slowdown of zfs (tx->tx)

[Artem Belevich]

On Thu, Jan 10, 2013 at 11:39 AM, Nicolas Rachinsky
<fbsd-mas-0 at ml.turing-complete.org> wrote:
> There is an UDMA_CRC_Error_Count of 17 and 20 for the two disks with
> checksum errors. The other disks have values between 0 and 5.
>
> And yes, there have been timeouts some time ago. Since the problem did
> occur without the timeout occuring again, I considered the timeouts to
> be unrelated. And then I forgot them. :(
>
>
> But shouldn't timeouts either produce correct data after a retry or
> a read/write error otherwise?

if I see CRC counter incrementing often enough that's a good
indication that something is wrong. It does not mean that those
transactions were the ones that corrupted data, but rather as an
indication that things are not right with particular device. It may be
a false alarm as CRC errors may happen under normal conditions, but
non-trivial number of them is a good sign of trouble.

--Artem