Estou escrevendo uma consulta que será usada para os resultados da página de um feed social. O conceito é que o aplicativo móvel solicitará N itens e fornecerá uma data e hora de início que chamei @CutoffTime
abaixo. O objetivo do tempo limite é estabelecer quando a janela de paginação deve começar. A razão pela qual estamos usando um registro de data e hora em vez de um deslocamento de linha é que o registro de data e hora nos permitirá paginar de um local consistente ao obter postagens mais antigas, mesmo se um conteúdo social mais recente for adicionado.
Como os itens do feed social podem ser seus ou de seus amigos, estou usando um UNION
para combinar os resultados desses dois grupos. Originalmente, tentei a TheQuery_CTE
lógica sem o UNION
e foi lento demais.
Isto é o que eu fiz (incluindo esquema de tabela pertinente):
CREATE TABLE [Content].[Photo]
(
[PhotoId] INT NOT NULL PRIMARY KEY IDENTITY (1, 1),
[Key] UNIQUEIDENTIFIER NOT NULL DEFAULT NEWID(),
[FullResolutionUrl] NVARCHAR(255) NOT NULL,
[Description] NVARCHAR(255) NULL,
[Created] DATETIME2(2) NOT NULL DEFAULT SYSUTCDATETIME(),
);
CREATE TABLE [Content].[UserPhotoAssociation]
(
[PhotoId] INT NOT NULL,
[UserId] INT NOT NULL,
[ShowInSocialFeed] BIT NOT NULL DEFAULT 0,
CONSTRAINT [PK_UserPhotos] PRIMARY KEY ([PhotoId], [UserId]),
CONSTRAINT [FK_UserPhotos_User] FOREIGN KEY ([UserId])
REFERENCES [User].[User]([UserId]),
CONSTRAINT [FK_UserPhotos_Photo] FOREIGN KEY ([PhotoId])
REFERENCES [Content].[Photo]([PhotoId])
);
CREATE TABLE [Content].[FlaggedPhoto]
(
[FlaggedPhotoId] INT NOT NULL PRIMARY KEY IDENTITY(1,1),
[PhotoId] INT NOT NULL,
[FlaggedBy] INT NOT NULL,
[FlaggedOn] DATETIME2(0) NOT NULL DEFAULT SYSDATETIME(),
[FlaggedStatus] INT NOT NULL DEFAULT 1,
[ReviewedBy] INT NULL,
[ReviewedAt] DATETIME2(0) NULL
CONSTRAINT [FK_Photos_PhotoId_to_FlaggedPhotos_PhotoId] FOREIGN KEY ([PhotoId])
REFERENCES [Content].[Photo]([PhotoId]),
CONSTRAINT [FK_FlaggedPhotoStatus_FlaggedPhotoStatusId_to_FlaggedPhotos_FlaggedStatus] FOREIGN KEY ([FlaggedStatus])
REFERENCES [Content].[FlaggedContentStatus]([FlaggedContentStatusId]),
CONSTRAINT [FK_User_UserId_to_FlaggedPhotos_FlaggedBy] FOREIGN KEY ([FlaggedBy])
REFERENCES [User].[User]([UserId]),
CONSTRAINT [FK_User_UserId_to_FlaggedPhotos_ReviewedBy] FOREIGN KEY ([ReviewedBy])
REFERENCES [User].[User]([UserId])
);
CREATE TABLE [User].[CurrentConnections]
(
[MonitoringId] INT NOT NULL PRIMARY KEY IDENTITY,
[Monitor] INT NOT NULL,
[Monitored] INT NOT NULL,
[ShowInSocialFeed] BIT NOT NULL DEFAULT 1,
CONSTRAINT [FK_Monitoring_Monitor_to_User_UserId] FOREIGN KEY ([Monitor])
REFERENCES [dbo].[User]([UserId]),
CONSTRAINT [FK_Monitoring_Monitored_to_User_UserId] FOREIGN KEY ([Monitored])
REFERENCES [dbo].[User]([UserId])
);
CREATE TABLE [Content].[PhotoLike]
(
[PhotoLikeId] INT NOT NULL PRIMARY KEY IDENTITY,
[PhotoId] INT NOT NULL,
[UserId] INT NOT NULL,
[Created] DATETIME2(2) NOT NULL DEFAULT SYSUTCDATETIME(),
