gvinum + geli + gjournal

[Anthony Chavez]

On 11/21/2011 11:25 PM, Lev Serebryakov wrote:
> Hello, Anthony.

Hi Lev, thank you for responding!

> You wrote 22 ноября 2011 г., 3:37:53:
> 
>> I have 2 questions regarding graid5:
>> 1. Why hasn't it made its way to the base FreeBSD distribution yet?
>   Because its code is ugly, almost non-understandable, raises some
> questions, and lays from style(9) a miles away.
>   I'm rewriting it now, to support same functionality (and to be
> compatible with existing arrays), but as I didn't see any interest in
> it (yes, I'm using it by myself, but this your message is first
> "external" interest I could see for few years), I spend not much time
> on it.

It seems most admins prefer hardware RAID5, but having a good
implementation in software for those under tight budgets seems like it
would be very worthwhile.

>> 2. Has anyone done any rigorous testing with it?
>   It depends on your meaning of word ``rigorous'' :)
>   I'm using it for couple of years, I had several server failures in
>  this time (panics), Ive chaged two failed drives for same size and
>  after that migrate to larger HDDs (5x500Gb -> 5x2Tb) non-stop
>  (almost, growfs need you to unmount FS), and I run some tests in VM
>  after each change. But there is no formal test-suite (yet, but,
>  again, see above about lack of "external" interest, which discourage
>  me).

Were the panics that you mentioned directly caused by graid5?  If so,
under what conditions?

Please don't get discouraged.  I think graid5 could be very useful.
About 5 years ago, I had a client that wanted to set up a fileserver but
the lack of software RAID5 at the time turned them off of FreeBSD
initially.  I managed to convince them that hardware RAID5 was a smart
investment, but they weren't terribly happy about having to spend the
extra cash.

You might want to base your opinion on the number of downloads the code
receives rather than responses here.  I, for one, have been using
FreeBSD since version 1.0, but haven't posted to the mailing lists very
often at all. ;-)

>    One known semi-bug semi-feature. It cache writes aggressively
>  (which allows it to have perfect linear write speed and avoid
>  re-reads to compute checksum in most real-life cases), and
>  in case of failures, this could lead to "unexpected softupdates
>  inconsistences," as FFS doesn't send FLUSH commands at all, and module
>  could not now, which writers should go synchronous, with minimal
>  possible delay.

If that's the case, this would not affect me, as I'm using gjournal and
mounting asynchronously rather than SoftUpdates.

Any other features / bugs I should be aware of?  The data that I'm
intending to migrate is rather precious.  I cannot afford to lose any of
it, and will use graid3 if I must.

>> I have 2 questions regarding graid5:
>> 1. Why hasn't it made its way to the base FreeBSD distribution yet?
>> 2. Has anyone done any rigorous testing with it?
>   BTW, if you could write list of good test-cases which should be in
> (VM-based) automated test-suite, it will be very helpful.
>   I could easily miss some corner cases.
>   But, please, be aware, that this test-suite should be implemented
> via standard UNIX API plus "disk disconnection," which is possible
> under VirtualBox 4.x, so tests should be formulated via newfs, file
> operations, fsck, etc. toolset, not via some internal kernel APIs.

When time permits, I'll have a look at the code, but it may be a few
months before I will be able to do so.  I'm rather busy at the moment.

When you mention "disk disconnection," do you mean *physical*
disconnection (e.g., before a umount)?

-- 
Anthony Chavez                                  http://hexadecagram.org/
mailto:acc at hexadecagram.org                    xmpp:acc at hexadecagram.org