performance - 不等式が等式よりも速く結果を返すのはなぜですか?

Question

この特定のクエリに関して：

クエリされたデータを理解することを期待しないでください。それは無関係です。

WHERE句を見てください：

SELECT 
    de.DocumentEntryId,
    de.Number,
    currentdev.ToIgnore,
    previousdev.ToIgnore,
    currentdev.DocumentEntryValueId AS CurrentDocumentEntryValueId, 
    SUM(currentper.Value) AS CurrentPaymentRate,
    SUM(currentdevr.ConversionRate) AS CurrentConversionRate,
    MAX(previousdev.DocumentEntryValueId) AS PreviousDocumentEntryValueId,
    SUM(previousper.Value) AS PreviousPaymentRate,
    SUM(previousdevr.ConversionRate) AS PreviousConversionRate
FROM 
    DocumentEntry de
    INNER JOIN PaymentEntry currentpe ON currentpe.PaymentEntryId = de.CurrentPaymentEntryId
    INNER JOIN DocumentEntryValue currentdev ON currentpe.DocumentEntryValueId = currentdev.DocumentEntryValueId
    INNER JOIN DocumentEntryValue previousdev ON previousdev.DocumentEntryId = de.DocumentEntryId
    INNER JOIN PaymentEntryRate currentper ON currentpe.PaymentEntryId = currentper.PaymentEntryId
    INNER JOIN DocumentEntryValueRate currentdevr ON currentdev.DocumentEntryValueId = currentdevr.DocumentEntryValueId
    INNER JOIN PaymentEntry previouspe ON previousdev.DocumentEntryValueId = previouspe.DocumentEntryValueId
    INNER JOIN PaymentEntryRate previousper ON previouspe.PaymentEntryId = previousper.PaymentEntryId
    INNER JOIN DocumentEntryValueRate previousdevr ON previousdevr.DocumentEntryValueId = previousdev.DocumentEntryValueId
WHERE 
    previousdev.DocumentEntryValueId <> currentdev.DocumentEntryValueId
    AND currentdev.ToIgnore <> 1
    AND previousdev.ToIgnore <> 1
    AND currentpe.PaymentId = previouspe.PaymentId
GROUP BY 
    de.DocumentEntryId, 
    de.Number,
    currentdev.ToIgnore,
    previousdev.ToIgnore,
    currentdev.DocumentEntryValueId
ORDER BY DocumentEntryId

特に次の 2 つの不等式:

AND currentdev.ToIgnore <> 1
AND previousdev.ToIgnore <> 1

注: ToIgnore はビット列です。

このクエリは、約 1300 行を返すのに約 10 秒かかります。

ただし、これらの 2 行を等値演算子を使用するように変更すると、次のようになります。

AND currentdev.ToIgnore = 0
AND previousdev.ToIgnore = 0

何かを返すには永遠に時間がかかります。

役に立ちます。各テーブルのインデックスは次のとおりです。

テーブルDocumentEntryインデックス:

ALTER TABLE [dbo].[DocumentEntry] ADD  CONSTRAINT [DocumentEntry$0] PRIMARY KEY CLUSTERED 
(
    [DocumentEntryId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [IDX_DOCUMENTENTRY_REFNUMBER] ON [dbo].[DocumentEntry]
(
    [EntryReferenceNumber] 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 [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [missing_index_14227] ON [dbo].[DocumentEntry]
(
    [MasterDocumentEntryId] 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 [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [missing_index_186297_186296_DocumentEntry] ON [dbo].[DocumentEntry]
(
    [Number] ASC
)
INCLUDE (   [DocumentEntryId]) 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 [Data Filegroup 1]

テーブルDocumentEntryValueインデックス:

ALTER TABLE [dbo].[DocumentEntryValue] ADD  CONSTRAINT [DocumentEntryValue$0] PRIMARY KEY CLUSTERED 
(
    [DocumentEntryValueId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [IDX_DOCUMENTENTRYVALUE_DOCENTRYID] ON [dbo].[DocumentEntryValue]
(
    [DocumentEntryId] 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 [Data Filegroup 1]


CREATE NONCLUSTERED INDEX [IDX_DOCUMENTENTRYVALUE_REFNUMBER] ON [dbo].[DocumentEntryValue]
(
    [EntryValueReferenceNumber] 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 [Data Filegroup 1]


CREATE NONCLUSTERED INDEX [missing_index_4022_4021_DocumentEntryValue] ON [dbo].[DocumentEntryValue]
(
    [ReferenceDocumentId] 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 [Data Filegroup 1]

テーブルDocumentEntryValueRateインデックス:

ALTER TABLE [dbo].[DocumentEntryValueRate] ADD  CONSTRAINT [DocumentEntryValueRate$0] PRIMARY KEY CLUSTERED 
(
    [DocumentEntryValueRateId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [Data Filegroup 1]


CREATE NONCLUSTERED INDEX [missing_index_56865_56864_DocumentEntryValueRate] ON [dbo].[DocumentEntryValueRate]
(
    [DocumentEntryValueId] ASC
)
INCLUDE (   [timestamp],
    [DocumentEntryValueRateId],
    [RateId],
    [RateVersionId],
    [RateTypeId],
    [RateGroupId],
    [RatePeriodValueId],
    [ConversionRate]) 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 [Data Filegroup 1]

テーブルPaymentEntryインデックス:

ALTER TABLE [dbo].[PaymentEntry] ADD  CONSTRAINT [PaymentEntry$0] PRIMARY KEY CLUSTERED 
(
    [PaymentEntryId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [IDX_PAYMENTENTRY_DOCENTRYVALUEID] ON [dbo].[PaymentEntry]
(
    [DocumentEntryValueId] 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 [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [missing_index_2517_2516_PaymentEntry] ON [dbo].[PaymentEntry]
(
    [PaymentId] 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 [Data Filegroup 1]

表PaymentEntryRateインデックス:

ALTER TABLE [dbo].[PaymentEntryRate] ADD  CONSTRAINT [PaymentEntryRate$0] PRIMARY KEY CLUSTERED 
(
    [PaymentEntryRateId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [IDX_PAYMENTENTRYRATE_PAYMENTENTRYID] ON [dbo].[PaymentEntryRate]
(
    [PaymentEntryId] 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 [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [missing_index_2618_2617_PaymentEntryRate] ON [dbo].[PaymentEntryRate]
(
    [RateId] ASC
)
INCLUDE (   [timestamp],
    [PaymentEntryRateId],
    [PaymentEntryId],
    [RateTypeId],
    [RateGroupId],
    [RateVersionId],
    [RatePeriodValueId],
    [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) ON [Data Filegroup 1]

CREATE NONCLUSTERED INDEX [missing_index_2686_2685_PaymentEntryRate] ON [dbo].[PaymentEntryRate]
(
    [RateTypeId] ASC,
    [RateId] ASC,
    [RatePeriodValueId] ASC
)
INCLUDE (   [PaymentEntryId]) 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 [Data Filegroup 1]

等値演算子 (= 0) を使用した場合の推定実行計画の一部を次に示します。

この場合（またはおそらく多くの場合）、不等式演算子が等式演算子よりも高速に見える理由を誰かが説明できますか？

performance - 不等式が等式よりも速く結果を返すのはなぜですか?

1 に答える 1

Related

Reference