Re: Consensus drops node

[Hans Schnehl <torvallenator@xxxxxxxxx>]

On Mon, Mar 23, 2009 at 03:53:16PM -0700, Kyle Williams wrote:
> Yes, I've seen this too.  The node "JanusPAv2" just dropped out of the mix
> about 7 days ago.I wasn't too concerned since I was testing the max through
> put of a embedded device.
> 
> I though it was probably something I did when I was half-awake, but now I'm
> starting to question it a bit more...hrm..
> Last I saw my node, it had a Guard and Fast status.  Then one morning, it
> was gone and no traffic was passing through.
> 
> 
> 
> On Mon, Mar 23, 2009 at 2:51 PM, Hans Schnehl <torvallenator@xxxxxxxxx>wrote:
> 
> >
> > Hi all,
> >
> > six hours ago the node 'vallenator's performance  began to decrease
> > gradually
> > from around 18 mbit/s to at present about 1 mbit/s.
> >
> > The node was running rather stable at it's upper level (see above) with an
> > uptime
> > of now 32 days.
> >
> > I did not do anything on this running machine, so I initially suspected my
> > hoster
> > to intercept or interfere, not necessarily with a motivation against Tor,
> > but
> > maybe for more or less genious technical reasons. But they did not do so.
> >
> > What happened is, the node was actually 'forgotte'n, or dropped out of
> > consensus
> > by the authoritive routers, it still is as of now. Since I do not have
> > logging
> > enabled I am unable to search or provide logs with possible traces of what
> > was
> > happening.
> > Anybody else experiencing this?
> >
> > I'll keep the node running 'as is' and watch for any changes to occur
> > within the
> > next hours but probably I'll take the chance to rebuild a few things on
> > that
> > machine if if there is no change.
> >
> > Any ideas about this ?
> >
> > Regards
> >
> > Hans
> >


The node came up again by itself.  It did so after 18 hours of gradually but
signifigantly loosing connections to the rest of the 'Tor-world' until there was
only noise left. I assume the built in scheduled 18hourly publishing of descriptors
caught some or all of the authorities by surprise in realising this box is still
there and never stopped working ;). For these 18 hours the node was not listed
in the current consensus. By now it is, again, with the number of connections
and utilized bandwidth increasing.

If you find vallenator on kprogs torstatus (trunk) this incident shows nicely
in the graphics.

Apparently there was a short but complete loss of connectivity or the node became
unresponsive for just a few minutes in the beginning of this.  Due to the number
of at that time already established connections, these and throughput reestablished
very fast back on the previous level. But since the node appeared to have gone
for the authorities, gradually connections to other nodes and the 'Tor-world'
vanished.

The blackout itself might have been caused by some apprentice (like me) practising
to pull and reinsert plugs on the hosters side or other reasonable activity.

Kyle mentioned, this happened to him when he 'was testing the max througphut of
a embedded device' and it sounds a little like he was expecting this device to
fail sooner or later, and maybe it didn't fail and just became unresponsive at
the wrong moment.

My node was for a short period of time unresponsive to or disconnected from the
general network. The turn to publish descriptors fell exactly within this period
of unresponsiveness.  So there probably were no descriptors sent to the authorities
at the time they expected these to be there, ergo them dropping the router from
the lists. The node didn't really notice and sat there chewing it's fingernails
and wondering about all the evil in this world. ahemm.  Earlier versions of Tor
did publish descriptors  when sending it a SIGHUP, but this is not possible
anymore.  So probably presently  the only way of solving this is to either restart
Tor, or just wait 18 hours until Tor will republish it's descriptors, and hopefully
this doesn't happen in another unresponsive period.

Regards