A sequence of all brokers down errors can be received in an idle client, a producer for example, even if the broker disconnections are just idle disconnections, that is normal broker TCP closes, and not TCP resets or connection failures.
To test it, start a cluster with a short idle disconnection time like with
python interactive_broker_version.py --kraft 4.0.0 --conf '{"conf": ["connections.max.idle.ms=10000"]}'
then start a examples/producer with cluster bootstrap servers but don't produce anything. After some time the errors will start appearing.
It doesn't happen only when the disconnections happen all at the same time, but it's the down state that isn't reset if other broker connections aren't needed (sparse connections) contributing to raising the error even if a single connection is closed.
A sequence of all brokers down errors can be received in an idle client, a producer for example, even if the broker disconnections are just idle disconnections, that is normal broker TCP closes, and not TCP resets or connection failures.
To test it, start a cluster with a short idle disconnection time like with
then start a
examples/producerwith cluster bootstrap servers but don't produce anything. After some time the errors will start appearing.It doesn't happen only when the disconnections happen all at the same time, but it's the down state that isn't reset if other broker connections aren't needed (sparse connections) contributing to raising the error even if a single connection is closed.