cppkafka::Consumer Destructor Can Throw about cppkafka

[mfontanini]

Good catch, I didn't notice that one. Sounds like a good idea! There's no logging strategy at this point. At some point I wanted to use rdkafka's logging as it already exists but it's not exposed publicly, which kind of sucks. I guess for now just catching that would work.

[Cfretz244]

Oh yeah, it would be great if it could tap into the existing logging.
Personally, I've never been a big fan of libraries logging in the first place, but there are situations like this where it's really the only viable response.

[mfontanini]

I think cppkafka could actually log through rdkafka via the configured log callback or rd_kafka_log_print (the default) in case there's none. Reopening so I don't forget about this one.

[arvidn]

One way to make it possible for a client to detect these kinds of errors would be to expose close() to the public, which can then throw. But yes, the destructor should be assumed to be called while unwinding because of another exception.

[accelerated]

Would this work? Not sure about the facility name...I used LOG_USER

Consumer::~Consumer() {
    try {
        // make sure to destroy the function closures. in case they hold kafka
        // objects, they will need to be destroyed before we destroy the handle
        assignment_callback_ = nullptr;
        revocation_callback_ = nullptr;
        rebalance_error_callback_ = nullptr;
        close();
    }
    catch (const Exception& ex) {
        auto& callback = get_configuration().get_log_callback();
        if (callback) {
            callback(*this, LOG_ERR, "user", ex.what());
        }
        else {
            rd_kafka_log_print(get_handle(), LOG_ERR, "user", ex.what());
        }
    }
}

[mfontanini]

lol at "Reopening so I don't forget about this one.".

That looks good. I would probably change "user" to "cppkafka" so it's clear where the error is coming from.

[accelerated]

Maybe this should be closed now...