Samba 2.2.3a on Solaris 8

[David Lee]

On Thu, 18 Apr 2002, Jeremy Allison wrote:

> [...]
> Ok - discussed this with Andrew last night. It seems that this is only
> a problem on Solaris. Solaris seems to have *serious* issues with fcntl
> locks with multiple processes contending for locks. No other system we
> run on seems to have this problem (they have their own problems :-).

Disclaimer:  I'm not a Sun sales-droid.  Absolutely not.  The following is
not intended to be Sun "sales-speak" (even if, at initial glance, it might
appear to be so). 

But I wonder whether at least part of the reason this tends only to be
seen on Solaris just might be because Solaris is robust and _good_ enough
to be pushed so far that it exposes these problems (whereas perhaps other
OSes have other problems that kick in before this sort of loading ever
arises in the first place). 

In other words, perverse though it may at first seem, might Solaris sites
be seeing this problem because Solaris is a _good_ OS, capable of being
pushed to such limits?  (Have other OSes stumbled on the lower slopes of
the load mountain?)

Mere speculation, of course, as our only significant Samba is on Solaris. 

Straw poll: how many non-Solaris sites exist at all, handling, say, 300+
simultaneous connections, including high smbd turnover?

(We are around the 1,000 mark; a typical hour-long "pacct" record contains
some 2600 "smbd" processes: in excess of one "smbd" every two seconds
*sustained*.  That would be quite a load on "connections.tdb", even if it
were non-bursty.  We had to backtrack to 2.0.x. some months ago: prudence
and politics prevent us risking moving forwards again.)


> Dave CB - can you investigate this within Sun please. This is a *critical*
> part of Samba, we may have to look into a solaris-specific workaround and
> this would be bad.

Any thoughts, Dave (CB), with your inside knowledge of Sun?  (Technics and
politics!)

If there is a Sun bug, it would be good to get this at least recognised,
and preferably fixed.  And if there were a patch, then we could quickly
knock up some "configure.in" stuff to detect the presence/absence of the
relevant patch and perhaps issue suitable warning messages.  (Or even a
run-time detection, with a warning in "log.smbd"?)

-- 

:  David Lee                                I.T. Service          :
:  Systems Programmer                       Computer Centre       :
:                                           University of Durham  :
:  http://www.dur.ac.uk/t.d.lee/            South Road            :
:                                           Durham                :
:  Phone: +44 191 374 2882                  U.K.                  :