如何编写和优化LINQ to SQL中具有一对多关系的三表联接？

Question

Ideal Class Structure

A Game has many Players, each of which has many Statistics. In other words, each List<Game> contains a List<Player> and each Player contains a List<Statistic> .

Game -> Player1 -> Statistic1
                   ....
                   Statistic30
        ....
        Player10 -> Statistic1
                    ....
                    Statistic30

Basic Table Schema

Game
----
GameId (int)
Region (nvarchar(4))

Player
------
GameId (int)
Region (nvarchar(4))
AccountId (int)

Statistic
---------
GameId (int)
Region (nvarchar(4))
AccountId (int)

My Attempt

var b = (from g in db.Games
         select new GameDTO()
         {
             GameId = g.GameId,
             Players = (from p in db.PlayerGames
                        where p.GameId == g.GameId && p.Region.Equals(g.Region)
                        select new PlayerGameDTO()
                        {
                            AccountId = p.AccountId,
                            GameId = p.GameId,
                            Region = p.Region,
                            Statistics = (from r in db.Statistics
                                          where r.AccountId == p.AccountId && r.GameId == p.GameId && r.Region.Equals(p.Region)
                                        select r).ToList()
                        }).ToList()
         });

This solution (obviously) does not employ Join , largely because I'm not sure how to perform the Join s in the correct order to achieve the desired result.

I should mention that each day we aggregate ~100K new games, ~1M players, and ~30M statistics. The current query can select ~1.4 games per second and uses 99% of the hyper threaded quad core CPU.

If anything is muddy, please feel free to ask for clarification.

Update #1

var d = (from g in db.Games
         join p in db.PlayerGames on new { g.GameId, g.Region } equals new { p.GameId, p.Region }
         join r in db.Statistics on new { p.GameId, p.Region, p.AccountId } equals new { r.GameId, r.Region, r.AccountId }
         select new StatisticsDTO()
         {
             GameId = r.GameId, 
             AccountId = r.AccountId, 
             StatType = r.StatType,
             Value = r.Value
         });

Something this simple is churning out ~9K (22x faster than the original) rows per second. SQL Server is clearly doing all the work, using ~90% of the CPU. HOWEVER, instead of nested objects, I'm left with a 1 dimensional query.

If you have any suggestions on this update, I'd love to hear them.

Answer 1

It sounds like it may be more appropriate to let your database handle some of this workload, especially if you're simply running queries and not writing to the database. Consider creating a View in your database that implements the joins. Then you can query the View and avoid joining on your client machine. You can still use the entity data model and LINQ to run queries against the view. You should see a pretty good performance increase with this approach.

//Possible SQL for creating the view
CREATE VIEW vw_GameData AS 
SELECT g.GameId, g.Region, p.AccountId, etc...
FROM Game g JOIN Player p ON (g.GameId = p.GameId AND g.Region = p.Region)
JOIN Statistic s ON (s.GameId = p.GameId AND s.RegionId = p.RegionId AND s.AccountId = p.AccountId)

Answer 2

First try a simple linq join.

 Game ---- GameId (int) Region (nvarchar(4)) Player ------ GameId (int) Region (nvarchar(4)) AccountId (int) Statistic --------- GameId (int) Region (nvarchar(4)) AccountId (int) 

var b = (from t in db.Games
         join t1 in t.Player on t.GameId equals t1.GameId
         join t2 in t.Statistic on t.GameId equals t2.GameId
         select new PlayerGameDTO
         {
            AccountId = t1.AccountId,
            GameId = t1.GameId,
            Region = t1.Region,
            //RawStats <-- what are you trying to do here?
            //RawStats = (from r in db.RawStats
            //where r.AccountId == p.AccountId && r.GameId == p.GameId && r.Region.Equals(p.Region) select r).ToList()
         }).ToList();