[Archived] DATETIME2(2) NULL,
CONSTRAINT [FK_PhotoLike_PhotoId_to_Photo_PhotoId] FOREIGN KEY ([PhotoId])
REFERENCES [Content].[Photo]([PhotoId]),
CONSTRAINT [FK_PhotoLike_UserId_to_User_UserId] FOREIGN KEY ([UserId])
REFERENCES [User].[User]([UserId])
);
CREATE TABLE [Content].[Comment]
(
[CommentId] INT NOT NULL PRIMARY KEY IDENTITY,
[PhotoId] INT NOT NULL,
[UserId] INT NOT NULL,
[Comment] NVARCHAR(255) NOT NULL,
[Created] DATETIME2(2) NOT NULL DEFAULT SYSUTCDATETIME(),
[CommentOrder] DATETIME2(2) NOT NULL DEFAULT SYSUTCDATETIME(),
[Archived] DATETIME2(2) NULL,
CONSTRAINT [FK_Comment_PhotoId_to_Photo_PhotoId] FOREIGN KEY ([PhotoId])
REFERENCES [Content].[Photo]([PhotoId]),
CONSTRAINT [FK_Comment_UserId_to_User_UserId] FOREIGN KEY ([UserId])
REFERENCES [User].[User]([UserId])
);
/*
End table schema
*/
DECLARE @UserId INT,
@NumberOfItems INT,
@CutoffTime DATETIME2(2) = NULL -- Stored Proc input params
-- Make the joins and grab the social data we need once since they are used in subsequent queries that aren't shown
DECLARE @SocialFeed TABLE ([Key] UNIQUEIDENTIFIER, [PhotoId] INT
, [Description] NVARCHAR(255), [FullResolutionUrl] NVARCHAR(255)
, [Created] DATETIME2(2), [CreatorId] INT, [LikeCount] INT
, [CommentCount] INT, [UserLiked] BIT);
-- Offset might be different for each group
DECLARE @OffsetMine INT = 0, @OffsetTheirs INT = 0;
IF @CutoffTime IS NOT NULL
BEGIN
-- Get the offsets
;WITH [GetCounts_CTE] AS
(
SELECT
[P].[PhotoId] -- INT
, 1 AS [MyPhotos]
FROM [Content].[Photo] [P]
INNER JOIN [Content].[UserPhotoAssociation] [UPA] ON
[UPA].[PhotoId] = [P].[PhotoId]
AND
[UPA].[ShowInSocialFeed] = 1
LEFT JOIN [Content].[FlaggedPhoto] [FP] ON
[FP].[PhotoId] = [P].[PhotoId]
AND
[FP].[FlaggedStatus] = 3 -- Flagged photos that are confirmed apply to everyone
WHERE
[FP].[FlaggedPhotoId] IS NULL -- Filter out flagged photos
AND
[UPA].[UserId] = @UserId -- Show the requesting user
AND
[P].[Created] >= @CutoffTime -- Get the newer items
UNION
SELECT
[P].[PhotoId] -- INT
, 0 AS [MyPhotos]
FROM [Content].[Photo] [P]
INNER JOIN [Content].[UserPhotoAssociation] [UPA] ON
[UPA].[PhotoId] = [P].[PhotoId]
AND
[UPA].[ShowInSocialFeed] = 1
INNER JOIN [User].[CurrentConnections] [M] ON
[M].[Monitored] = [UPA].[UserId]
AND
[M].[Monitor] = @UserId AND [M].[ShowInSocialFeed] = 1 -- this join isn't present above
LEFT JOIN [Content].[FlaggedPhoto] [FP] ON
[FP].[PhotoId] = [P].[PhotoId]
AND
(
[FP].[FlaggedStatus] = 3
OR
([FP].[FlaggedBy] = @UserId AND [FP].[FlaggedStatus] = 1)
) -- Flagged photos that are confirmed apply to everyone, pending flags apply to the user
WHERE
[FP].[FlaggedPhotoId] IS NULL -- Filter out flagged photos
AND
[P].[Created] >= @CutoffTime -- Get the newer items
)
SELECT
@OffsetMine = SUM(CASE WHEN [MyPhotos] = 1 THEN 1 ELSE 0 END)
, @OffsetTheirs = SUM(CASE WHEN [MyPhotos] = 0 THEN 1 ELSE 0 END)
FROM [GetCounts_CTE]
END
-- Prevent absence of social data from throwing an error below.
