Re: recent rev. take a very long time to get client working

[Roger Dingledine <arma@xxxxxxx>]

On Tue, Oct 23, 2007 at 02:58:47PM +0800, Li-Hui Zhou wrote:
> I have no idea why recent svn revs take a very long time to get the
> client function working. A lot of
> Oct 23 14:50:48.239 [notice] I learned some more directory information, but not
> enough to build a circuit.
> Oct 23 14:50:55.840 [notice] I learned some more directory information, but not
> enough to build a circuit.
> Oct 23 14:52:06.682 [notice] I learned some more directory information, but not
> enough to build a circuit.
> such as informations in the log.

It's because recent svn revisions use the new v3 directory protocol,
and there are currently only two v3 directory authorities. Both of
these authorities upgraded to a new version a few hours ago that had a
crash bug, and one of them is still down. It takes a majority to build a
networkstatus consensus, and a majority of 2 is 2. Without a consensus,
your Tor client refuses to build any circuits.

Later, there will be more than two v3 authorities, so this will happen
less.

We're pondering letting Tor clients start with an obsolete networkstatus
consensus (say, if you have one from 12 hours ago). This is good from
a "now it works" perspective, but bad because a malicious directory
cache can hand you one from 12 hours ago and claim there aren't any
newer ones, and now you behave differently from all the other users who
weren't tricked.

--Roger