The expected lock order is cs_vNodes => cs_peerManager. Unfortunately the ReattemptInitialBroadcast method locks the opposite way, and causes a potential deadlock that has been uncovered by the getavaaddr functional test that mocks the scheduler time (for the statistics to be updated) and increase the chance of the reverse lock condition to occur.

Ref T1696.

ninja all check-all

With debug:

for i in {1..1000}; do ./test/functional/test_runner.py abc_p2p_getavaaddr || break; done