Re: [9fans] venti ports and productization

[jmk@plan9.bell-labs.com]

On Wed Oct 22 10:58:46 EDT 2003, miller@hamnavoe.demon.co.uk wrote:
> > We have had some problems
> > with the big Venti server, but my understanding is that
> > all of the problems appeared to be disk corruption rather
> > than software error.
>
> The software could do a bit better in recovering from hardware errors.
> If a bad sector makes a venti block unreadable, rdarena and wrarena
> could use the redundant information in the arena trailer to skip the
> bad block and continue, but they just give up.  Worse, if checkarenas
> (with -f) fails to read a block, instead of using the trailer information
> to skip the block, it tries to correct the mismatch between block label
> and trailer by copying garbage into the trailer entry from the buffer
> it just failed to read into.  Oops.
>
> -- Richard

That's pretty much right, in my limited experience. When I started looking
at the errors we have I hacked up the arena read programme I have to do some
resynchronisation, it needs more work, though.
Looking at the errors we have, I don't think I'd ever want to write back
into the arena to try to fix it, that stuff with VtTypeCorrupt just seems
wrong (as does the definition of VtTypeCorrupt as VtMaxType).