Question

I know one is supposed to use a database sequence to obtain the value for the next key on a table (let’s assume a single-column numeric Primary Key), but in case a sequence is not used, do you see any obvious code smells with the following code (code is in Java for JPA but the pattern is really language and technology independent):

  boolean haveSucceeded = false;
  int _i = 0 ;
  while ((!haveSucceeded) && (_i++ < MAX_RETRIES)) {
     try {
          user.setId(getFacade().maxId()+1); // line A
          getFacade().create(user);          // line B
          haveSucceeded = true;
     } catch (javax.ejb.EJBTransactionRolledbackException exc) {
          debug("ConstraintValidationException, "+ ( (_i<MAX_RETRIES)?"retrying":"giving up"));
     }
  }

Where the retries are happening to account for possible clashes due to concurrent accesses (possibly from other application instances as well) and since it cannot be guaranteed that line A (where the max is calculated) and line B (where the row is inserted) will operate on the same data.

Answer 1

This doesn’t look thread safe.
You should, in a thread-safe way:

• lock the user table
• read the current max(id)
• add 1 and use that as the next id
• unlock the table

This approximates what happens in the database with an autoincrement.

If you can’t lock the table, synchronize the method, either making it static synchronized or using a static lock object, eg:

 private static final Object lock = new Object();

 void yourMethod() {
     synchronized (lock) {
         // your code here
     }
 }

If you have multiple instances of your server running, this approach won’t be enough.