web-dev-qa-db-fra.com

Ordre lent par SQL Server

Dans mon application, j'ai une requête qui effectue une recherche dans le tableau "fichiers".

La table "fichiers" est partitionnée par "f". "Créée" (voir la définition de la table et contient ~ 26 millions de lignes pour le client 19 ("f". "Cid = 19).

Le point ici est, si je fais cette requête:

SELECT "f"."id" AS "FileId"  
, "f"."name" AS "FileName"  
, "f"."year" AS "Fileyear"  
, "f"."cid" AS "clientId"
, "f"."created" AS "FileDate"
, CASE WHEN ("vnVE0"."value" is not null AND "vnVE0"."value" != '')                           
                                THEN CAST("vnVE0"."value" AS decimal(28,2))
                                ELSE 0 END AS "keywordValueCol0_numeric"
FROM files "f"  
OUTER APPLY
(
    SELECT DISTINCT
        VT.[value]
    FROM dbo.value_number AS VT
    WHERE
        VT.id_file = F.id
        AND VT.id_field = 260
) AS "vnVE0"
WHERE "grapado" IS NULL AND "masterversion" IS NULL AND ("f"."year" = 2013 OR "f"."year" = 0) AND "f"."cid" = 19
GROUP BY "f"."id", "f"."name", "f"."year", "f"."cid", "f"."created", CASE WHEN ("vnVE0"."value" is not null AND "vnVE0"."value" != '')                           
                            THEN CAST("vnVE0"."value" AS decimal(28,2))
                            ELSE 0 END
ORDER BY (SELECT NULL)
OFFSET 0 ROWS FETCH NEXT 50 ROWS ONLY;

J'obtiens les résultats en 0 secondes, avec le plan d'exécution suivant: https://www.brentozar.com/pastetheplan/?id=SkV0-FDcG

Query without ORDER BY

Mais si j'essaye de commander par "nom" la requête devient trop lente:

SELECT "f"."id" AS "FileId"  
, "f"."name" AS "FileName"  
, "f"."year" AS "Fileyear"  
, "f"."cid" AS "clientId"
, "f"."created" AS "FileDate"
, CASE WHEN ("vnVE0"."value" is not null AND "vnVE0"."value" != '')                           
                                THEN CAST("vnVE0"."value" AS decimal(28,2))
                                ELSE 0 END AS "keywordValueCol0_numeric"
FROM files "f"  
OUTER APPLY
(
    SELECT DISTINCT
        VT.[value]
    FROM dbo.value_number AS VT
    WHERE
        VT.id_file = F.id
        AND VT.id_field = 260
) AS "vnVE0"
WHERE "grapado" IS NULL AND "masterversion" IS NULL AND ("f"."year" = 2013 OR "f"."year" = 0) AND "f"."cid" = 19
GROUP BY "f"."id", "f"."name", "f"."year", "f"."cid", "f"."created", CASE WHEN ("vnVE0"."value" is not null AND "vnVE0"."value" != '')                           
                            THEN CAST("vnVE0"."value" AS decimal(28,2))
                            ELSE 0 END
ORDER BY "f"."name"
OFFSET 0 ROWS FETCH NEXT 50 ROWS ONLY;

Cette requête prend 11 minutes pour me renvoyer les résultats, avec le plan d'exécution suivant: https://www.brentozar.com/pastetheplan/?id=Sk3Fbtv9M

Query with ORDER BY

De plus, si je modifie l'ordre par colonne, le résultat est le même.

Comme vous pouvez le voir dans le plan d'exécution, l'index "files_mv" a un coût de 61%, voici la définition de l'index:

CREATE NONCLUSTERED INDEX [files_mv] ON [dbo].[files]
(
    [masterversion] ASC,
    [year] ASC,
    [cat_id] ASC,
    [cid] ASC,
    [eid] ASC,
    [grapado] ASC,
    [sub_id] ASC,
    [tip_id] ASC
)
INCLUDE (   [id],
    [name]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
GO

J'utilise SQL Server avec Azure. Plus précisément, Azure SQL Database avec tarification/niveau de modèle "S4 Estándar (200 DTU)".

Je reçois beaucoup de données, mais je pense que la connexion Internet n'est pas un goulot d'étranglement, car dans d'autres requêtes, je reçois également beaucoup de données et elles sont plus rapides.

De plus, je fais une insertion massive de données dans ces tables, et dans quelques jours, j'aurai jusqu'à ~ 240 millions de lignes dans la table des fichiers (pour un cid) et ~ 480 millions de lignes dans la table value_number.

