CAM status: CCB request completed with an error

[Ronald F. Guilmette]

In message <20140403074009.GA73196 at neutralgood.org>, 
kpneal at pobox.com wrote:

>On Wed, Apr 02, 2014 at 03:48:21AM -0700, Ronald F. Guilmette wrote:
>> 
>> What the bleep is a "CCB request", and why would it fail, over and over and 
>over
>> again, ad infinitum?
>
>Dunno. Try asking on either one of the freebsd-scsi or freebsd-geom lists.

OK.  I will.

One footnote to my earlier rant:  The problems all seem to be arising from
my USB sticks... *not* from ordinary (sATA) drives as I had previously
believed.

I have had _many_ problems, over time, with the FreeBSD USB driver.  I guess
this is just another one.  Sigh.

>> Did somebody fail to implement some logic that
>> would recognize when there have been several thousand such errors already...

>Well, syslogd does eliminate duplicate messages with a message about having
>avoiding repeating itself X number of times.

Yes.

>But that only works for log
>lines that are a single line long.

Yes.

>It would be a useful addition to have syslogd recognize repeated groups
>of log lines and give them the same treatment.

I disagree.  I believe that it would be a helluva lot more useful, in
this case, if the USB driver just simply kept track of which devices are
"failing" as far as it is concerned, and then just simply avoided per-
forming an *infinite* set of retries on those, all in rapid succession.
Performing such a set of retries... in rapid succession... is arguably
stupid, wasteful, and potentially hazardous to the health of innumerable
/var partitions, worldwide.

(And by the way, there isn't a damn thing wrong with my USB sticks.  *They*
are perfectly fine.  It is the ______ FreeBSD USB driver that is screwing
up.  It apparently gets confused easily... and then just stays that way.)

>It would be a shame to have various subsystems in the kernel have to roll
>their own.

I would argue that *every* program and/or driver should, ideally, make at
least some effort *on its own* to avoid generating a infinite number of
syslog messages, particularly if all the messages are relating directly
to a single allegedly syslog-worthy event, as in the case that I began
this rant about. It seems to me that this is just a matter of common
courtesy... or perhaps "uncommon" would be more accurate.

Infinite sequences of syslog messages, occurring on unattended machines,
lead inevitably (and sometimes rapidly) to "/var: No space left on device".
This is an undesirable outcome of the highest order.


Regards,
rfg