SET @OffsetMine = ISNULL(@OffsetMine, 0);
SET @OffsetTheirs = ISNULL(@OffsetTheirs, 0);
-- Actually select the data I want
;WITH TheQuery_CTE AS
(
SELECT
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
, COUNT(DISTINCT [PL].[PhotoLikeId]) AS [LikeCount] -- Count distinct used due to common join key
, COUNT(DISTINCT [C].[CommentId]) AS [CommentCount]
, CAST(ISNULL(MAX(CASE WHEN [PL].[UserId] = @UserId THEN 1 END), 0) AS BIT) AS [UserLiked]
FROM [Content].[Photo] [P]
INNER JOIN [Content].[UserPhotoAssociation] [UPA] ON
[UPA].[PhotoId] = [P].[PhotoId]
AND
[UPA].[ShowInSocialFeed] = 1
LEFT JOIN [Content].[PhotoLike] [PL] ON
[PL].[PhotoId] = [P].[PhotoId]
AND
[PL].[Archived] IS NULL
LEFT JOIN [Content].[Comment] [C] ON
[C].[PhotoId] = [P].[PhotoId]
AND
[C].[Archived] IS NULL
LEFT JOIN [Content].[FlaggedPhoto] [FP] ON
[FP].[PhotoId] = [P].[PhotoId]
AND
[FP].[FlaggedStatus] = 3 -- Flagged photos that are confirmed apply to everyone
WHERE
[FP].[FlaggedPhotoId] IS NULL -- Filter out flagged photos
AND
[UPA].[UserId] = @UserId -- Show the requesting user
GROUP BY
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
ORDER BY
[P].[Created] DESC
, [P].[Key] -- Ensure consistent order in case of duplicate timestamps
OFFSET @OffsetMine ROWS FETCH NEXT @NumberOfItems ROWS ONLY
UNION
SELECT
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
, COUNT(DISTINCT [PL].[PhotoLikeId]) AS [LikeCount]
, COUNT(DISTINCT [C].[CommentId]) AS [CommentCount]
, CAST(ISNULL(MAX(CASE WHEN [PL].[UserId] = @UserId THEN 1 END), 0) AS BIT) AS [UserLiked]
FROM [Content].[Photo] [P]
INNER JOIN [Content].[UserPhotoAssociation] [UPA] ON
[UPA].[PhotoId] = [P].[PhotoId]
AND
[UPA].[ShowInSocialFeed] = 1
INNER JOIN [User].[CurrentConnections] [M] ON
[M].[Monitored] = [UPA].[UserId]
AND
[M].[Monitor] = @UserId AND [M].[ShowInSocialFeed] = 1
LEFT JOIN [Content].[PhotoLike] [PL] ON
[PL].[PhotoId] = [P].[PhotoId]
AND
[PL].[Archived] IS NULL
LEFT JOIN [Content].[Comment] [C] ON
[C].[PhotoId] = [P].[PhotoId]
AND
[C].[Archived] IS NULL
LEFT JOIN [Content].[FlaggedPhoto] [FP] ON
[FP].[PhotoId] = [P].[PhotoId]
AND
(
[FP].[FlaggedStatus] = 3
OR
([FP].[FlaggedBy] = @UserId AND [FP].[FlaggedStatus] = 1)
) -- Flagged photos that are confirmed apply to everyone, pending flags apply to the user
WHERE
[FP].[FlaggedPhotoId] IS NULL -- Filter out flagged photos
GROUP BY
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
ORDER BY
[P].[Created] DESC
, [P].[Key] -- Ensure consistant order in case of duplicate timestamps
OFFSET @OffsetTheirs ROWS FETCH NEXT @NumberOfItems ROWS ONLY
)
INSERT INTO @SocialFeed ([Key], [PhotoId], [Description], [FullResolutionUrl]
, [Created], [CreatorId], [LikeCount], [CommentCount], [UserLiked])
SELECT TOP (@NumberOfItems)
[Key]
, [PhotoId]
, [Description]
, [FullResolutionUrl]
, [Created]
, [UserId]
, [LikeCount]
, [CommentCount]
, [UserLiked]
FROM [TheQuery_CTE]
ORDER BY -- Order here so the top works properly
[Created] DESC
, [Key] -- Ensure consistent order in case of duplicate timestamps
-- Output the social feed
SELECT
[P].[Key]
, [P].[PhotoId]
, [P].[Description] AS [PhotoDescription]
, [P].[FullResolutionUrl]
, [P].[Created] AS [Posted]
, [P].[CreatorId]
, [LikeCount]
, [CommentCount]
, [UserLiked]
FROM @Photos [P]
-- Select other data needed to build the object tree in the application layer
Percebo que posso me livrar do UNION
in the, GetCounts_CTE
mas não acho que isso realmente resolverá nenhum dos problemas que vejo abaixo.
