When the broker is using ASYNCIO and customer is using AMQP client, and they (simulate) disk device failure, the broker does not fail over to the slave to continue operation.

This does not occur if only using CORE client.

This issue does not occur when using NIO on the broker.

Customer (Mike Pearce) has discussed this with Martyn Taylor who has already coded a fix based on a similar issue on another instance. Mike Pearce has discussed with Martyn one extra bit of code that may need a similar fix applied to fully overcome the issue.