PRB: Scrolling Message "Record not available ... Please Wait" (125576)

RESOLUTION

A workstation that gets this message will eventually be able to complete the pending process, although this could take a considerable amount of time.

If TTS is running, try turning it off. FoxPro does not provide direct support for Novell TTS. However, Novell provides a library (Nwtts.dll) that works with the FoxPro Library Construction Kit (LCK) to accomplish transactional processing.

Support for transactions has been implemented in Visual FoxPro. For more information, search for 'Transactions' in the Visual FoxPro Online Help. If TTS is not running, the problem is most likely a contention problem on the server. This means that the server is receiving more requests for processing transactions than it can deal with so they're backing up in the server's process queue. Eventually the server will catch up, the queue will be emptied out, and things will be back to normal.

Avoiding This Network Storm

It's important to note that because there is no error condition, there is no error to trap. Once this message appears, the network is already experiencing a network storm and the only resolution is to have all users stop what they're doing and wait for it to clear. The preferred approach is to take steps to avoid the storm in the first place.

Two known approaches involve upgrading to a more powerful network or moderating the flow of process requests to the network server. While upgrading the network may be prohibitive, it is often relatively easy to moderate the flow of process requests to the server.

Moderating the Flow of Process Requests

Here are two approaches you can use to moderate the flow of process requests to the network server:

• Change over to a client/server database system such as SQL Server.
  
  This concentrates more of the processing directly on the server and not over the network thereby reducing the amount of transaction processing handled by the network server.
• Change your record locking/adding scheme to a more efficient one, and/or introduce a small delay in the routines to allow the network server time to catch up after each process.
  
  NOTE: Minimizing the number and complexity of index tags may help this situation. However, this problem has been reproduced on a network with only four workstations and with as few as four tags in the index.

Coding Techniques Effective in Dealing with This Problem

Two coding techniques that have proven effective in dealing with this problem are: changing from record locking to file locking and introducing a wait state (delay) after adding to or changing any tables.

Changing to file locking means that you use FLOCK() rather than RLOCK() when to add or change a record. It's important that you keep each lock only long enough to make the addition or change, and then release it.

Although going to a file locking scheme may seem to be less efficient, it has the added benefit of spreading out process requests and leaves your program code in complete control. Because you are requesting a lock on the file before you process any records, you can provide more appropriate options when locks are unavailable due to heavy traffic.

In effect, you are shifting processing control from the server to your program. You can add a delay by including a WAIT "" TIMEOUT n command immediately after releasing the file lock. To help spread out process requests to the server, you can vary the amount of time to wait (specified by n) depending on the time it takes to obtain the lock. One algorithm to accomplish this calls for requesting the lock within a loop and counting the iterations until the lock is obtained. Then use this count in a formula to arrive at a delay time. The longer it takes to obtain a lock, the longer the wait state. When you consider this taking place on several (or many) workstations, the net effect is to flatten out the flow of process requests to the server. This eliminates the peak conditions that result in the network storm.