NFS ACCESS requests lock up 10.6.8 NFS/AFP/SMB server for 60 seconds

[foop]

I’m posting this in the forlorn hope that somebody will know what on earth is going on. I’m an eternal optimist.

We have an elderly Xeon-based XServe attached to an XServe Raid that serves network home directories to Macs (by AFP), Linux boxes (by NFS) and a couple of Windows machines (by SMB). It’s running 10.6.8. It authenticates to an Open Directory server (running 10.4.11; it’s not broken so I haven’t touched it).

Over the last couple of weeks it has occasionally been locking up for around 60 seconds. This happens at roughly the same time past the hour, although not every hour. I’ve been running Wireshark to see what has been going on.

This is the usual course of events:

1. A rapid series of NFS v3 ACCESS requests from a linux box, usually for a directory in the root of a user’s home directory, reach the server.
2. The file server sends a LDAP request to the OD server for “cn=groups,dc=[machine],dc=[our domain],dc=ac,dc=uk” wholeSubtree.
3. The OD server sends a reply announcing success with two groups (“admin” and “staff”, neither of which are OD groups).
4. The file server sends a retransmission of the original search request.
5. Packets come into the file server, but nothing goes out.
6. The request times out?
7. 60 seconds later everything gets back to normal in a flood of outgoing packets.

If anyone has any ideas why this happens, that would be really good. Maybe I could tweak the timeout on the OD requests somehow (assuming that’s the problem)?

[gleventhal]

Aren’t Admin and Staff both built-in OD groups? I thought they were fairly necessary. Not saying that this is necessarily the problem you are having, just an observation. As for your problem, perhaps some type of check is occurring and then timing out? I have seen things like this happen with a Proxy check. Try running fs_usage on the Mac File Server while it is timing out to see what processes are getting it tied up. What does the console or System.log say during the timeout period? Try looking into the sar command to log some of the data perhaps as well.

[Author]

Posts