I don't yet understand the use case - you seem to be implying that you are writing code for others to use, and so are not requiring them to use logback, but you wish to take responsibility for clearing up after logback for them? That seems to me to mix layers of abstraction incorrectly. I would suggest that the layer of abstraction at which logback is chosen as the slf4j implementation, and the socket appender is configured, is also the layer of abstraction at which responsibility for shutting down logback is taken. If you are writing code against SLF4J only, for downstream users to integrate into their apps using an SLF4J implementation of their choice, you should not be concerned with logback at all (including configuring it to use an appender and shutting it down).