I'm chasing down some performance problems on a Linux server, and I've setup BSD process accounting to track down what might be causing the load. However, I'm seeing a lot of processes called "Filesystem", like this:
sed root ?? 0.00 secs Thu Oct 22 06:00 Filesystem F root ?? 0.00 secs Thu Oct 22 06:00 sed root ?? 0.00 secs Thu Oct 22 06:00 Filesystem F root ?? 0.00 secs Thu Oct 22 06:00 Filesystem F root ?? 0.00 secs Thu Oct 22 06:00 which root ?? 0.00 secs Thu Oct 22 06:00 which root ?? 0.00 secs Thu Oct 22 06:00 cut root ?? 0.00 secs Thu Oct 22 06:00 grep root ?? 0.00 secs Thu Oct 22 06:00 Filesystem F root ?? 0.00 secs Thu Oct 22 06:00 cut root ?? 0.00 secs Thu Oct 22 06:00 Filesystem root ?? 0.01 secs Thu Oct 22 06:00 Filesystem F root ?? 0.00 secs Thu Oct 22 06:00 grep root ?? 0.00 secs Thu Oct 22 06:00 Filesystem root ?? 0.00 secs Thu Oct 22 06:00
I don't have a program called "Filesystem" on this machine, and Google isn't telling me anything useful about what it might represent. Has anyone dug into the source code or otherwise knows why these would be popping up?
I am just grasping at straws here, don't know that much about psacct, but I'm guessing it's a "dummy" process used to represent Kernel time spent reading/writing to the filesystem? What's your iowait% look like? Are you performing many disk intensive operations? What is your disk subsystem like?
Found it.
I was being insufficiently thorough when I determined that I didn't have a program called
Filesystem
on the machine -- in fact, there is a script calledFilesystem
in/usr/lib/ocf/resource.d/heartbeat/Filesystem
that manages the mounted filesystems in the cluster manager (DRBD volumes). It turns out it's the monitor processes for those filesystems that were causing all my consternation, and there was nothing supernatural about them after all.