Database design: Matching sql database keys to php constants?

Well this is a simple design question I've wondered about many times and never found a satisfying solution for. My example is with php-sql, but this certainly applies to other languages too.

I have a small database table containing only very few entries, and that almost never needs updating. eg this usertype table:

usertype_id (primary key)  | name       | description
---------------------------+------------+-------------------
1                          | 'admin'    | 'Administrator'
2                          | 'reguser'  | 'Registered user'
3                          | 'guest'    | 'Guest'

Now in the php code, I often have to check or compare the type of user I'm dealing with. Since the user types are stored in the database, I can either:

1) Select * from the usertype table at class instantiation, and store it in an array.

Then all the ids are available to the code, and I can do a simple select to get the rows I need. This solution requires an array and a db query every time the class is instantiated.

$query = "SELECT info, foo FROM user WHERE usertype_id = ".$usertypes['admin'];

2) Use the name column to select the correct usertype_id, so we can effectively join with other tables. This is more or less equivalent to 1) but without needing to cache the whole usertype table in the php object:

$query = "SELECT info, foo FROM user JOIN user开发者_Python百科type USING (usertype_id) WHERE usertype.name = 'admin' ";

3) Define constants that match the keys in the usertype table:

// As defines
define("USERTYPE_ADMIN",1);
define("USERTYPE_REGUSER",2);

//Or as class constants
const USERTYPE_ADMIN = 1;
const USERTYPE_REGUSER = 2;

And then do a simple select.

$query = "SELECT info, foo FROM user WHERE usertype_id = " . USERTYPE_ADMIN;

This is probably the most resource-efficient solution, but it is bad to maintain, as you have to update both the table and the code if you need to modify something in the usertype table..

4) Scrap the usertype table and only keep the types in the php code. I don't really like this because it lets any value get into the database and get assigned to the type of user. But maybe, all things considered, it isn't so bad and i'm just complicating something that should be simple..

Anyways, to sum it up the solution I like most is #2 because it's coherent and with an index on usertype.name, it can't be that bad. But what I've often ended up using is #3, for efficiency.

How would you do it? Any better solutions?

(edit: fixed query in #2)

I would suggest #3 to avoid useless queries, and prevent risk of behavior changes if existing DB table rows are incidentally modified:

• Adding the necessary constants in the model class:

  class Role // + use namespaces if possible
  {
    // A good ORM could be able to generate it (see @wimvds answer)
    const ADMIN = 1;
    const USER = 2;
    const GUEST = 3;
  
    //...
  

  }

• Then querying like this makes sense:

  $query = "SELECT info, foo FROM user WHERE role_id = ".Role::ADMIN;
  

  With an ORM (e.g. Propel in the example below) you'll end up doing:

  $isAdminResults = UserQuery::create()->filterByRoleId(Role::ADMIN);
  

I almost always go for option 3). You could generate the code needed automatically based on what is available in the DB. The only thing you have to remember then is that you have to run the script to update/rewrite that info when you add another role (but if you're using phing or a similar build tool to deploy your apps, just add a build rule for it to your deploy script and it will always be run whenever you deploy your code :p).

Why not denormalize the DB table so instead of having usertype_id, you'd have usertype with the string type (admin). Then in PHP you can just do define('USERTYPE_ADMIN', 'admin');. It saves you from having to modify two places if you want to add a user type...

And if you're really worried about any value getting in, you could always make the column an ENUM data type, so it would self manage...

For tables that will contain "type" values especially when is expected such table to change over time I tend to use simple approach: Add Varchar column named hid (comes from "human readable id") with unique key. Then I fill it with id meaningful to humans like:

usertype_id (primary key)  | name       | description       | hid (unique key)
---------------------------+------------+-------------------+---------------
1                          | 'admin'    | 'Administrator'   | 'admin'
2                          | 'reguser'  | 'Registered user' | 'user'
3                          | 'guest'    | 'Guest'           | 'guest'

When you need the actual id you will have to do select based on hid column, i.e.

select usertype_id from tablename where hid = "admin"

This is not an efficient approach but it will ensure compatibility of your application among different deployments (i.e. one client may have 1.admin, 2. guest; other client 1.admin, 2. user, etc.). For your case I think #3 is pretty suitable but if you expect to have more than 10 different user roles - try the "hid" approach.

Are you using any kind of framework here? Could these values be stored in a single source - a config file - which both creates a list of the objects in PHP and also populates the table when you bootstrap the database? I'm thinking from a Rails perspective, as it's been a while since I've written any PHP. Solution there would probably be fixtures.

Why not to make it just

foreach (getdbarr("SELECT * FROM usertype") as $row)  {
  define($row['name'],$row['id']);
}

You shouldn't need a JOIN in every query to fetch the information about types/roles. You can keep your 'user' model and 'role' models separate in the data access objects (DAO) -- especially since there are so few records for user types.

In most cases where I have a limited number of options that I'd otherwise be joining against a large table, I cache them in memcached as an associative array. In the event I need some information about a particular relationship (like a role) I just lazy load it.

