使用Web服务时避免多次呼叫

Question

I have a task where a user consumes XML from a third party. The XML feed is only updated once a day. The XML is stored in a database and returned to the user when requested. If the XML is not in the database, then it is retrieved from the third party, stored in the database and returned to the user. All subsequent requests will simply read the XML from the database.

Now my question. Say it takes 10 seconds for the request to the third party to return. In this period, there are multiple server calls for the same data. I don't want each of these to fire off requests to the third party and I don't want the user to receive nothing or an error. They should probably wait for the first request to complete at which point the XML would be available. This is a relatively simple problem but I want to know what the best way of catering for it is.

Do I just use a simple flag to control requests or maybe something like a semaphore? Are there better solutions based on the stack I intend to use which is the Play framework and a cassandra backend. Is there something I could do with callbacks or triggers?

By the way, I need to lazy load the data when the first request comes in. So, in this task it isn't an option to get the data in a separate process or when the app starts...

Thanks

Answer 1

All you need to do is create a separate component that is responsible to get the XML from the third party and save it to the database.
In your code the various thread try to "fetch" the XML from this component.
This component returns the XML from the database if it exists. If it does not exist then you use a ReentrantLock to synchronize.
So you do a trylock and only one of your threads succeeds. The rest will be blocked. When the lock is released the other threads are unblocked but the XML has already been fetched from the third party and stored to the database by the thread that first managed to gain the lock. So the other threads just return the XML from the DB.

Example code (this is just a "pseudo code" to get you started. You should handle exceptions etc but the main skeleton can be used. Do NOT forget to unlock in a finally so that your code does not block indefinitelly):

public String getXML() {  
  String xml = getXMLFromDatabase();  
  if(xml == null) {  
     if(glocalLock.tryLock()) {  
        try{  
            xml = getXMLFromThirdParty();  
            storeXMLToDatabase(xml);       
        }  
        finally {  
            globalLock.unlock(); //ok! got XML and stored in DB. Wake-up others!  
        }  
     }
    else {  
         try{ //Another thread got the lock and will do the query. Just wait on lock!     
             globalLock.lock();  
         }  
         finally {
             //woken up but the xml is already fetched  
             xml = getXMLFromDatabase();  
             globalLock.unlock();  
         }   
     }    
  return xml;  
}