Vejo alguns possíveis problemas:
- Isso é muita lógica duplicada, então provavelmente estou tornando a vida mais difícil para mim.
- Se ocorrer uma inserção entre o cálculo da contagem e a seleção dos dados, estarei desligado. Eu não acho que isso aconteceria com frequência, mas levaria a erros estranhos/difíceis de depurar.
- Todos os problemas que as pessoas mais inteligentes/mais experientes encontrariam com a configuração acima.
Qual é a melhor maneira de escrever esta consulta? Pontos de bônus a solução torna minha vida mais simples.
Editar:
Não quero selecionar todos os dados e deixar o cliente exibir preguiçosamente os itens, porque não quero abusar dos planos de dados das pessoas, forçando-as a baixar itens que nunca verão. É certo que os dados provavelmente não serão tão grandes no grande esquema das coisas, mas pilhas de centavos ....
Editar 2:
Suspeito fortemente que esta não seja a solução ideal, mas é a melhor que encontrei até agora.
Mover minha UNION
consulta para algo VIEW
semelhante ao que Greg sugeriu funcionou bem para ocultar essa lógica e fornecer uma consulta mais concisa em meu procedimento armazenado. A view também abstrai a feiúra/complicação da união, o que é legal porque estou usando duas vezes no meu select. Segue o código da view:
CREATE VIEW [Social].[EverFeed]
AS
SELECT
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
, COUNT(DISTINCT [PL].[PhotoLikeId]) AS [LikeCount] -- Distinct due to common join key
, COUNT(DISTINCT [C].[CommentId]) AS [CommentCount]
, CAST(ISNULL(
MAX(CASE WHEN [PL].[UserId] = [UPA].[UserId] THEN 1 END), 0) AS BIT) AS [UserLiked]
, NULL AS [Monitor]
FROM [Content].[Photo] [P]
INNER JOIN [Content].[UserPhotoAssociation] [UPA] ON
[UPA].[PhotoId] = [P].[PhotoId]
AND
[UPA].[ShowInSocialFeed] = 1
LEFT JOIN [Content].[PhotoLike] [PL] ON
[PL].[PhotoId] = [P].[PhotoId]
AND
[PL].[Archived] IS NULL
LEFT JOIN [Content].[Comment] [C] ON
[C].[PhotoId] = [P].[PhotoId]
AND
[C].[Archived] IS NULL
LEFT JOIN [Content].[FlaggedPhoto] [FP] ON
[FP].[PhotoId] = [P].[PhotoId]
AND
[FP].[FlaggedStatus] = 3 -- Flagged photos that are confirmed apply to everyone
WHERE
[FP].[FlaggedPhotoId] IS NULL -- Filter out flagged photos
GROUP BY
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
UNION
SELECT
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
, COUNT(DISTINCT [PL].[PhotoLikeId]) AS [LikeCount]
, COUNT(DISTINCT [C].[CommentId]) AS [CommentCount]
, CAST(ISNULL(
MAX(CASE WHEN [PL].[UserId] = [M].[Monitor] THEN 1 END), 0) AS BIT) AS [UserLiked]
, [M].[Monitor]
FROM [Content].[Photo] [P]
INNER JOIN [Content].[UserPhotoAssociation] [UPA] ON
[UPA].[PhotoId] = [P].[PhotoId]
AND
[UPA].[ShowInSocialFeed] = 1
INNER JOIN [User].[CurrentConnections] [M] ON
[M].[Monitored] = [UPA].[UserId]
AND
[M].[ShowInSocialFeed] = 1
LEFT JOIN [Content].[PhotoLike] [PL] ON
[PL].[PhotoId] = [P].[PhotoId]
AND
[PL].[Archived] IS NULL
LEFT JOIN [Content].[Comment] [C] ON
[C].[PhotoId] = [P].[PhotoId]
AND
[C].[Archived] IS NULL
LEFT JOIN [Content].[FlaggedPhoto] [FP] ON
