Odysseas Georgoudis

Hey thanks for the ideas! it can be easily implemented as a stand alone sink It’s something i was considering to add 1-2 months ago. I put it on the...

I've opened a draft PR; feel free to take a look and let me know your thoughts. ### Other things I tried: - Tried different block sizes and chunks sizes...

The benchmark numbers above were from before the latest commits on master. With recent improvements I introduced in the backend worker, it seems that the sync IO performance has now...

These are the numbers i am currently getting after syncing this PR with master branch Without `io_uring` ``` $ /home/odygrd/git/quill/cmake-build-release/benchmarks/backend_throughput/BENCHMARK_quill_backend_throughput Throughput is 5.50 million msgs/sec average, total time elapsed: 727...

Thanks for pointing this out. I hadn't noticed the additional io_uring thread. I assumed that io_uring operates in an OS thread not on the isolated cores. This also explains why...

hey, thanks for reporting. I can not preproduce on freebsd with gcc 13. I have also tried with clang18 but the test passes There should be no changes there compared...

There are few tests making some assumptions. This is one of those tests expecting a loop to run N times but it won’t run N times if the system is...

‘multi_frontend_threads’ and ‘user_sink’ tests might be unstable depending on the system load. The reason for the former is that we can take a timestamp and then block on a full...

I have stabilised those tests in https://github.com/odygrd/quill/pull/551 I tested on freebsd with `stress --cpu 2000 --timeout 120 &` and they are passing on my VM Will be in release `7.1.0`...

Hey, i am not sure how easy would be to do that. We just call `format` and that code runs internally by libfmt. I am not sure how easy it...