I believe this is fixed in amqp v5.0.8 via https://github.com/celery/py-amqp/commit/f4fd4f952dded9ea1006e82642f2c15008bb68d3. I suspect that an exception is thrown during connection cleanup for TLS connections (in sock.unwrap() - hence why we didn't see it before for non-TLS), leading to a dangling socket. We've not seen any logging of this exception, so this must have been caught elsewhere.
Upper constraints suggest this fixed version should be available from Yoga, but for earlier releases the upper pin for amqp means you could hit the issue if using TLS in Xena and earlier.
A workaround should be to increase 'conn_pool_min_size' to be closer to (or equal to) 'rpc_conn_pool_size' so that connections are never dynamically closed in normal operation, although this has the downside of requiring a lot more connections to RabbitMQ by default.
I believe this is fixed in amqp v5.0.8 via https:/ /github. com/celery/ py-amqp/ commit/ f4fd4f952dded9e a1006e82642f2c1 5008bb68d3. I suspect that an exception is thrown during connection cleanup for TLS connections (in sock.unwrap() - hence why we didn't see it before for non-TLS), leading to a dangling socket. We've not seen any logging of this exception, so this must have been caught elsewhere.
Upper constraints suggest this fixed version should be available from Yoga, but for earlier releases the upper pin for amqp means you could hit the issue if using TLS in Xena and earlier.
A workaround should be to increase 'conn_pool_ min_size' to be closer to (or equal to) 'rpc_conn_ pool_size' so that connections are never dynamically closed in normal operation, although this has the downside of requiring a lot more connections to RabbitMQ by default.