Maybe mark it abandoned? fwiw, I think both KIPs break the worker-vs-connector boundary that, for better or worse, is fundamental to Connect's architecture and APIs.
An alternative came up a few times during discussions for KIP-382: support multiple "herders" within the same Connect cluster. Each herder would use a separate worker config, and each would be addressable through the REST API somehow. For example, perhaps you could POST a connector to a specific herder, which would run the connector using that herder's worker config. In most cases, you'd be fine with a single default herder, in which case nothing would change from today's experience. But when necessary an operator could add a new herder without creating an entirely new Connect cluster. For example, you might have one herder with compression enabled and one without. Ryanne On Fri, Mar 15, 2019 at 8:55 AM Randall Hauch <rha...@gmail.com> wrote: > > https://cwiki.apache.org/confluence/display/KAFKA/KIP-407%3A+Kafka+Connect+support+override+worker+kafka+api+configuration+with+connector+configuration+that+post+by+rest+api > was created by the author and a discussion thread was never started. > > This appears to very similar to (or rather a subset of) > https://issues.apache.org/jira/browse/KAFKA-6890 / > > https://cwiki.apache.org/confluence/display/KAFKA/KIP-296%3A+Connector+level+configurability+for+client+configs > , > which IMO has the correct scope and where a discussion is taking place > about the requirements and user experience. This proposal seems to differ > from KAFKA-6890 / KIP-296 is that the approach proposed here only addresses > connector configurations specified via the REST API, and not via > configuration files passed to the standalone Connect worker. This would be > a significant departure from the current behavior, where the REST API and > file configurations are completely compatible. > > It's fine to leave this open for further discussion, but I would prefer to > withdraw this KIP and instead focus the discussion on KIP-296. > > Thoughts? > > Randall >