Informations supplémentaires

Fonction de partition "PF_files_partitioning":

CREATE PARTITION FUNCTION PF_files_partitioning (DATETIME2(7))
AS
RANGE LEFT FOR VALUES ( '2013-03-31 23:59:59', 
                        '2013-06-30 23:59:59',
                        '2013-09-30 23:59:59',
                        '2013-12-31 23:59:59',
                        '2014-03-31 23:59:59', 
                        '2014-06-30 23:59:59',
                        '2014-09-30 23:59:59',
                        '2014-12-31 23:59:59',
                        '2015-03-31 23:59:59', 
                        '2015-06-30 23:59:59',
                        '2015-09-30 23:59:59',
                        '2015-12-31 23:59:59',
                        '2016-03-31 23:59:59', 
                        '2016-06-30 23:59:59',
                        '2016-09-30 23:59:59',
                        '2016-12-31 23:59:59',
                        '2017-03-31 23:59:59', 
                        '2017-06-30 23:59:59',
                        '2017-09-30 23:59:59',
                        '2017-12-31 23:59:59',
                        '2018-03-31 23:59:59')

Schéma de partition "PS_files_partitioning":

CREATE PARTITION SCHEME PS_files_partitioning AS PARTITION PF_files_partitioning ALL TO ([PRIMARY]);

** J'aurai environ 15 millions de lignes dans chaque partition.

table des fichiers:

CREATE TABLE [dbo].[files](
    [id] [bigint] IDENTITY(1,1) NOT NULL,
    [cid] [tinyint] NOT NULL,
    [eid] [bigint] NOT NULL,
    [cat_id] [bigint] NOT NULL,
    [tip_id] [bigint] NULL,
    [sub_id] [bigint] NULL,
    [year] [smallint] NOT NULL,
    [caducidad] [smallint] NULL,
    [grapadopri] [int] NOT NULL,
    [grapado] [bigint] NULL,
    [name] [nvarchar](255) NOT NULL,
    [extension] [tinyint] NOT NULL,
    [size] [bigint] NOT NULL,
    [id_doc] [bit] NOT NULL,
    [observaciones] [nvarchar](255) NOT NULL,
    [indexed] [bit] NOT NULL,
    [signed] [bit] NOT NULL,
    [created] [datetime2](7) NOT NULL,
    [name_lower] [nvarchar](255) NOT NULL,
    [modified] [datetime2](7) NULL,
    [related] [bit] NOT NULL,
    [masterversion] [bigint] NULL,
    [versioned] [bit] NOT NULL,
    [hwsignature] [tinyint] NOT NULL,
    [blockedUserId] [smallint] NULL,
 CONSTRAINT [PK_files_id] PRIMARY KEY CLUSTERED 
(
    [id] ASC,
    [created] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created]),
 CONSTRAINT [files$estructure_unique] UNIQUE NONCLUSTERED 
(
    [cat_id] ASC,
    [tip_id] ASC,
    [sub_id] ASC,
    [year] ASC,
    [name] ASC,
    [grapado] ASC,
    [created] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
)

table value_number:

CREATE TABLE [dbo].[value_number](
    [id] [bigint] IDENTITY(1,1) NOT NULL,
    [id_file] [bigint] NOT NULL DEFAULT ((0)),
    [id_field] [bigint] NOT NULL DEFAULT ((0)),
    [value] [nvarchar](255) NULL DEFAULT (NULL),
    [id_doc] [bigint] NULL DEFAULT (NULL)
 CONSTRAINT [PK_value_number_id] PRIMARY KEY CLUSTERED 
(
    [id] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
)

