Closed vkomenda closed 5 years ago
Although… this still doesn't handle the case of two subsequent InProgress changes, does it? Those would leave the failed candidates as zombie entries in the peer_epochs map?
I'm putting together a workaround. Let's see if we can optimise it.
Still failling; but it makes sense that the outputs differ, of course, since the re-joined node didn't receive all the batches. So I guess it's just the test?
I've just ran drop_and_readd
1000 times and it never failed!
I expected the QHB test to fail though because I'm still updating it. In addition to missed batches, node 0 does not send all of its user transactions, and the unmodified verify_batches
will fail.
Rebased and ready for review. Comments welcome!
Closes #341
Prior to this PR we had only addition of new peers in the sender queue. This PR allows removal of those peers which are validators that become superseded by other validators in the next epoch.