peer discovery failing on consecutive network join

[konradkonrad]

Here is an issue I could only reproduce in a private network, but seems bad enough to report.

Steps to reproduce:

1) Run a private network with the docker-compose.yml from https://gist.github.com/konradkonrad/54e623673741b3704726#file-docker-compose-yml like this:
docker-compose scale bootstrap=1 miner=1 eth=2 debug=0

2) Join with an interactive node like this:

docker run -it --name connector --link bootstrap:bootstrap ethereum/client-python -c eth.network_id=1337 -b 'enode://288b97262895b1c7ec61cf314c2e2004407d0a5dc77566877aad1f2a36659c8b698f4b56fd06c4a0c0bf007b4cfb3e7122d907da3b005fa90e724441902eb19e@bootstrap:30303' run --fake

watch it sync to head and fill the peers (watch log for INFO:p2p.peermgr missing peers known=4 min_peers=5 num_peers=3)

3) Stop and restart the node (Ctrl-C followed by docker start -i connector)

4) Check the peers count/watch the log fill up with connection attempts with the ever same node:

INFO:p2p.peermgr    connecting random   node=<Node(288b9726)>
INFO:p2p.peermgr    missing peers   known=1 min_peers=5 num_peers=1
INFO:p2p.peermgr    connecting random   node=<Node(288b9726)>
INFO:p2p.peermgr    missing peers   known=1 min_peers=5 num_peers=1
INFO:p2p.peermgr    connecting random   node=<Node(288b9726)>
INFO:p2p.peermgr    missing peers   known=1 min_peers=5 num_peers=1
INFO:p2p.peermgr    connecting random   node=<Node(288b9726)>
INFO:p2p.peermgr    missing peers   known=1 min_peers=5 num_peers=1

Expected outcome: connector-node reconnects to the whole private network on second attempt.

Observed outcome: connector-node seems to only see the bootstrap-node on second attempt (block propagation is still working though).