Ok, we found the problem, but not the resolution. The problem is caused by a limitation in NT 4 that the paged kernel memory cannot exceed 192 MB. In 2000+ the limit is 256 MB. Once it hits that limit, the read errors start. The problem is that all the memory is used by FW! Once the limit is hit, if we stop the foxweb service all the memory is released back to the OS. It basically goes back to almost where it was when the machine is first booted (about 16 megs). The reason this is a new occurrence is because of increased table sizes and traffic. It is all related to the size and frequency of large tables and files being opened. Turns out this is really well documented.
But why isnt the memory released when the channel completes it's process? The processes gradually increase the memory usage until it blows - anywhere from 5-10 hours.
Sent by FoxWeb Support on 07/27/2004 11:29:34 PM:
Did you get a chance to look at fwstart.log? It may contain useful information that may help you diagnose this problem. If you see anything suspicious please send us a copy via email.
Also, have any errors been logged in the FoxWeb error log since the problem started happening? It's always possible that the error log files (error.dbf, cdx and fpt) are corrupted or inaccessible.
Yet another possibility is hard-drive corruption, or insufficient user rights.
Sent by Jack Evans on 07/27/2004 11:40:25 AM:
I am baffled - We are still running FW 1.2 on a NT4.0 box (Sorry Pandelis) and all of a sudden in the last couple of weeks we are getting 1104 Error reading file errors. It is completely intermittentent and not related to any specific line of code. Once it occurs, it clogs up all the channels and they all report the same errors, but in different processes. This almost always occurs in a USE statement. In watching it, I also notice that there are zero length temp files in the working temp folder. It would appear that fw_error is trying to dump to that folder and cannot. I have changed temp folders to a completely different drive, no change. Other processes that use the tables seem unaffected. No other server processes are having trouble reading files either (this is a very heavily accessed server, so we know the problem is somehow related to the FW/VFP combination.