We use librdkafka mainly for Producing at a high msg rate like 30K/50K per second. If I enable librdkafka statistics by specifying a value for statistics.interval.ms, I am worried that it will impact producer performance.
However if the performance impact occurs only when firing the statistics callback and it's impact is negligible on msg producing or acknowledgment reception (I understand that even these actions will include minor performance impacting actions like calculating the latency, updating a sample count and total latency etc.), then I have the option of using a large value forstatistics.interval.ms. Specially I am worried about latency calculations like 99th percentile.
Of course this needs to be tested and compare performance, BUT
If anyone has past experience related to this
OR
someone familiar with librdkafka source code
could please answer/share your experience on this ?