[tahoe-dev] newbie question

erpo41 at gmail.com erpo41 at gmail.com
Mon Jul 30 12:13:53 UTC 2012


I believe "the introducer is down" is self-correctable by replacing (or
supplementing) the introducer with gossip. Gnutella clients use a similar
system and it is very reliable even when peers often go down or change IPs.

The difficult problem, from my point of view, is that tahoe's performance
and reliability characteristics are not like a vanilla filesystem on a
spinning disk. Treating it like one is an abstraction, and the abstraction
leaks. Example: it is a very infrequent occurrence that my ext4 filesystem
encounters an error in the middle of writing a file, so the OS doesn't
handle this very well. If a file manager (e.g. nautilus) has to deal with a
dircap mounted using FUSE, it won't have any clue that it's not a regular
HDD due to the abstraction layer and won't be able to handle failure
conditions. This happens right now when I browse my file server from my
laptop using nautilus on sshfs. If the wireless connection drops, nautilus
will sit there for hours trying to complete an operation instead of
detecting that the network is down and giving up.

As far as filing tickets, I believe there is a lot of preliminary work to
be done first--if a person can't troubleshoot a problem, a person can't
program a computer to troubleshoot that problem. In this case, that means
really good error messages and diagnostic tools. I'm happy to write them,
but having to learn darcs takes it from "fun programming time" into "work"
territory for me, and the docs say that the git bridge is not 100% yet.

Thanks,
Eric
 On Jul 30, 2012 5:37 AM, "Greg Troxel" <gdt at ir.bbn.com> wrote:

>
> "erpo41 at gmail.com" <erpo41 at gmail.com> writes:
>
> >> There should be a command-line version of the status page, but there
> >> seems not to be.
> >> I use "wget http://127.0.0.1:3456/ && more index.html" :-)
> >> Seriously, am I the only one who runs nodes on computers I am not
> >> sitting at?
> >
> > I also run tahoe at a computer I am not sitting at. There is a storage
> > server on my home network responsible for running tahoe, and it listens
> on
> > an address in my private IP range. I call up the web UI from my desktop.
>
> I never let the tahoe  web gateway listen on other than 127.0.0.1
>
> > When I'm away from home and want to check my node status, an ssh tunnel
> > with port forwarding does the job.
>
> Sure, I can do that too, and have.
>
> > A CLI status checking command would be cool. However, a better solution
> > would be to make tahoe so reliable and self-correcting that users don't
> > ever need to check the status. (I.e. 5-second wizard-based configuration,
> > automatic firewall/NAT penetration, retry everything until it succeeds or
> > the world comes to an end, check and repair data automatically, etc...).
>
> File those tickets!
>
> But "the introducer is down" isn't self-correctable - my node is fine,
> as far as I can tell, and the web page shows what's going on instantly.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tahoe-lafs.org/pipermail/tahoe-dev/attachments/20120730/535514ea/attachment.html>


More information about the tahoe-dev mailing list