RabbitMQ channel_max configuration: per-node or cluster-wide scope?

Goal

Diagnose intermittent connection‑pool exhaustion in a RabbitMQ cluster where clients appear to exceed the configured channel_max without triggering a CHANNEL_ERROR.

Uncertainty

The RabbitMQ documentation specifies a channel_max setting that limits the number of channels per connection, defaulting to 2047. However, it does not explicitly state whether this cap is enforced per node in a cluster or globally across all nodes. In practice, a client that opens many channels on a single node can reach the limit, but when mirrored queues or sharded connections distribute traffic across nodes, the same client may continue creating channels beyond the configured value until one node’s threshold is hit.

Questions

• Is the channel_max limit applied independently to each node in a RabbitMQ cluster, or is it a global constraint across the entire cluster?
• If the limit is per-node, how does a multi‑node connection (e.g., via mirrored queues) affect the overall channel count seen by the broker?
• What diagnostic metrics or commands can reliably reveal whether a client is exceeding the channel_max on any node in a cluster?