Langsame Bestellung von SQL Server


12

In meiner Anwendung habe ich eine Abfrage, die eine Suche in der Tabelle "files" durchführt.

Die Tabelle "files" ist durch "f". "Created" partitioniert (siehe Tabellendefinition) und hat ~ 26 Millionen Zeilen für den Client 19 ("f". "Cid = 19).

Der Punkt hier ist, wenn ich diese Abfrage mache:

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;

Ich erhalte die Ergebnisse in 0 Sekunden mit folgendem Ausführungsplan: https://www.brentozar.com/pastetheplan/?id=SkV0-FDcG

Abfrage ohne ORDER BY

Aber wenn ich versuche, nach "Name" zu bestellen, wird die Abfrage zu langsam:

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;

Diese Abfrage dauert 11 Minuten, um mir die Ergebnisse mit dem folgenden Ausführungsplan zurückzugeben: https://www.brentozar.com/pastetheplan/?id=Sk3Fbtv9M

Abfrage mit ORDER BY

Wenn ich die Reihenfolge nach Spalte ändere, ist das Ergebnis dasselbe.

Wie Sie im Ausführungsplan sehen können, kostet der Index "files_mv" 61%. Dies ist die Indexdefinition:

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

Ich verwende SQL Server mit Azure. Insbesondere Azure SQL-Datenbank mit Preis- / Modellstufe "S4 Estándar (200 DTUs)".

Ich erhalte viele Daten, aber ich denke, dass die Internetverbindung kein Engpass ist, weil ich bei anderen Abfragen auch viele Daten erhalte und sie schneller sind.

Außerdem füge ich Daten in diese Tabellen ein, und in wenigen Tagen werden bis zu 240 Millionen Zeilen in der Dateitabelle (für eine CID) und 480 Millionen Zeilen in der Tabelle value_number vorhanden sein.

zusätzliche Information

Partitionsfunktion "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')

Partitionsschema "PS_files_partitioning":

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

** Ich werde ungefähr 15 Millionen Zeilen in jeder Partition haben.

Datei-Tabelle:

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)
)

value_number Tabelle:

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)
)

Indexe der Dateitabelle

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])

Indizes der Tabelle 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)

Statistiken sind aktuell. Ich habe die Datentypen für das Jahr und andere Spalten geändert und jetzt scheint die Leistung etwas besser zu sein, aber der Ausführungsplan ist immer noch der gleiche. Ich versuche, die Kardinalitätsschätzung (Indexänderung) zu korrigieren, hatte aber noch keinen Erfolg. Laut Azure-Dokumentation sollte die Datenbank eine Kompatibilitätsstufe von 130 aufweisen, und mit ProductVersion 12.0 habe ich bereits eine Kompatibilitätsstufe von 100.

Antworten:


9

Analyse

Für die Abfrage ohne Auftragspräferenz kann SQL Server gruppierte Zeilen mit einem Hash Match Flow Distinct streamen. Wenn es schnell auf die erforderliche Anzahl eindeutiger Einträge stößt, ist die Ausführungszeit kurz.

Wenn eine bestimmte Reihenfolge erforderlich ist, muss SQL Server jede Zeile testen. Um beispielsweise Zeilen in eine nameReihenfolge zu bringen, müssen alle Zeilen nach Namen sortiert werden. Dies ist langsam, wenn viele Zeilen vorhanden sind und kein Index vorhanden ist, um diese Reihenfolge ohne Sortieren bereitzustellen.

Es gibt eine Reihe grundlegender Komplikationen in Ihrem Fall, insbesondere die Aufteilung und die Disjunktion [year]. Die Partitionierung bedeutet, dass Ihre Indizes nicht die erwartete Reihenfolge liefern können. Beispielsweise wird ein Index für namezuerst nach der Partitionsnummer und dann nach dem Namen sortiert. Es können keine Zeilen geliefert werden, die nameeinzeln sortiert sind .

Sie haben auch FORCED PARAMETERIZATIONfestgelegt. Dies mag insgesamt von Vorteil sein, bringt jedoch Auswirkungen mit sich, die Sie vollständig verstehen sollten. In Kombination mit der Partitionierung und den mehrspaltigen Indizes sind Ihre Statistiken weitgehend unbrauchbar.

Die Disjunktion yearwirkt sich auch negativ auf die Bestellung aus und bedeutet, dass SQL Server nur year >= 0 and year <= 2013in Ihrem Plan suchen kann. Dies ist viel weniger selektiv als das Suchen auf year = 0und year = 2013getrennt.

Empfehlungen

In Anbetracht all dessen:

Ein guter Index für die ORDER BY nameAbfrage ist:

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;

Ein besserer Index für die value_numberTabelle ist:

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

Die Abfrage kann dann so geschrieben werden, dass für jedes Jahr und jede Partition höchstens 50 Zeilen abgerufen werden. Wir nehmen dann die ersten 50 in der Reihenfolge aus dem kombinierten Satz:

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;

Der Ausführungsplan sortiert höchstens 100 Zeilen aus der filesTabelle:

geschätzter Plan

In der Regel ist für jede Bestellung ein anderer Index erforderlich.

Wenn Sie die Gesamtzahl der Ergebnisse zählen müssen, verwenden Sie:

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 <> Geige

Durch die Nutzung unserer Website bestätigen Sie, dass Sie unsere Cookie-Richtlinie und Datenschutzrichtlinie gelesen und verstanden haben.
Licensed under cc by-sa 3.0 with attribution required.