[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] Desynchronizing dm-raid1



>>>>> "Malahal" == malahal  <malahal us ibm com> writes:

Malahal> Your problem is very similar to an iSCSI problem sumitted
Malahal> here:
Malahal> http://now.netapp.com/NOW/cgi-bin/bol?Type=Detail&Display=137902

Can't see it without a netapp account...


Malahal> Fortunately, you can detect the problem and the I/O can be
Malahal> retried if possible.

That's not entirely trivial.  Obviously the original data that matches
the checksum is gone.  And I don't want to blindly regenerate the
checksum from the new data (how do I know this was an "ok" kind of
corruption?).

The only one that has that knowledge is the filesystem.  And if the
filesystem needs to be integrity-aware, I'd much rather have it Do The
Right Thing than teach it to inspect and reissue I/Os that come back
with -EDANGERWILLROBINSON.

With ext2 at least this is not some rare corner case.  It happens
hundreds of times during an unpack of a kernel tarball.  We'd end up
burning many, many cycles doing retries.

-- 
Martin K. Petersen	Oracle Linux Engineering


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]