Display rows when scrolling as twitter, using a stored procedure - sql-server

Question

I have a site that displays posts. I want the site's scrolling to behave like twitter - scrolling down will display more and more posts, endlessly.
Suppose I have the following tables:
A Post table to hold all the posts. Every post is related to a single person
CREATE TABLE [dbo].[Post](
[Id] [bigint] IDENTITY(1,1) NOT NULL,
[PersonId] [int] NOT NULL,
[PublishDate] [datetime] NOT NULL,
CONSTRAINT [PK_Post] PRIMARY KEY CLUSTERED
(
[Id] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF,
ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY] TEXTIMAGE_ON [PRIMARY]
A PostTag table to hold all the related tags of each post.
CREATE TABLE [dbo].[PostTag](
[PostId] [bigint] NOT NULL,
[TagId] [int] NOT NULL,
CONSTRAINT [PK_PostTag] PRIMARY KEY CLUSTERED
(
[PostId] ASC,
[TagId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF,
ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY]
For each user of the site, the UserPersonStatistics table holds the number of times he showed interest in a person related post.
CREATE TABLE [dbo].[UserPersonStatistics](
[UserId] [bigint] NOT NULL,
[PersonId] [int] NOT NULL,
[Counter] [bigint] NOT NULL,
CONSTRAINT [PK_UserPersonStatistics] PRIMARY KEY CLUSTERED
(
[UserId] ASC,
[PersonId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF,
ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY]
For each user of the site, the UserPostStatistics table holds the number of times he showed interest in a post.
CREATE TABLE [dbo].[UserPostStatistics](
[UserId] [bigint] NOT NULL,
[PostId] [bigint] NOT NULL,
[Counter] [bigint] NOT NULL,
CONSTRAINT [PK_UserPostStatistics] PRIMARY KEY CLUSTERED
(
[UserId] ASC,
[PostId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF,
ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY]
For each user of the site, the UserTagStatistic table holds the number of times he showed interest in a tag related post.
CREATE TABLE [dbo].[UserTagStatistics](
[UserId] [bigint] NOT NULL,
[TagId] [int] NOT NULL,
[Counter] [bigint] NOT NULL,
CONSTRAINT [PK_UserTagStatistics] PRIMARY KEY CLUSTERED
(
[UserId] ASC,
[TagId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF,
ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY]
What I need is a stored procedure that for each user returns 35 different posts each time, that "remembers" the last 35 post so it will not return the same posts again, and the 35 posts should be consist of:
15 posts for the most popular tag (UserTagStatistics)
15 posts for the most popular person (UserPersonStatistics)
5 most popular posts (UserPostStatistics)
One problem is that the procedure should return 35 different posts each time.
One more problem is that a post can return once as the most popular post, once as a post of the most popular tag, and once as a post for the most popular person. This post should be counted once, not three times.
The performance of the stored procedure is crucial.
I know its a very complicated question.
Any thoughts are appreciated.
kruvi

Answer 1

Add a "LastViewed" datetime field to all tables then use a proc like this. For performance, just make sure to have an index on UserID+LastViewed+Counter and UserID+PersonID for each of the three tables and it should scream. Actually, since UserID+LastViewed+Counter is practically the whole table, if possible I'd recommend you make it the clustered index on each of your tables so that you avoid creating that second index which would basically be the same size as the raw table.
create proc GetInfo(#UserId bigint) as
begin
update userpersonstatistics
set
lastviewed=getdate()
where
userid=#UserID and personid in
(
select top 15 personid from userpersonstatistics
where
userid=#UserID and
(
lastviewed is null or lastviewed !=
(select max(lastviewed) from userpersonstatistics
where userid=#UserID)
)
order by counter desc
)
select * from UserPersonStatistics
where UserID=#UserID and LastViewed =
(select max(lastviewed) from UserTagStatistics)
--**Repeat the above code for UserPostStatistics and UserTagStatistics
end
REVISED PROC BASED ON INPUT:
create proc GetInfo(#UserId bigint) as
begin
declare #lastviewed datetime
declare #results TABLE
(
StatType varchar(10),
Counter int,
PostID
)
set #lastviewed = getdate()
--Person
insert into #results(StatType,Counter,PostID)
select
'Person',counter,PostID
from
UserPersonStatistics
where
userid=#UserID and personid in
(
select top 35 personid from userpersonstatistics
where
userid=#UserID and
(
lastviewed is null or lastviewed !=
(select max(lastviewed) from userpersonstatistics
where userid=#UserID)
)
order by counter desc
)
--Post
insert into #results(StatType,Counter,PostID)
select
'Post',counter,PostID
from
UserPostStatistics
where
userid=#UserID and Postid in
(
select top 35 Postid from userPoststatistics
where
userid=#UserID and
(
lastviewed is null or lastviewed !=
(select max(lastviewed) from userPoststatistics
where userid=#UserID)
)
order by counter desc
)
--Tag
insert into #results(StatType,Counter,TagID)
select
'Tag',counter,TagID
from
UserTagStatistics
where
userid=#UserID and Tagid in
(
select top 35 Tagid from userTagstatistics
where
userid=#UserID and
(
lastviewed is null or lastviewed !=
(select max(lastviewed) from userTagstatistics
where userid=#UserID)
)
order by counter desc
)
--At this point you could have 105 rows of the various types (35*3).
--You can use whatever algorithm you need to decide the top 35.
--That may include some weighting.
--You may want to consider using the Rank() function.
end
If your algorithm should consider the #1 top counter from each category before the #2's, take a look at the Rank() function.