bdf consuming high CPU

[George Denyer_1]

Hi,
I have a problem with three servers running hp-ux 11i v2. All three servers were performing poorly and we found out the reason was that on all three there was a process with very high CPU usage, the strange thing is that the process was a "bdf /net". After killing this process all three servers went back to normal, from 0% idle to 80% or 90% idle. Even stranger is that this is the second time we find this in a few months.

The output from "ps -ef | grep bdf" showed this:
root 2462 2461 255 Jul 12 ? 7214:01 bdf /net
root 2461 1 0 Jul 12 ? 0:00 sh -c bdf /net 2>&1

The servers are two rx4640 running SG and SAP and a rx2620 running SAP.

I executed "bdf /net" and it seemed to hang but inmediatly started consuming a high percentage of CPU.

Has anybody seen this before, or have any ideas?

Thank you,

George Denyer.

[Peter Nikitka]

Hi,

the directory /net is normally used for the default -hosts map of the automounter.

It should be configured with an option '-nobrowse' (NB: this is the Solaris flavoured option, check for the correct HP-UX name!), so no mounts are performed when accessing this directory directly.

Perhaps check for patches of bdf and automount.

mfG Peter

[Steven E. Protter]

Shalom George,

If bdf is using a lot of cpu resources I can think of several possible causes.

1) There is a stale nfs link and bdf is hanging up trying to get information on the remote fs
2) bdf has a problem and needs to be patched.

SEP

[Josh Dalziel]

You need to make sure that all of you mount file systems are ok and not stale

[Dave Olker]

Hi George,

I'm not able to reproduce this on my 11i v2 systems, so if you are able to reproduce it at will I'd be interested in investigating the reason why it's hanging. Let me know if this is something you'd be interested in.

Regards,

Dave

[George Denyer_1]

Hi Dave,
Yes, I can reproduce this behaviour at will on this three servers, and yes, I'd be interested in finding out the reason why this happens.

To me, there are two misteries here, one is why the command hangs, and the other is what is executing it.

What would you need to investigate this issue?

G. Denyer

[Dave Olker]

Hi George,

Before we deal with the "how is it getting launched" let's focus on "why is it hanging".

The first things I'd want to see are copies of your /etc/fstab, /etc/mnttab, and any/all AutoFS map files. That would at least show me what's currently mounted and what options you're telling AutoFS to use. Hopefully that would help me reproduce the problem locally, assuming it's a class problem.

I'd also like to see the output of "swlist -l product" as that would show me what patches you have installed. Again - more information to help me reproduce it.

If I'm unable to reproduce the problem locally then I'll have you reproduce it again and collect some data, but for now I'd prefer to try reproducing it locally so I don't have to disrupt your environment by collecting lots of data.

Regards,

Dave

[Bill Hassell]

bdf (and df and find and du) are not a good tools to run on an NFS node. bdf itself has a fairly high overhead to simply gather the required data from the directories but going over the network for this information may lead to problems. bdf /net seems like a mistake in some script, so the next time it happens, you need the complete parent tree (ppid's up to the top) so you can see what started bdf /net. It is not clear why this command would ever be run on a production server. Note that in a production server, all NFS servers must never go down.