IBD stalls if addnode peer is in IBD #34096

Starting with two fresh nodes A and B: If you start node A and (before node A completes IBD) start a second node B with -addnode={address of node A}, node B will stall and not proceed in IBD.

mkdir -p /tmp/nodea /tmp/nodeb

cat > /tmp/nodea/bitcoin.conf << EOF
prune=1000
bind=127.0.0.1:8333
EOF

cat > /tmp/nodeb/bitcoin.conf << EOF
prune=1000
rpcport=18331
bind=127.0.0.1:18335
addnode=127.0.0.1:8333
EOF

and then start both nodes however you prefer, e.g.

bitcoind -datadir=/tmp/nodea --daemonwait
bitcoind -datadir=/tmp/nodeb --daemonwait

And node B will not advance:

tail /tmp/nodeb/debug.log

Shutdown node A and node B comes to life!:

bitcoin-cli -datadir=/tmp/nodea stop
tail /tmp/nodeb/debug.log

I don't think that this is a permanent stall - just a delay. If we have other peers, I would expect us to sync headers with them after 1) a new block is found and INVed to us by some peer or 2) m_headers_sync_timeout is reached - whichever comes first.

In case 2), we would sync headers with a new peer - which may or may not be the old peer, this is depending on the order in which peers are processed by net, see https://github.com/bitcoin/bitcoin/blob/master/src/net_processing.cpp#L6492-L6500.