GitLab

When running Tor in small testing networks, much of the time the bandwidth-weights line does not appear in the directory-footer in the consensus files. The log file shows messages like this:

Consensus with empty bandwidth: G=852123 M=0 E=0 D=569253 T=1421376

The code that counts up these bandwidth values is in networkstatus_compute_consensus in dirvote.c, specifically around line 1590 in Tor master as of now.

The code that prints this error is in networkstatus_compute_bw_weights_v10 in dirvote.c.

I believe that it is an error not to produce bandwidth-weights in the event that we have no knowledge of bandwidth for a given position. For example, if D is zero because there are no nodes that serve as exits+guards, shouldn't we just adjust the weights accordingly? We may still have functional guards and functional exits just because we have no node that serves as both.

Since this is for weighting purposes, why are T, D, E, G, and M all initialized to 0 instead of 1? I think the default weight should be 1, meaning all positions are selected equally, and any bandwidth above 1 should be used to increase the weight. Does this sound right?

If that is not desired, then I request that we at least initialize these values to one for testing networks. One patch is attached for each of these options.