[tor-relays] "stable" flag voting differences

[Karsten Loesing]

On 10/27/13 7:32 PM, starlight.2013q4 at binnacle.cx wrote:
> I observed an interesting behavior in the
> authority votes regarding relay stability
> and am curious if anyone can comment.
> 
> Have a new relay, about ten days old.
> Relay is marked
> 
>    Fast Guard Running Stable Valid
> 
> One 10 minute network outage three days ago
> and then a 68 minute outage one day ago
> for repair.  Overall about 4680 seconds down
> over the last week, and if one uses a
> 646220 second timeframe per this post
> 
>    https://lists.torproject.org/pipermail/tor-relays/2013-July/002255.html
> 
> It works out to about 99.27% uptime,
> which seems within the 98% threshold
> quoted in the posting.
> 
> What was interesting was that
> 
>    Faravahar
>    gabelmoo
>    maatuska
>    tor26
>    urras
> 
> all immediately removed the "stable" flag when
> the relay came back online after maintenance,
> while
> 
>    dannenberg
>    moria1
> 
> left the stable flag in place.  The two
> remaining authorities
> 
>    dizum
>    turtles
> 
> have never set the stable flag due to heavy
> link congestion between AS286/AS3265
> for 'dizum' and congestion within AS6461
> for 'turtles'.
> 
> Then exactly 24-hours after the end of the
> outage the five naysayers turned the
> stable flag back on.
> 
> Mainly a point of curiosity, but I'd guess
> from this that the first five authorities
> listed above are using a much shorter
> window for evaluating stability while
> the second two are using the seven-day
> window.  I couldn't find per-authority
> logging of this category of information
> so it's hard to know.
> 
> Perhaps it's relevant that the node was
> removed from the consensus directory 
> briefly immediately after the outage ended.
> 
> Finally it is interesting that 'dizum' and
> 'turtles' have set the "guard" flag even
> though they consider the relay unstable.
> This is a recent development.  Earlier
> the two impaired-path authorities had
> the guard flag set off.

Hi,

thanks for looking into these flag voting differences!  This is not an
explanation of the issues you raise, but just in case you want to dig
deeper, here are some starting points:

https://metrics.torproject.org/consensus-health.html#overlap

"flag-thresholds" in votes-20??-??.tar.bz2 tarballs on
https://metrics.torproject.org/data.html

https://trac.torproject.org/projects/tor/ticket/8797

In short, it would be neat to find out exact reasons for authorities
voting differently, and having fixes would be really cool!

All the best,
Karsten