Table des index des fichiers

CREATE NONCLUSTERED INDEX [files_clientes] ON [dbo].[files]
(
    [cid] ASC
)
INCLUDE ([id]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [files_grapado] ON [dbo].[files]
(
    [grapado] ASC
)
INCLUDE (   [id],
    [name]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [files_mv] ON [dbo].[files]
(
    [masterversion] ASC,
    [year] ASC,
    [cat_id] ASC,
    [cid] ASC,
    [eid] ASC,
    [grapado] ASC,
    [sub_id] ASC,
    [tip_id] ASC
)
INCLUDE (   [id],
    [name]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [files_ocr] ON [dbo].[files]
(
    [cid] ASC,
    [grapado] ASC,
    [indexed] ASC,
    [masterversion] ASC,
    [extension] ASC
)
INCLUDE (   [id],
    [eid],
    [cat_id],
    [tip_id],
    [sub_id],
    [year],
    [name]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [files_ocr2] ON [dbo].[files]
(
    [cid] ASC,
    [eid] ASC,
    [grapado] ASC,
    [indexed] ASC,
    [masterversion] ASC,
    [extension] ASC
)
INCLUDE (   [id],
    [cat_id],
    [tip_id],
    [sub_id],
    [year],
    [name]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [files_ocr3] ON [dbo].[files]
(
    [cid] ASC,
    [cat_id] ASC,
    [grapado] ASC,
    [indexed] ASC,
    [masterversion] ASC,
    [extension] ASC
)
INCLUDE (   [eid],
    [tip_id],
    [sub_id],
    [year],
    [name]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [busqueda_name] ON [dbo].[files]
(
    [cid] ASC,
    [eid] ASC,
    [grapado] ASC,
    [year] ASC
)
INCLUDE (   [id],
    [cat_id],
    [tip_id],
    [sub_id],
    [grapadopri],
    [name],
    [size],
    [id_doc],
    [signed],
    [created],
    [modified],
    [related],
    [masterversion]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [busqueda2] ON [dbo].[files]
(
    [cid] ASC,
    [eid] ASC,
    [cat_id] ASC,
    [grapado] ASC,
    [masterversion] ASC,
    [year] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [cid] ON [dbo].[files]
(
    [cid] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [eid] ON [dbo].[files]
(
    [eid] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [extension] ON [dbo].[files]
(
    [extension] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [FK_files_archivo] ON [dbo].[files]
(
    [grapado] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [FK_files_tipo] ON [dbo].[files]
(
    [tip_id] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [grapadopri] ON [dbo].[files]
(
    [grapadopri] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [index_all] ON [dbo].[files]
(
    [cid] ASC,
    [eid] ASC,
    [grapado] ASC,
    [masterversion] ASC
)
INCLUDE (   [cat_id],
    [tip_id],
    [sub_id],
    [year],
    [grapadopri],
    [name],
    [size],
    [id_doc],
    [signed],
    [created],
    [modified],
    [related],
    [versioned]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [missing_index_7_6] ON [dbo].[files]
(
    [cid] ASC,
    [eid] ASC,
    [grapado] ASC,
    [name] ASC,
    [year] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) 
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [ocrCloudClients] ON [dbo].[files]
(
    [grapado] ASC,
    [indexed] ASC,
    [extension] ASC
)
INCLUDE (   [cid],
    [eid],
    [cat_id],
    [tip_id],
    [sub_id]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [searchEntity] ON [dbo].[files]
(
    [cid] ASC,
    [eid] ASC,
    [grapado] ASC,
    [masterversion] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

CREATE NONCLUSTERED INDEX [sub_id] ON [dbo].[files]
(
    [sub_id] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)
ON PS_files_partitioning([created])

Index de la table value_number

CREATE NONCLUSTERED INDEX [searchValues] ON [dbo].[value_number]
(
    [id_field] ASC
)
INCLUDE (   [id_file],
    [value]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)

CREATE NONCLUSTERED INDEX [search] ON [dbo].[value_number]
(
    [id_file] ASC,
    [id_field] ASC
)
INCLUDE (   [value]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)

CREATE NONCLUSTERED INDEX [id_field] ON [dbo].[value_number]
(
    [id_field] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)

CREATE NONCLUSTERED INDEX [FK_valueesN_documento] ON [dbo].[value_number]
(
    [id_doc] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)

CREATE NONCLUSTERED INDEX [FK_valueesN_archivo] ON [dbo].[value_number]
(
    [id_file] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)

Les statistiques sont à jour. J'ai changé les types de données pour l'année et d'autres colonnes et maintenant les performances semblent être un peu meilleures, mais le plan d'exécution est toujours le même. J'essaie de corriger l'estimation de cardinalité (changement d'index), mais je n'ai pas encore réussi. Selon la documentation Azure, je devrais avoir 130 niveaux de compatibilité dans la base de données et j'en ai déjà 100 avec ProductVersion 12.0.

12
RuSSe

Une analyse

Pour la requête sans préférence d'ordre, SQL Server peut diffuser des lignes groupées à l'aide d'un Hash Match Flow Distinct. S'il rencontre rapidement le nombre requis d'entrées distinctes, le temps d'exécution est court.

Lorsqu'une commande spécifique est requise, SQL Server doit tester chaque ligne. Par exemple, pour placer les lignes dans l'ordre name, il doit trier toutes les lignes par nom. Ce sera lent s'il y a beaucoup de lignes et aucun index pour fournir cet ordre sans trier.

