What does TRANSPORT error mean?

[johnrgregg3]

I get lots of these in my error_cb routine, (error -195), but then sometimes I get my on_delivery callback called, with no indication of an error. This suggests to me that the TRANSPORT error isn't really an error. Maybe it means we can't get through to all the brokers in the cluster, but only some of them? If it is not accompanied by an ALL_BROKERS_DOWN, can I safely ignore it?

[edenhill]

The TRANSPORT error represent connection errors with the broker, typically connection closed or connection refused.
The error_cb should give you more context through the error string.

If you are on osx this is typically caused by 'localhost' resolving to both ipv4 and ipv6 but the broker is only listening on one of those address families, in which case you can set 'broker.address.family': 'v4'.

But yes, you are correct, TRANSPORT errors can be ignored, they are typically transient.

[johnrgregg3]

OK. I'm running on Ubuntu. I used to log the string, but I seem to recall that sometimes it would hit me with an exception, so I stopped doing that. I'll put that back and see if I can reproduce it. If so, I'll post here. Thanks!

[davidblewett]

But yes, you are correct, TRANSPORT errors can be ignored, they are typically transient.

I would appreciate updating the documentation as to what exceptions are transient and what are permanent. Even better, it would be very nice to encode that in the exception class hierarchy, so that people can have an exception handler for say KafkaTransientException that just logs it and moves on, and one for KafkaPermanentException that logs and triggers graceful shutdown.

[edenhill]

That's a good idea, @davidblewett . We'll look into it.