[Samba] diagnosing broken pipe errors

[Ernie Buford]

Andrew Bartlett wrote:
> On Tue, 2008-02-26 at 16:20 -0500, Ernie Buford wrote:
>   
>> I'm hoping to get some guidance diagnosing intermittent smbd
>> errors that have been cropping up in the last 2 months.  They
>> are sporadic, difficult to reproduce, and don't reveal much of
>> a pattern so far except that the majority are generated by a
>> couple of clients, although several clients appear to be
>> affected.  Typical symptom on the client is a hang when
>> attempting to look at the share in Windows Explorer, or
>> table data suddenly disappears when working in MS Access,
>> or sometimes no symptom at all.  It seems the most repeatable
>> error condition is when a drive is first mapped after bootup,
>> but even that is not consistent.
>>     
>
>   
>> I've found some indication via web search that this might be
>> related to network problems, so that's one reason I'm
>> hoping for diagnostic tips.  I'm aware of no recent network
>> changes that might be causing this problem, and I've tried
>> the worst-behaving machine on a different ethernet cable
>> and switch port with no improvement.
>>     
>
> I would replace your network cards and switches with better quality
> gear, if possible.
>
>   

I'd be more suspicious of the network gear if there were
other networking symptoms pointing in that direction,
but so far, these samba errors are the only indication that
anything is wrong.  Other means of pushing data to and from
these workstations work fine.  We are on a Cisco Catalyst
3650G Series Gigabit switch and use Dell Precision
workstations, except for the most error-plagued machine,
which is an Optiplex GX620.  These aren't exactly cheap,
entry-level machines.  Still, I'll try to find a way to see if
I can isolate the problem to particular NICs or the
switch itself (I can fall back to a 100Mb switch that we
still have online).

Thanks for your input.
EB


> Broken pipe errors should not occour on a LAN, unless the client is
> rebooted.  Multiuser control of MS Access can be a particular pain in
> this regard, and many network set 'veto oplock files' to exclude *.mdb.
> This prevents oplocks (which require server-initiated communication)
> from being granted. 
>
> The getpeername error may well be harmless, simply due to Windows being
> silly and connecting to both 139 and 445, then dropping one. 
>
> Andrew Bartlett
>
>