptyList' error.
toNonEmpty :: MonadValidate (NonEmpty Error) m => [x] -> m (NonEmpty x)
toNonEmpty = \case
[] -> refute (UnexpectedEmptyList :| [])
x : xs -> pure (x :| xs)
-- | Get the remote field from a pair (see 'fromMappingFieldNames' for more
-- information) and produce a 'Projection'.
prepareJoinFieldProjection :: (FieldName, FieldName) -> Projection
prepareJoinFieldProjection (fieldName', _) =
FieldNameProjection
Aliased
{ aliasedThing = fieldName',
aliasedAlias = fieldName fieldName'
}
selectProjectionsFromFieldSources :: Bool -> [FieldSource] -> FromIr (NonEmpty Projection)
selectProjectionsFromFieldSources keepJoinField fieldSources = do
projections <- tolerate do
projections' <- traverse (fieldSourceProjections keepJoinField) fieldSources
toNonEmpty projections'
case projections of
Just (x :| xs) -> pure (foldl' (<>) x xs)
Nothing -> refute (pure NoProjectionFields)
trueExpression :: Expression
trueExpression = ValueExpression (BoolValue True)
--------------------------------------------------------------------------------
-- Constants
aggFieldName :: Text
aggFieldName = "agg"
existsFieldName :: Text
existsFieldName = "exists_placeholder"
--------------------------------------------------------------------------------
-- Name generation
data NameTemplate
= ArrayRelationTemplate Text
| ArrayAggregateTemplate Text
| ObjectRelationTemplate Text
| TableTemplate Text
| ForOrderAlias Text
| IndexTemplate
| UnnestTemplate
| FunctionTemplate FunctionName
generateEntityAlias :: NameTemplate -> FromIr EntityAlias
generateEntityAlias template = do
FromIr
( modify'
( \FromIrState {..} ->
FromIrState {indices = M.insertWith (+) prefix start indices, ..}
)
)
i <- FromIr (gets indices)
pure (EntityAlias (prefix <> tshow (fromMaybe start (M.lookup prefix i))))
where
start = 1
prefix = T.take 20 rendered
rendered =
case template of
ArrayRelationTemplate sample -> "ar_" <> sample
ArrayAggregateTemplate sample -> "aa_" <> sample
ObjectRelationTemplate sample -> "or_" <> sample
TableTemplate sample -> "t_" <> sample
ForOrderAlias sample -> "order_" <> sample
IndexTemplate -> "idx"
UnnestTemplate -> "unnest"
FunctionTemplate FunctionName {..} -> functionName
fromAlias :: From -> EntityAlias
fromAlias (FromQualifiedTable Aliased {aliasedAlias}) = EntityAlias aliasedAlias
fromAlias (FromSelect Aliased {aliasedAlias}) = EntityAlias aliasedAlias
fromAlias (FromSelectJson Aliased {aliasedAlias}) = EntityAlias aliasedAlias
fromAlias (FromFunction Aliased {aliasedAlias}) = EntityAlias aliasedAlias
fieldTextNames :: Ir.AnnFieldsG 'BigQuery Void Expression -> [Text]
fieldTextNames = fmap (\(Rql.FieldName name, _) -> name)
unEntityAlias :: EntityAlias -> Text
unEntityAlias (EntityAlias t) = t
--------------------------------------------------------------------------------
-- Global limit support
getGlobalTop :: FromIr Top
getGlobalTop =
FromIr
( asks
( \FromIrReader {config = FromIrConfig {globalSelectLimit}} ->
globalSelectLimit
)
)