Using a vended product that relies on J2SE and uses basic listFiles(java.io.filter) to get at contents of a directory. For some reason, on one of our platforms (HP intel running Linux RHEL 3, JDK 1.5), this call is, according to our vendor, "occasionally" returning null (implying that pathname has suddenly become unaccessible) when trying to read a directory.
Most of the time there's no issue (i.e. we discover, open, read and parse the 1000s of files resident here), but when there is, it happens w/out pattern (varying time to error) and happens whether or not directory is local or remote disk. The app using this framework is viewing a dir w/ 130k files, reads about 50 at a time (via separate threads), and moves the "finished" files to an archive directory.
We've had better success moving handful of files into a working directory and running against the smaller dir, but for various reasons this isn't ideal for us.
Also oddly, we have no issues running on our older IBM hardware (slower, less RAM, different bus, etc., but same RHEL version and patch level and JDK).
Vendor is working to come up w/ better error description and a recovery method (currently, program simply ends).
Any ideas on what underlying issue is would be much appreciated.
It's not ideal but you might try decompiling the vendors code and checking out what they are actually doing under the covers (using Jad - http://www.kpdus.com/jad.html or something similar). Sometimes this can reveal the real bug and at least maybe you can find a better work-around (or push on the vendor a bit harder).
It seems odd that such a core java method would fail like that... you might also search Sun's bug database to see if there are any bug tickets reported on this (unless it is truly the vendors issue).