Orden lento por SQL Server

En mi aplicación, tengo una consulta que realiza una búsqueda en la tabla "archivos".

La tabla "archivos" está dividida por "f". "Creado" (vea la definición de la tabla y tiene ~ 26 millones de filas para el cliente 19 ("f". "Cid = 19).

El punto aquí es, si hago esta consulta:

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;

Obtengo los resultados en 0 segundos, con el siguiente plan de ejecución: https://www.brentozar.com/pastetheplan/?id=SkV0-FDcG

Pero si trato de ordenar por "nombre", la consulta se vuelve demasiado lenta:

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;

Esta consulta tarda 11 minutos en devolverme los resultados, con el siguiente plan de ejecución: https://www.brentozar.com/pastetheplan/?id=Sk3Fbtv9M

Además, si cambio el orden por columna, el resultado es el mismo.

Como puede ver en el plan de ejecución, el índice "files_mv" tiene un costo del 61%, esta es la definición del índice:

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

Yo uso SQL Server con Azure. Específicamente, Azure SQL Database con nivel de precios / modelo "S4 Estándar (200 DTU)".

Estoy recibiendo muchos datos, pero creo que la conexión a Internet no es un cuello de botella, porque en otras consultas también recibo muchos datos y son más rápidos.

Además, estoy haciendo una inserción masiva de datos en estas tablas, y en unos días tendré hasta ~ 240 millones de filas en la tabla de archivos (para un cid) y ~ 480 millones de filas en la tabla value_number.

información adicional

Función de partición "PF_files_partuling":

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')

Esquema de partición "PS_files_partitions":

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

** Tendré alrededor de 15 millones de filas en cada partición.

tabla de archivos:

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)
)

tabla 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)
)

Tabla de índices de archivos

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])

Índices de la tabla 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)

Las estadísticas están actualizadas. He cambiado los tipos de datos para el año y otras columnas y ahora el rendimiento parece ser un poco mejor, pero el plan de ejecución sigue siendo el mismo. Estoy tratando de arreglar la estimación de cardinalidad (índice cambiante), pero aún no tuve éxito. De acuerdo con la documentación de Azure, debería tener un nivel de compatibilidad de 130 en la base de datos y ya tengo 100 con ProductVersion 12.0.

Análisis

Para la consulta sin preferencia de orden, SQL Server puede transmitir filas agrupadas utilizando un Hash Match Flow Distinct. Si encuentra rápidamente el número requerido de entradas distintas, el tiempo de ejecución es corto.

Cuando se requiere un pedido específico, SQL Server debe probar cada fila. Por ejemplo, para colocar filas en nameorden, debe ordenar todas las filas por nombre. Esto será lento si hay muchas filas y ningún índice para proporcionar ese orden sin ordenar.

Hay una serie de complicaciones fundamentales en su caso, especialmente la partición y la disyunción [year]. La partición significa que sus índices no pueden entregar el pedido que podría esperar. Por ejemplo, un índice nameactivado se ordena primero por número de partición, luego por nombre. No puede entregar filas ordenadas namesolas.

También has FORCED PARAMETERIZATIONestablecido. Esto puede ser beneficioso en general, pero viene con impactos que debe comprender completamente. Eso combinado con la partición y los índices de múltiples columnas significa que sus estadísticas son en gran medida inútiles.

La disyunción en yeartambién se mete con los pedidos, y significa que SQL Server solo puede buscar year >= 0 and year <= 2013en su plan. Esto es mucho menos selectivo que buscar year = 0y por year = 2013separado.

Recomendaciones

Entonces, a la luz de todo lo anterior:

Un buen índice para la ORDER BY nameconsulta es:

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 mejor índice para la value_numbertabla es:

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

La consulta se puede escribir para obtener como máximo 50 filas para cada año y partición. Luego tomamos los primeros 50 en orden del conjunto combinado:

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;

El plan de ejecución ordenará como máximo 100 filas de la filestabla:

Generalmente se requeriría un índice diferente para cada pedido diferente.

Si necesita contar el número total de resultados, use:

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 <> violín