Fixed: OpLocks caused the corruptions/slowness (Was: How Samba let us down)

[Chris de Vidal]

--- Neil Hoggarth <neil.hoggarth at physiol.ox.ac.uk>
wrote:
> On Thu, 24 Oct 2002, Chris de Vidal wrote:
> 
> > I'd be happy to let the group know.  I'm not
> positive
> > we'll reenable anything but kernel oplocks,
> though.
> > We have work to do.
> 
> The "kernel oplocks" parameter affects how Unix
> processes accessing the
> file interact with SMB oplocks. Enabling kernel
> oplocks on a share which
> doesn't have SMB oplocks turned on shouldn't make
> any difference, I'd
> have thought.
> 
> If I understand your description correctly you don't
> have Unix processes
> interacting with the stored files; your Linux box is
> acting purely as an
> SMB file server for Windows clients? All the file
> accesses come in from
> the net, via Samba?

Yes.

>  In this case you probably want
> to leave kernel
> oplocks off ('cos they buy you nothing,
> functionally, and there have
> been suggestions that Linux kernel bugs causing
> problems with them). The
> interesting test is whether either of:
> 
> 	oplocks = yes
> 	level2 oplocks = no
> 
> or
> 
> 	oplocks = yes
> 	level2 oplocks = yes
> 
> work.
> 
> If your corruption returns *and you can show that
> your network and
> clients are working properly* (ie. no oplock break
> messages are getting
> lost or being ignored by client machines - which
> probably requires
> Ethernet packet captures) then it's probably "Red
> Alert" time.

I'll keep these guidelines in mind.

> Also: don't think that if you establish the
> existence of a priority 1
> bug then it is all over - if you're experiencing a
> bug that the team
> can't reproduce themselves then it doesn't mean that
> there isn't a bug,
> but it does mean that they're going to need a lot of
> help characterizing
> and finding it.

The team probably would have to install Elixir's Opus
and process large flat db files (Fox Pro, I think)
with multiple processes on multiple servers... in
other words, it probably isn't going to happen.  The
corruption will remain possible with other users.

On the other hand, several people have confirmed it to
be a problem with multiple clients accessing Microsoft
Access, which is a relatively cheap test.

Anyway, for us it is working fine, so I really have
little motivation for fixing the problem other than to
give back to the contributions given.  We are probably
going to run the checks as you mentioned above,
probably going to submit a bug report if we find
something, and probably going to submit a
documentation patch, but can't do much more (can't
spend much more time on it.. life goes on here).

/dev/idal

__________________________________________________
Do you Yahoo!?
Y! Web Hosting - Let the expert host your web site
http://webhosting.yahoo.com/