Process hangs with 100% SYS

[Fredrik Lännergren]

Uhm...this got a bit odder now that I finally managed to get it to happen when I start the program under tusc.

When five of them are started at the same time, three succeed while two freeze with 100% Sys. The output between successful and stopped processes are identical up to that point.

The tusc output halts for the freezing processes between a close() (on liblber.so on a mounted ClearCase path) call and a stat() (on libc.2 at a local opt path) call.

I.e., nothing to do with our code at all... those are not our libs. I guess this thread is too old now for any replies, but if anyone has an idea on how to attack this, I'd be grateful to hear it.

/Fredrik

[A. Clay Stephenson]

The tusc output halts for the freezing processes between a close() (on liblber.so on a mounted ClearCase path) call and a stat() (on libc.2 at a local opt path) call.

I suppose that a "mounted ClearCase path" means an NFS mounted filesystem. My first approach would be to move this shared library to the local machine and then use SHLIB_PATH to point to the local copy. This would get NFS out of the picture because this may be NFS related rather than a problem directly with the code itself.

[Fredrik Lännergren]

Finally.

It was the network mount that hung the entire machine at 100% Sys. For some reason, starting several programs that access the same file simultaneously sometimes hangs some of the processes. When I moved the libs to local paths, everything worked just fine.

[Fredrik Lännergren]

It was a network mount that caused the hangs when several instances of the same program were started simultaneously. Moving the libs to local paths cleared it.