[FP].[PhotoId] = [P].[PhotoId]
AND
(
[FP].[FlaggedStatus] = 3
OR
([FP].[FlaggedBy] = [M].[Monitor] AND [FP].[FlaggedStatus] = 1)
) -- Flagged photos that are confirmed (3) apply to everyone
-- , pending flags (1) apply to the user
WHERE
[FP].[FlaggedPhotoId] IS NULL -- Filter out flagged photos
GROUP BY
[P].[Key]
, [P].[PhotoId]
, [P].[Description]
, [P].[FullResolutionUrl]
, [P].[Created]
, [UPA].[UserId]
, [M].[Monitor]
Using that view I shortened my query to the following. Note I'm setting the OFFSET
with a subquery.
DECLARE @UserId INT, @NumberOfItems INT, @CutoffTime DATETIME2(2);
SELECT
[Key]
, [PhotoId]
, [Description]
, [FullResolutionUrl]
, [Created]
, [UserId]
, [LikeCount]
, [CommentCount]
, [UserLiked]
FROM [Social].[EverFeed] [EF]
WHERE
(
([EF].[UserId] = @UserId AND [EF].[Monitor] IS NULL)
OR
[EF].[Monitor] = @UserId
)
ORDER BY -- Order here so the top works properly
[Created] DESC
, [Key] -- Ensure consistant order in case of duplicate timestamps
OFFSET CASE WHEN @CutoffTime IS NULL THEN 0 ELSE
(
SELECT
COUNT([PhotoId])
FROM [Social].[EverFeed] [EF]
WHERE
(
([EF].[UserId] = @UserId AND [EF].[Monitor] IS NULL)
OR
[EF].[Monitor] = @UserId
)
AND
[EF].[Created] >= @CutoffTime -- Get the newer items
) END
ROWS FETCH NEXT @NumberOfItems ROWS ONLY
The view nicely separates the complexity of the UNION
from the filtering. I think the subquery in the OFFSET
clause will prevent concurrency issues that I was worried about by making the whole query atomic.
One problem I just found while typing this is: in the code above if two photos with the same creation date are on different "pages" then the photos on the subsequent pages will be filtered out. Consider the following data:
PhotoId | Created | ...
------------------------
1 | 2015-08-26 01:00.00
2 | 2015-08-26 01:00.00
3 | 2015-08-26 01:00.00
With a page size of 1 on the initial page, PhotoId 1
will be returned. With the same page size on the second page no results will be returned. I think in order to resolve this I'm going to have to add the Key
Guid as a parameter....
I suggest you go down a different path with this. It looks like a pattern of
select top() .... order by
should be sufficient.To handle cases where Created happens to be equal you need to include the
Key
value in the cutoff handling. Perhaps a where clause something like this.First, factorize the select statement in a view, the simplest illustration hereafter :
Edit : And if you want to include the UNION and filter on userId, instead include them in an inline table user defined function with a @UserId parameter.
Second, consider (cautiously) using something like SET TRANSACTION ISOLATION LEVEL SNAPSHOT on your session in order to avoid discrepancies between the COUNT and the SELECT statements. However you could use the highest PhotoId at a given moment to prevent selecting more recent photos.
Third, you could use a lazy loading mechanism on your front end, loading posts progressively as the users scroll down the page, and avoid the (somehow outdated) paging system.