[PATCH] New debug backend "ringbuf"

[Andrew Bartlett]

On Sun, 2017-01-08 at 16:59 +0100, Ralph Böhme wrote:
> On Sun, Jan 08, 2017 at 12:49:57PM +0100, Ralph Böhme wrote:
> > On Sun, Jan 08, 2017 at 10:48:26AM +0100, Volker Lendecke wrote:
> > > On Sun, Jan 08, 2017 at 10:08:53AM +0100, Ralph Böhme wrote:
> > > > This one made my day when trying to fix a subtle bug in some
> > > > asyc
> > > > code in ctdbd_conn.c I'm currently working on. The bug was
> > > > subject
> > > > to timing effects, I could easily reproduce it at log level 1,
> > > > but
> > > > as soon as I cranked up log level it hid away. Arg.
> > > > 
> > > > I tried moving the logfile to a memdisk, but even with that,
> > > > timings
> > > > were different enough so I still couldn't reproduce it.
> > > > Ahhhhhh...
> > > > 
> > > > That's when I thought of adding a new logging backend that
> > > > simply
> > > > writes all log messages into a memory buffer, et voila, with
> > > > this I
> > > > was able to reproduce the issue at log level 10. Bug found and
> > > > fixed. :)
> > > > 
> > > > To make use of it when hunting for a bug, you must be able to
> > > > identify a place in the codepath where you know things are gone
> > > > south and you can cause a panic. You then attach with gdb to
> > > > the
> > > > panicked (and in developer mode sleeping process, otherwise
> > > > grab the
> > > > corefile) and run:
> > > > 
> > > > (gdb) dump binary memory samba.log debug_ringbuf
> > > > debug_ringbuf+SIZE
> > > > 
> > > > The default size is 1 MB, but this can be changed by a backend
> > > > option:
> > > > 
> > > >  logging = ringbuf:size=NBYTES
> > > > 
> > > > Fwiw, the ringbuffer is of course only allocated if the backend
> > > > is
> > > > used.
> > > 
> > > This looks really interesting. Question -- would you be fine if I
> > > wrote a smbcontrol to grab the log, similar to for example pool-
> > > usage?
> > 
> > great idea! If you're not going to do it, I will. Let's see who's
> > faster. :)
> 
> me. :)
> 
> Updated patchset attached. While I was at it I found it worthwhile
> improving
> messaging.idl and pidl to use and add support for mixed with/without
> value
> enums. What do you think?

Can you register it in source4/lib/messaging.c as well, so we can ask a
'samba' process the same question?

Thanks,

Andrew Bartlett