A change to maintain keys (and open up a path to node restarts), has caused some issues in the test (nodes rejoining as opposed to being lost).
That test is sorted now (though client mem is on the high side now, that’s a different issue). So we should be back on the straight and narrow for PRs there.