Entry Filter performance overhead

[klevy-toast]

Hello,

I've been experimenting with Entry Filters as a means of delivering messages only to the subscription specified by a custom header. In my testing, I have observed pretty negligible performance differences on the broker side, but surprisingly see that the bookkeeper CPU is higher when the filter is active. Since the filter runs on the brokers, I'm having trouble understanding why that could be the case. Does anybody have ideas?

[klevy-toast]

I realized that the entry filter implementation always results in individual acknowledgement whereas consumers use group acknowledgement. In my test I am generating high throughput and filtering (skipping) every message for 7 out of 8 consumers. I think it does make sense that this implementation would result in more cursor operations and therefore higher bookkeeper CPU. Can anybody confirm this hypothesis?