Managing Share Memory Segments

Concerns

1. When we ask for an ID to a shared memory segment and a segment already exists with that key we don't do any checking to see that APPX created this memory and we aren't colliding with some other non-APPX process. I think our shared memory header record should include some known identifier at the beginning that we can inspect.
2. When we ask for an ID to a shared memory segment and a segment already exists with that key we don't use the number of slots out of the shared memory header. We use the number of slots defined in the APPX environment variable. Those values may not be the same. If the value from the environment variable is smaller we limit our access to the memory slots, if larger then our attach to the memory fails completely and silently. It's up the an administrator to manually locate and remove the shared memory segment, run APPX to create the new larger segment, then either wait for all sessions to log out and back in or force everyone to log out and back in to get attached to the new memory segment. Seems like we could put a stale indicator byte in our shared memory header block and when we need to extend the shared memory, set the stale bit, copy the contents of the old segment, delete it, allocate a larger segment, copy the data into it, and continue running. As processes go to update their slot if the header stale bit is set just un-attach/reattach to get to the current segment.
3. It looks like the SessionID used to find a memory slot is based on the row-id of the USAGE file record. Doesn't this create a problem if more than one APPX install path on a system share the same shared memory segment? Since this does not seem like a good thing, each copy can already be configured to use a different key to isolate each to it's own shared memory segment. But for safety I think we should have some information in our shared memory header that tells us the instance of APPX this segment is tied to. That way we would could catch attempts to cross over and use the wrong shared memory key.
4. Synchronized access to our shared memory segment(s) is controlled by locking either a byte in a dummy file on unix or a mutex on windows. In either case the name to the file or mutex is a fixed name. If we have several keyed memory segments in use for different installs of APPX they are all sharing the same synchronization lock. I think this locking should be based on the path to the install.
5. The license server has code that will clean up the shared memory slots left in use by orphan processes. However this will only work if the license server is running on the system that also contains the shared memory slot. This model does not work on Novell or other shared usage file installs. Not sure what to do about this.
6. I think the whole process is too manual. Coming up with key values and setting both Keys and Sizes as environment variable to activate the functionality seems wrong. Why don't we just have it on by default with a way to turn it off. Why don't we just determine a key based on the APPXPATH using the ftok() function, determine a nice initial size based on the license count and re-allocate as mentioned in item 2 above if we need more slots.
7. Problem: If you run a process like Vendors from the 1EX Main Menu and it blows up with a File Can't Be Accessed error, upon returning from the error screen to the menu the monitor still says it's running Vendors Input.
   

Test Results, Comments, Open Issues

The notes above also stated, "If APPX_MONITOR_SLOTS is too small, nothing tragic happens - new processes just won't update their status information in the monitor table." That sounds like it should be tested, so we killed all sessions, including the license server and appxd, edited the .env file to set the value to 2, and restarted APPX. We used the first login to run the APPX Monitor, and it properly showed that session plus the license server. Using Alt-F1, we started a second session, which also showed up on the Monitor display, a bit of a surprise there. In fact, subsequent sessions also showed up, until we had six sessions running. The seventh did not appear on the table. Closing each of those in the table worked normally, and closing the session(s) that were not shown did not cause any problems, although we did notice that the entry for the license server disappeared temporarily one time. It might be worth testing this with a slight variation - logging in as different users, to check to see if the user count is somehow a factor. But the fact that it allowed six sessions (not four) seems to make that unlikely. Follow-up note: Pete looked into the code to determine why we more than two sessions were being logged, and discovered that the actual number of slots is the next higher multiple of 8 from the value entered.

One facet that is not thoroughly tested is what the APPX Monitor will show for background tasks. It appears that an additional session will be displayed, but the background session we were testing with died too quickly to see what was really happening. This should be investigated further.

Suggested Enhancements:

1. It would be nice if the APPX Monitor showed the process ID as well as the User ID.
2. It would be nice if, during query processes, it would show the name of the process itself, rather than QSLCT and QSORT.
3. The configuration process is too "manual".  The APPX Monitor should always be on and should automatically configure itself.

• icps Command on Linux: