NPE in AbstractConnectionFactory that causes connection listeners to be skipped

[deje07]

In what version(s) of Spring AMQP are you seeing this issue?

2.4.6, but current main branch is also affected

Describe the bug

The following line can produce NPE:

spring-amqp/spring-rabbit/src/main/java/org/springframework/amqp/rabbit/connection/AbstractConnectionFactory.java

Line 663 in b27a18f

int protocolClassId = cause.getReason().protocolClassId();

as according to the documentation of ShutdownSignalException.getReason(): "the reason, if any". Under some circumstances, like a low-level TCP Connection Reset on the connection, the Rabbit client library will not fill in the reason in the exception, therefore this code will throw NPE and connection listeners will not be called.

To Reproduce

To reproduce:

1. set up a ConnectionListener on a ConnectionFactory (like CachingConnectionFactory)
2. establish connection to a RabbitMQ server
3. force the TCP connection to RST via some networking/os tools. I used Linux and forced the connection to break using sudo iptables -A INPUT -p tcp --dport 5672 -j REJECT --reject-with tcp-reset

Expected behavior

The ConnectionListener's onShutdown(cause) should be called. If the Reason does not define channel, then connection-level problem should be assumed.

Sample

I can create a sample if really required but based on the linked source code line it is clearly visible that NPE is possible.

[deje07]

Additional note: Rabbit's AMQConnection class does create null reason for shutdown in multiple places:

1. in handleHeartbeatFailure()
2. in handleFailure(Throwable) where it handles unexpected exceptions on connection's communications

In all code paths after invoking shutdown with null reason, doFinalShutdown is called, tearing down the connection itself.

Based on these I think it is safe to assume that if reason is missing, it should be treated as a connection level problem.

[garyrussell]

No need for a sample; thanks for reporting.