Hi there,

I have a heavily used linux data server. It is running RH7.3 with a custom compiled kernel. The problem with it is that after about a week of running the NFS system locks up. The computer is still running but the whole NFS system is broken and wont die. The scripts will not shutdown the nfs system neither will issuing a kill –9 on nfs processes (I also tried –15). At the same time there where about 50 smb –D processes running that also could not be killed. When I looked at the dmesg there where several of the following errors

Kernel: NFS : Task xxxxx cannot get a request slot

I do not know if this error is relevant to the problem. When I tried to reboot the computer the shutdown process immediately hung up. I then tried a halt. This also hung up not getting anywhere. In the end I resorted to a hard reset.

Does anyone have any ideas why all this may be happening?

Thanks

paul

Take a look at 'ps -aux' and look for processes chewing cpu time.

You can ssh into the box for this.

There is a process allocation system called 'nice' which can alter priorities.
'man nice' for more info.

Regards,
Peter

I did check the cpu usage. I have an old laptop that constantly displays xosview or perfmeter outputs from key computers so I can see at a glance if any are being hammered.

The problem seems to be due to either the massive spawning of smb –D processes or something else causing the NFS system to lock solid.

I did not try niceing any of the processes because none of them where using any cpu.

The main problem is that when the computer goes into this state I cannot shut it down nicely. The shutdown gets to the NFS system and then hangs. I have tried leaving it in this state over night but no change. It does not timeout or complete. I have to resort to the reset button. This is not good for the system and due to the fact it has 2x75GB scsi drives on it, it takes ages to check the filesystems. The drives exist from a prior installation of Linux and the boss did not want me to try upgrading the filesystems to ext3 incase of data loss.

Thanks

Paul

Check out 'man nfs' and look at the "hard, soft & intr" options.

You can add these to the startup script

Regards,
Peter

I have all my mounts set to hard. I have had problems with soft mounts. I have seen some file corruption using soft mounting. This is apparently a known problem and it is recomended that soft mounting is only used for read-only systems. At least this is what I have read in the admin books I have. The server contains our source code and so must not become corrupted. I have not seen the Intr option before. I will look into this one and see if it is recomended for a read/write file system.

Thanks for the suggestion

paul