sujitn commented on PR #3038:
URL: https://github.com/apache/iggy/pull/3038#issuecomment-4144899997

   Agreed on the shard 0 bottleneck that's the real architectural issue. Fault 
isolation is also a fair point given the size of the Kafka protocol surface.
   
   the security/correctness gaps (auto-auth, no CRC validation, dropped record 
keys) are implementation maturity issues, not in-process vs gateway issues. The 
gateway will have the same Phase 1 limitations until they're addressed.
   
   Questions on the gateway approach before I open a Discussion:
   
   - send_messages() returning base_offset is a hard blocker without it a Kafka 
ProduceResponse is invalid regardless of architecture. Is there an existing 
issue for this, or should I propose the binary protocol change? Happy to 
contribute that work.
   
   - For ListOffsets I need earliest/latest offset per partition. Is that 
exposed through the SDK today, or would it require a new SDK method?
   
   - The connector runtime comparison makes sense for lifecycle isolation, but 
the interaction pattern is different connectors are async pipelines, a Kafka 
gateway is synchronous request-response on the same TCP connection. I'll look 
at the runtime structure but the internal design will diverge.
   
   The protocol codec from this spike (varint, RecordBatch v2, request/response 
types, flexible header negotiation) transfers directly to the gateway.  I'll 
close this PR and open a Discussion with the gateway design.
   
   Point taken on discussions-before-code will do that going forward.


-- 
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.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to