While using a queue can be as simple as write and read, sometimes you need a little extra, and that's where keyed data queues come in.

On the IBM i, it's easy to create a physical file that has no keys and write to and read from that file (it's a little harder in the non-IBM i SQL world, but it can be done). A simple data queue is like an unkeyed physical file: you add records to the file and then read them off in sequential order. You can jump around by relative record number, but we really don't use that technique a lot these days, at least not in production programs. Instead, we key our files and use those keys to access the data.

The Key to the Data

You can either key the physical file itself or add a logical view, which provides access to the physical file in the order defined in the logical. Even if you've keyed the physical, you may want to provide alternate access paths to the data by creating more views. For a DDS-defined logical view over a physical file, you create a key by specifying the fields that make up the key. In SQL, you do the same thing by creating a VIEW over your table (remember, on the IBM i, tables and physical files are nearly identical) and specifying the columns. You can do some additional magic using operations on the keys as well, taking substrings or changing the case or converting to numeric.

With data queues, you don't have nearly so much flexibility. A non-keyed data queue has a single data element whose length you define when creating the queue. A keyed data queue is almost identical except it has a second element, the key. Like the data element, your only real option when defining the key is to define the length. You then specify the key as an alphanumeric value when you add an entry to the queue.

Creating a Keyed Data Queue

In my previous article, I talked about creating and deleting data queues. Deleting a keyed data queue is exactly the same as deleting a no-keyed data queue, so I won't repeat that part here. Creating a keyed data queue requires only a couple of minor changes.

Data queue . . . . . . . . . . .             Name                

Library . . . . . . . . . . .    *CURLIB    Name, *CURLIB      

Type . . . . . . . . . . . . . . *STD         *STD, *DDM          

Maximum entry length . . . . . .               1-64512            

Force to auxiliary storage . . . *NO          *NO, *YES          

Sequence . . . . . . . . . . . . *KEYED       *FIFO, *LIFO, *KEYED

Key length . . . . . . . . . . .               1-256                        

Include sender ID . . . . . . .  *NO          *NO, *YES                    

Queue size:                                   

Maximum number of entries . .  *MAX16MB      Number, *MAX16MB, *MAX2GB    

Initial number of entries . .  16            Number                      

Automatic reclaim . . . . . . .  *NO          *NO, *YES                    

Text 'description' . . . . . . .  *BLANK                                     

Figure 1: After specifying *KEYED as the sequence, the CRTDTAQ command presents a few more parameters.

The name is the qualified name of the data queue. Leave *STD as the type, although I hope to delve into the entire concept of DDM data queues another day. In fact, as with the last article, I'm not going to go through all of the keywords in an effort to make you a data queue expert. Instead, I just want to cover the key parameters (no pun intended) that you'll need to use a keyed data queue, and they're pretty minimal. Make sure to specify *KEYED as the sequence, and then the maximum entry length is the length of the data element while the key length is the length of the key element.

Programming for a Keyed Data Queue

Now that your keyed data queue is created, you can program for it. You'll have to use a slightly extended version of the APIs in order to take advantage of the keys. I've included the new prototypes here.

     D SendData       PR                 ExtPgm('QSNDDTAQ')

     D   Dtaqnam                     10a   const

     D   Dtaqlib                     10a   const

     D   Dtaqlen                     5p 0 const

     D   Data                            const like(myMessage)

     D   Keylen                       3p 0 option(*nopass) const

     D   Key                               option(*nopass) const like(myKey)

     D ReceiveData     PR                 ExtPgm('QRCVDTAQ')

     D   Dtaqnam                     10a   const

     D   Dtaqlib                     10a   const

     D   Dtaqlen                     5p 0

     D   Data                             like(myMessage)

     D   WaitTime                     5p 0 const

     D   Keyorder                    2a   option(*nopass) const

     D   Keylen                       3p 0 option(*nopass) const

     D   Key                               option(*nopass) const like(myKey)

     D   Senderlen                   3p 0 option(*nopass) const

   D   Sender                       1   option(*nopass) const

Like the last article, SendData and ReceiveData are the prototypes for QSNDDTAQ and QRCVDTAQ, the two APIs most used with data queues. These prototypes are the same as the prototypes in the other article but with a few additional parameters. For example, the SendData prototype has two additional fields, Keylen and Key, while ReceiveData has five additional parameters, which I'll explain in a moment. But before I do that, let me give you a real-world scenario in which I would use a keyed data queue.

Data queues are all about sending and receiving data asynchronously—that is, the sending program can send a message even if the receiving program isn't ready. In fact, if the receiver gets busy, lots and lots of people can send messages and they will be queued up for processing. The receiver can then pop the messages off the queue one at a time and process them as resources are available. This concept works perfectly as long as the data fits within a single message. The maximum message size is about 64KB, so a lot of messages will fit. But unfortunately, not all will fit, especially when we're dealing with more complex transactions such as orders. Take a big order with a few hundred lines and each line is a few hundred bytes and you're well over the limit. So what to do? Well, you could send the transaction on multiple messages, but then you run into the issue of trying to tie all those messages together. And that's where the keyed data queue comes in!

Think about this problem: when two or more users are pumping single-message requests into a data queue, sequence and timing don't matter. The processing program pops the next request and processes it. But if the data spans multiple messages, then you can run into problems of interleaving; the queue may have a few records from one request followed by one from a second, followed by another from the first again. Add more requests, the situation gets worst.

You avoid this by using two queues, one keyed and one not keyed. First, you design a unique transaction ID, which could be as simple as the next number from a data area. The requester writes all the transaction data to the keyed data queue using the transaction ID as the key. Then and only then does it send the transaction ID in a message to the non-keyed data queue. The processor gets that message, parses out the key, and then uses that key to read the data.

Here's the code:

     // Loop sending data

     While getNextMessage(myMessage);

       SendData( 'APPDATAQ': 'APPLIB': %size(myMessage): myMessage: %size(myKey): myKey);

     Enddo

     SendData( 'APPREQQ': 'APPLIB': %size(myKey): myKey);

This snippet assumes that you've already called the routine that fills myKey with the next transactionID. The routine getNextMessage populates myMessage with the next piece of the transaction and returns false if no more exists. If there is data, it's sent to the data queue (APPDATAQ) and then getNextMessage is called again. Once all the data has been loaded onto the queue, you send the transaction ID to the unkeyed request queue.

     ReceiveData( 'APPREQQ': 'APPLIB': lenReceived: yourKey: 60);

     If (lenReceived > 0);

       Dou lenReceived > 0;

         ReceiveData('APPREQQ': 'APPLIB': lenReceived: yourMessage: 0:

                     'EQ': %size(yourKey): yourKey: 0: ' ');

         If (lenReceived > 0);

           // Process data record

         Endif;

       Enddo;

     Endif;

The receiver sits on the requests queue. When a request is received, it's assumed to be the key to the data queue. The process then reads all the records from the data queue and processes them. No timeout is needed on this read because the data has been loaded ahead of time. The process routine could process the records individually as they are read or could save them all in an array or even a temporary data file. The point is that the processor can reliably get all the data for a transaction and process it together.

And that's a good look at my favorite use of keyed data queues!