C0urante commented on pull request #10530:
URL: https://github.com/apache/kafka/pull/10530#issuecomment-824847306


   I've just realized--this may not be a safe change to make. Although we 
should fail fast on startup if a worker tries to join an existing cluster with 
a bad hostname, it's still possible right now to run a single worker in that 
scenario. Although running a single worker in distributed mode may sound like 
an anti-pattern, there are still benefits to such a setup, such as persistent 
storage of configurations in a Kafka topic. It's also possible that there are 
Dockerized quickstarts and demos that use a single Connect worker and may have 
a bad hostname; these would suddenly break on upgrade if we merge this change.
   
   At the very least, I think we should be more explicit about _why_ we're 
failing workers on startup with our error message, try to give users a better 
picture of how workers with a bad hostname can be dangerous (since right now 
they don't fail on startup but instead begin to fail silently when forwarding 
user requests or task configs to the leader), and even call out Dockerized 
setups with instructions on how to fix the worker config in that case by 
changing the advertised URL to use a valid hostname and, if running a 
multi-node cluster, making sure that the worker is reachable from other workers 
with that advertised URL.
   
   But I think at this point we're doing a fair bit of work to try to 
circumvent this issue instead of addressing it head-on. Using a different HTTP 
client has already been discussed as an option but punted on in favor of a 
smaller, simpler change; given that this change seems less simple now, I wonder 
if it's worth reconsidering.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


Reply via email to