数据库设计：这是位置记录的好方法吗？

Question

In the diagram below you can see a simplified version of what I'm trying to do. I have to track the location of certain items, but I also have to efficiently retrieve the latest location of any given item. The simplest way of doing this, would be to query ItemLocationLog and search for the latest date for that item, but since this table is bound to be very large, I'm wondering whether this would be efficient (I guess indexing dateTime field would help, but I don't have the experience to determine just how much).

Another approach I thought about would be to add a foreign key for the log table on Item (as is shown in the diagram with the field "lastLocation"), which would always point to the latest log entry and it would thus spare me the search. Yet another option would be to add a foreign key to Location on Item and update it every time a log entry is added for any given item.

I'm sure that this is a common problem with a simple solution, but since I've had no experience with this, I'm skeptical about my own approaches. What are the best practices for this type of scenarios? Is it ok to add references to the Item table in order to avoid a costly query, or is the query trivial enough that I should just obtain this information from the log table itself?

Answer 1

As a matter of principle, only include redundancies in your model if you have measured the performance, determined the actual bottleneck and concluded the denormalization would actually help (enough to offset the risk of data corruption).

Which it won't in your case, curiously enough. One peculiarity of how B-Tree indexes work is that searching for MAX is essentially as fast as searching for exact value. You might have a little bit of a boost from better caching if INT is smaller than DATETIME on your DBMS, but not much.

Indexing is very powerful, if done right. And index on ItemLocationLog {idItem, dateTime} should facilitate lightning-fast SELECT MAX(dateTime) FROM ItemLocationLog WHERE idItem = ? .

Take a look at Use The Index, Luke! for a nice introduction on the topic.

Answer 2

Don't pre-optimize for a problem that you don't know you have.

Start with an index on the ItemLocationLog table covering idItem . Then SELECT TOP 1 idItemLocationLog from ItemLocationLog order by idItemLocationLog DESC - assuming that your PK is an autoincrement column. If this isn't fast enough, then try an index on idItem plus dateTime . If that still isn't fast enough, then you could start considering drastic denormalization, like keeping the last known location reference on Item .

Some people are really surprised how good RDBMS is at retrieving data. You shouldn't be!

Answer 3

Try this first (examples are for PostgeSQL).

-- Latest location of ItemID = 75
select
      a.ItemID
    , b.LocationID
    , ValidFrom
from Item         as a
join ItemLocation as b on b.ItemID     = a.ItemID
                      and b.ValidFrom  = (select max(x.ValidFrom) from ItemLocation as x
                                                                  where x.ItemID = a.ItemID) 
join Location     as c on b.LocationID = c.LocationID
where a.ItemID = 75 ;


-- Earliest location of ItemID = 75
select
      a.ItemID
    , b.LocationID
    , ValidFrom
from Item         as a
join ItemLocation as b on b.ItemID     = a.ItemID
                      and b.ValidFrom  = (select min(x.ValidFrom) from ItemLocation as x
                                                                  where x.ItemID = a.ItemID) 
join Location     as c on b.LocationID = c.LocationID
where a.ItemID = 75 ;

This may look scary, but is quite fast , the ItemID is part of primary keys

And if you need a list of all items at any point in time

-- Location of all items for point in time ('2012-05-01 11:00:00') 
select
      a.ItemID
    , b.LocationID
    , ValidFrom
from Item         as a
join ItemLocation as b on b.ItemID     = a.ItemID
                      and b.ValidFrom  = (select max(x.ValidFrom)
                                            from ItemLocation as x
                                           where x.ItemID = a.ItemID
                                             and x.ValidFrom <= '2012-05-01 11:00:00') 
join Location     as c on c.LocationID = b.LocationID
;