$user = DAO_User::get(1); // this pulls a JOIN-less record
$role = $user->getRole(); // lazy-load

The code for $user->getRole() can be something like:

public function getRole() { 
  // This comes from a cache that may be called multiple 
  // times per request with no penalty (i.e. store in a registry)
  $roles = DAO_UserRoles::getAll();

  if(isset($roles[$this->role_id]))
    return $roles[$this->role_id];

  return null; // or: new Model_UserRole();
}

This also works if you want to display a list with 1000 users on it. You can simply render values for that column from a single $roles associative array.

This is a major performance improvement on the SQL end, and it goes a long way to reducing complexity in your code base. If you have several other foreign keys on the user table you can still use this approach to grab the necessary information when you need it. It also means you can have dependable Model_* classes without having to create hybrids for every possible combination of tables you might JOIN -- which is much better than simply getting a result set, iterating it, and freeing it.

Even with more than 100 rows on both sides of your JOIN, you can still use the lazy load approach for infrequent or highly redundant information. With a reasonable caching service in your code, there's no penalty for calling DAO_UserRole::get(1500) multiple times because subsequent calls during the same request shouldn't hit the database twice. In most cases you're only going to be displaying 10-25 rows per page out of 1000s, and lazy loading will save your database engine from having to JOIN all the extraneous rows before you actually need them.

The main reason to do a JOIN is if your WHERE logic requires it, or if you need to ORDER BY data from a foreign key. Treating JOINs as prohibitively expensive is a good habit to be in.

For basicly static lookup tables, I generally make static constant files (such as your #3). I generally use classes such as:

namespace Constants;
class UserTypes {
    const ADMIN = 1;
    const USER = 2;
    const GUEST = 3;
}

$id = Constants\UserTypes::ADMIN;

When I'm using lookup takes that are a bit more variable, then I'll pull it into a object and then cache it for 24 hours. That way it only gets updated once a day. That will save you from making database round trips, but allow you to deal with things in code easily.

Yeah, you're right about avoiding #3 and sticking with #2. As much as possible, look-ups like when you use a usertype table to contain the roles and then relate them to the user table using the id values should stay in the database. If you use constants, then the data must always rely on your php code to be interpreted. Also, you can enforce data integrity by using foreign keys (where servers allow) and it will allow you to port the reporting from your php code to other reporting tools. Maintenance also becomes easier. Database administrators won't need to know php in order to derive the meanings of the numbers if you used #3, should they ever be asked to aid in reports development. It may not seem too relevant, but in terms of maintenance, using stored procedures than embedded sql in your php code would also be maintenance-friendly in several ways, and will also be advantageous to DBAs.

I'd go for option #2 and use the join as it is intended to be used. You never know what the future will throw up, it's always better to be prepared today!

With regards to leaving the database alone as much as possible for such operations, there is also the possibility of caching in the long term. For this route, within PHP an option is to use a file cache, one that will only get updated when time calls for it. For the framework I have created, here's an example; I'd be interested to know what people think:

Note:

(LStore, LFetch, GetFileName) belong to a Cache object which gets called statically.

(Blobify and Unblobify) belong to a SystemComponent object which is always alive

Each piece of cache data has a key. this is the only thing you ever have to remember

public function LStore($key,$data, $blnBlobify=true) {
    /* Opening the file in read/write mode */
    $h = fopen(self::GetFileName($key, 'longstore'),'a+');
    if (!$h) throw new Exception('Could not write to cache');
    flock($h,LOCK_EX); // exclusive lock, will get released when the file is closed
    fseek($h,0); // go to the start of the file
    /* truncate the file */
    ftruncate($h,0);
    if($blnBlobify==true) { $data = SystemComponent::Blobify(array($data)); }
    If (fwrite($h,$data)===false) {
        throw new Exception('Could not write to cache');
    }
    fclose($h);
}   
public function LFetch($key) {
    $filename = self::GetFileName($key, 'longstore');
    if (!file_exists($filename)){ return false;}
    $h = fopen($filename,'r');
    if (!$h){ return false;}
    /* Getting a shared lock */
    flock($h,LOCK_SH);
    $data = file_get_contents($filename);
    fclose($h);
    $data = SystemComponent::Unblobify($data);
    if (!$data) {
        /* If unserializing somehow didn't work out, we'll delete the file */
        unlink($filename);
        return false;
    }
    return $data;
}
/* This function is necessary as the framework scales different directories */

private function GetFileName($key, $strCacheDirectory='') {
    if(!empty($strCacheDirectory)){ 
        return SystemComponent::GetCacheAdd() . $strCacheDirectory.'/' .  md5($key); 
    } else {    
        return SystemComponent::GetCacheAdd() . md5($key);
    }
}
public function Blobify($Source){
    if(is_array($Source)) { $Source = serialize($Source); }

    $strSerialized = base64_encode($Source);

    return $strSerialized;

}
public function Unblobify($strSerialized){
    $Decoded = base64_decode($strSerialized);
    if(self::CheckSerialized($Decoded)) { $Decoded = unserialize($Decoded); }

    return $Decoded;    

}
function CheckSerialized($Source){
    $Data = @unserialize($Source);
    if ($Source === 'b:0;' || $Data !== false) {
        return true;
    } else {
        return false;
    }
}  

Now when it comes to accessing the actual data, I just call a fetch. For making sure it is up to date, I tell it to store. In your case, this would be after updating the usertype table.