[tahoe-dev] odd determination of unhealthy

[Greg Troxel]

David-Sarah Hopwood <david-sarah at jacaranda.org> writes:

> Greg Troxel wrote:
>> I did a 'tahoe --check --raw' on my main directory in the pubgrid, and
>> got an odd report: the object is 'unhealthy', even though all 10 shares
>> are present, plus two leftover shares of a previous version.  I would
>> call that healthy with some stale bits in need of garbage collection,
>> but maybe I don't get it.
>
> The current checker is supposed to consider a file to be unhealthy if fewer
> than N (in this case 10) shares are present on any servers.

ok; that's reasonable even if it doesn't match the 3/7/10 notion.

> So I don't
> understand why this file was considered unhealthy.

I suspect it has to do with the 2 shares of the previous sequence
number.  But I don't have any evidence.

> We want to change the health criterion to take into account share
> distribution (http://tahoe-lafs.org/trac/tahoe-lafs/ticket/614). By that
> criterion, this file should still be considered healthy, but only just
> for the default parameters. (It has a happiness of 7, because the shares
> are distributed between only 7 servers.)

That makes sense to me.

>> Another oddity is that running check on another client resulted in
>> seeing very few shares.  Both clients have real IP addresses.  But, the
>> one that saw only 4 shares is running (w/o storage) on a machine that
>> also runs a storage node.
>
> This seems to be because only 5 servers responded. The servers that did
> respond, reported the same shares that they had in the first check.
>
> Had the client just been started? In that case it might not have
> connected to all available storage servers yet
> (http://tahoe-lafs.org/trac/tahoe-lafs/ticket/719).

I tested this morning, and the client process was started on 28 July.
It claimed to be connected to ~10 servers.

But, there could be a firewall involved.  Are there keepalives on the
client/server connections.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 194 bytes
Desc: not available
URL: <http://tahoe-lafs.org/pipermail/tahoe-dev/attachments/20100816/1d72526f/attachment.pgp>