Il y a un certain nombre de complications fondamentales dans votre cas, notamment le partitionnement et la disjonction sur [year]. Le partitionnement signifie que vos index ne peuvent pas livrer la commande que vous attendez. Par exemple, un index sur name est en fait trié d'abord par numéro de partition, puis par nom. Il ne peut pas fournir des lignes triées sur name seul.

Vous avez aussi FORCED PARAMETERIZATION ensemble. Cela peut être bénéfique dans l'ensemble, mais cela comporte des impacts que vous devez bien comprendre. Cela combiné avec le partitionnement et les index multi-colonnes signifie que vos statistiques sont largement inutiles.

La disjonction sur year perturbe également la commande et signifie que SQL Server ne peut rechercher que year >= 0 and year <= 2013 dans votre plan. C'est beaucoup moins sélectif que de rechercher sur year = 0 et year = 2013 séparément.

Recommandations

Donc, à la lumière de tout ce qui précède:

Un bon indice pour le ORDER BY name la requête est:

CREATE INDEX [IX dbo.files cid, year, name : grapado IS NULL AND masterversion IS NULL] 
ON dbo.files (cid, [year], [name])
INCLUDE (grapado, masterversion)
WHERE grapado IS NULL AND masterversion IS NULL;

Un meilleur indice pour le value_number la table est:

CREATE INDEX [IX dbo.value_number id_file, id_field, value] 
ON dbo.value_number (id_file, id_field, [value]);

La requête peut ensuite être écrite pour extraire au plus 50 lignes pour chaque année et partition. Nous prenons ensuite les 50 premiers dans l'ordre de l'ensemble combiné:

WITH PartitionNumbers AS
(
    -- Each partition of the table
    SELECT P.partition_number
    FROM sys.partitions AS P
    WHERE P.[object_id] = OBJECT_ID(N'dbo.files', N'U')
    AND P.index_id = 1
)
SELECT
    FF.id,
    FF.[name],
    FF.[year],
    FF.cid,
    FF.created,
    vnVE0.keywordValueCol0_numeric
FROM PartitionNumbers AS PN
CROSS APPLY
(
    SELECT
        F100.*
    FROM 
    (
        -- 50 rows in order for year 2013
        SELECT
            F.id,
            F.[name],
            F.[year],
            F.cid,
            F.created
        FROM dbo.files AS F
        WHERE
            F.grapado IS NULL
            AND F.masterversion IS NULL
            AND F.[year] = 2013
            AND F.cid = 19
            AND $PARTITION.PF_files_partitioning(F.created) = PN.partition_number
        ORDER BY
            F.[name]
            OFFSET 0 ROWS
            FETCH FIRST 50 ROWS ONLY

        UNION ALL

        -- 50 rows in order for year 0
        SELECT
            F.id,
            F.[name],
            F.[year],
            F.cid,
            F.created
        FROM dbo.files AS F
        WHERE
            F.grapado IS NULL
            AND F.masterversion IS NULL
            AND F.[year] = 0
            AND F.cid = 19
            AND $PARTITION.PF_files_partitioning(F.created) = PN.partition_number
        ORDER BY
            F.[name]
            OFFSET 0 ROWS
            FETCH FIRST 50 ROWS ONLY
    ) AS F100
) AS FF
OUTER APPLY
(
    -- Lookup distinct values
    SELECT
        keywordValueCol0_numeric = 
            CASE
                WHEN VN.[value] IS NOT NULL AND VN.[value] <> ''
                THEN CONVERT(decimal(28, 2), VN.[value])
                ELSE CONVERT(decimal(28, 2), 0)
            END
    FROM dbo.value_number AS VN
    WHERE
        VN.id_file = FF.id
        AND VN.id_field = 260
    GROUP BY
        VN.[value]
) AS vnVE0
ORDER BY
    FF.[name]
    OFFSET 0 ROWS
    FETCH FIRST 50 ROWS ONLY;

Le plan d'exécution triera au plus 100 lignes de la table files:

estimated plan

Un index différent serait généralement requis pour chaque commande différente.

Si vous devez compter le nombre total de résultats, utilisez:

SELECT COUNT_BIG(*) 
FROM dbo.files AS F
OUTER APPLY
(
    SELECT DISTINCT VN.[value]
    FROM dbo.value_number AS VN
    WHERE
        VN.id_file = F.id
        AND VN.id_field = 260
) AS vnVE0
WHERE
    F.grapado IS NULL
    AND F.masterversion IS NULL
    AND F.[year] IN (0, 2013)
    AND F.cid = 19;

db <> violon

9
Paul White 9