[tor-relays] task-6329 / tor relays stats patch

[Karsten Loesing]

On 7/28/12 9:30 PM, Michael Zeltner wrote:
>> Thanks for sending the patch!  Most of it looks good, but I took out the
>> --family-for-fp option which, I think, isn't entirely correct.  This
>> option looks for relays having the given fingerprint in their family
>> line, but it doesn't cross-check if the given relay has those relays in
>> its family line, too.  The option also doesn't look at nicknames (of
>> relays with the Named flag) in family lines, which means it might miss
>> some family members.
> 
> Yes, I noticed the first bit as well, and have since better understood how
> families work - the attached patch should fix that. It looks for mutually
> listed relays via both fingerprints and nicknames (in case the relays have the
> named flag).
> 
> I hope I nailed it this time round.

Thanks for the patch!  I rebased your patch to current master and
rewrote some parts of it.  Please take a look at the task-6329-family
branch in my public metrics-tasks repository:

https://gitweb.torproject.org/karsten/metrics-tasks.git/shortlog/refs/heads/task-6329-family

If you like it, I'll squash our two commits, making you the commit
author, and will push it to the official repo.

>> Oh, and it would be really cool to have a -F or
>> --by-family option to aggregate by families.
> 
> Yeah, I'd like that too. I haven't worked on grouping yet but I'll look into it
> soon.
> 
>> I wonder if we should do
>> the whole family-calculating business inside Onionoo and add a
>> family_number field which has the same number for relays in the same
>> family.  The script could then look up the family number of the given
>> relay and display all relays with that family number.  Would that make
>> sense?  Want to extend the script once Onionoo has such a field?
> 
> It'd certainly make it quicker to filter out the relays one wants to take a
> closer look at - if one is interested in the impact of families. And sure. It'd
> certainly make the grouping business easier because there's a family-uid one
> can easily group by :)

Okay, I'll look into adding this new family_number field soon.  Hard to
say if I mean this week or this month by that. ;)  If you come up with a
patch for --by-family before that, let's merge that one first.

Thanks!
Karsten