[volunteergrid2-l] I'm live. :)

[Shawn Willden]

On Thu, Apr 5, 2012 at 2:28 PM, Shawn Willden <shawn at willden.org> wrote:

> On Thu, Apr 5, 2012 at 1:31 PM, David-Sarah Hopwood <
> david-sarah at jacaranda.org> wrote:
>
>> On 05/04/12 20:11, Shawn Willden wrote:
>> > On Thu, Apr 5, 2012 at 10:08 AM, David-Sarah Hopwood <
>> david-sarah at jacaranda.org
>> > <mailto:david-sarah at jacaranda.org>> wrote:
>> >
>> >     Even if there were no bugs and no network problems in the normal
>> case, setting
>> >     shares.happy = shares.total still puts you in a situation where a
>> single failure
>> >     will compromise upload availability.
>> >
>> > Only if shares.happy = shares.total = number_of_nodes -- which is what
>> I was suggesting.
>> >  What I would really like is, say, shares.happy = shares.total =
>> number_of_nodes - delta,
>> > where delta is a small factor, perhaps 1 or 2, depending on how
>> resilient you want uploads
>> > to be.
>>
>> Oh, ok. Yes, if that doesn't work then it's a bug.
>
>
> It didn't... until I got in a room with you and Zooko and Brian, then it
> worked fine :-)
>
> I haven't tried it since.  I should see if I can find a way to reproduce
> it.
>

I just updated my configuration to use:

shares.happy=20
shares.total=20
shares.needed=12

VG2 currently has 21 active nodes accepting shares (well, I thought it did;
see below), so this should work, but I can't upload anything.  The error
message I get is very interesting:

UploadUnhappinessError: shares could be placed on only 19 server(s) such
that any 12 of them have enough shares to recover the file, but we were
asked to place shares on at least 20 such servers. (placed all 20 shares,
want to place shares on at least 20 servers such that any 12 of them have
enough shares to recover the file, *sent 22 queries to 21 servers, 20
queries placed some shares, 2 placed none (of which 2 placed none due to
the server being full and 0 placed none due to an error)*)

So, two things jump out at me:  First, that Tahoe sent 22 queries to 21
servers trying to place 20 shares.  Huh?  Why 22 queries?

The second is that we have one full node... which actually I knew but had
forgotten (the owner of that node has committed to making more space
available sometime in the next three months).  So shares.happy and
shares.total should probably be set to 19 to room for one node to be down
or to have a transient problem without causing upload failures.  Still,
with 20 nodes with space available and actually accepting shares, this
should be fine.

Side question to anyone who knows the immutable upload code:  Would it help
if the full node were marked as "read-only"?  Would that prevent my node
from trying to contact it?

Thanks,

-- 
Shawn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://tahoe-lafs.org/cgi-bin/mailman/private/volunteergrid2-l/attachments/20120406/7a9484e9/attachment-0001.html>