From: Volker Lendecke on
On Mon, Jan 21, 2008 at 09:15:15AM +0100, Robert Gehr wrote:
> This morning our main Samba server was pretty unresponsive running at a
> 30+ load average.
> After digging a bit what I found in the logs was the following.
>
> No data on inotify fd?!
> [2008/01/21 07:02:59, 0] smbd/notify_inotify.c:inotify_handler(234)
>
> That entry appears a billion zillion times in the log file until the
> file system is filled up leaving no space on the device and sending
> samba into oblivion.

It filled the disk? Do you have "max log size = 0" set? I
would have expected Samba to rotate the logs. Or do you log
to syslog?

> I also found a couple of entries in the logs like this.
>
> [2008/01/18 15:50:46, 1] smbd/notify_inotify.c:watch_destructor(342)
> inotify_rm_watch returned Invalid argument

I'll have to look at the code, but this might be the same
reason.

As a workaround, you might want to set "kernel change notify = no"

Do you still have the log files around from when this all
started? Without a way to somehow reproduce it, I'd guess
this will be quite difficult to fix.

Volker