Andy Pan

The benchmark for commands of retrieving data with `pipeline` could call `aeApiAddEvent` for `AE_WRITABLE`. Posting these numbers for another reference. ```bash ./valkey-benchmark -r 10000000 -d 1024 -n 10000000 -P 100...

> Ok, so I'm still worried about the panic and how it might impact production systems for edge cases related to kqueue failures. I would like to get the performance...

Another approach could be Makefile flags, something like `BUILD_TLS` or `USE_SYSTEMD`. This looks more feasible.

@madolson Please check out the latest commit.

> @panjf2000, do you mind counting the number of `kevent` calls before and after this change for a given workload (`./valkey-benchmark -r 10000000 -d 1024 -n 10000000 -P 100 -t...

### Benchmark command ```bash ./valkey-benchmark -r 10000000 -d 1024 -n 10000000 -P 100 -t get,lpop,rpop,spop,zpopmin -q ``` ### Branch: valkey-io:unstable ```bash CALL COUNT … accept 64 fcntl 174 setsockopt 263...

Ping @madolson @PingXie

> Once that is done, we can update the kqueue batch handler. To make it clearer, by "update the kqueue batch handler", you meant removing the `USE_KQUEUE_BATCH` macro in this...

Or, did you mean that we need to fix #528 by changing the return type in the function signatures of `connSet(Read/Write)` from `int` to `void` (option 2 in #528, it...

> > To make it clearer, by "update the kqueue batch handler", you meant removing the USE_KQUEUE_BATCH macro in this PR and make it the only code path, right? >...