ppker / tailscale / 9357177100

52%

push

github

ipn/ipnlocal: improve testability of random node selection

In order to test the sticky last suggestion code, a test was written for
LocalBackend.SuggestExitNode but it contains a random number generator
which makes writing comprehensive tests very difficult. This doesn't
change how the last suggestion works, but it adds some infrastructure to
make that easier in a later PR.

This adds func parameters for the two randomized parts: breaking ties
between DERP regions and breaking ties between nodes. This way tests can
validate the entire list of tied options, rather than expecting a
particular outcome given a particular random seed.

As a result of this, the global random number generator can be used
rather than seeding a local one each time.

In order to see the tied nodes for the location based (i.e. Mullvad)
case, pickWeighted needed to return a slice instead of a single
arbitrary option, so there is a small change in how that works.

Updates tailscale/corp#19681

Change-Id: I83c48a752
Signed-off-by: Adrian Dewhurst <adrian@tailscale.com>

19 of 24 new or added lines in 1 file covered. (79.17%)

47459 of 93002 relevant lines covered (51.03%)

0.57 hits per line