Does Confluent Rest Proxy API (producer) retry event publish on failure

Question

We are planning to use Confluent Rest proxy as the platform layer for listening to all user events (& publishing them to Kafka). Working on micro-services model & having varied types of event generators, we want our APIs/event-generators to be decoupled from event listening/handling. Also, at event listening layer, event resilience is important for us.

1. From what I understand, if the Rest proxy layer fails to publish to Kafka(once) for any reason, it doesn't retry. This functionality needs to be handled by the caller (the client layer), which needs to make synchronous calls & retry on failure. However, couldn't find any details on this, in the product documentation. Could someone please confirm the same?
2. Confluent Rest Proxy developers claim that with the right Rest Proxy cluster set-up & right request batching by the client, performance as good as native producers' can be achieved. Any exceptions/(positive/negative)thoughts here?
3. Calls to the Rest Proxy Producer API are blocking. If the client doesn't need to know the partition & offset details, can we configure these calls to be non-blocking in anyway, such that once the request is received, resilience is managed by the Rest Proxy layer itself. The client just receives a 200 HTTP status as acknowledgement, whenever a produce msg request is received.

Answer 1

1. The REST Proxy is just a normal Kafka Producer and Kafka Consumer and can be configured with or without retries enabled, just as any other Kafka Producer can.
2. A single producer publishing via a REST Proxy will not achieve the same throughput as a single native Java Producer. However you can scale up many REST proxies and many HTTP producers to get high performance in aggregate. You can also mitigate the performance penalty imposed by HTTP by batching multiple messages together into a consolidated HTTP request to minimize the number of HTTP round trips on the wire.