Compatibility with old glibc

[brendene]

Hello,

Thanks again for the great software. On older systems with glibc < 2.17, std::chrono::steady_clock::now() is implemented as a syscall in libstdc++ instead of using the faster VDSO (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59177)

With the default logger config, the logging thread consumes close to 100% cpu due to the syscalls in an almost tight loop. Setting backend_thread_strict_log_timestamp_order=false will stop the clock_gettime syscalls in the logging thread or setting backend_thread_sleep_duration will reduce the frequency of calls.

FWIW, I came across this when compiling quill in an environment created with conda forge, the version of glibc that their libstdc++ library uses is old, perhaps for compatibility reasons.

I am not sure if its the responsibility of your program to handle these types but that I would point out that they exist and are not uncommon.

[odygrd]

hey, thanks for the post :) I haven't tried it in such old systems.
Yes even by design the backend logging thread spins instead of waking up via a condition variable for example as it is more performant for the hot threads. The logging thread can be pinned to a non-critical core to reduce the effect.

As you said setting backend_thread_strict_log_timestamp_order=false will greatly reduce calls to std::chrono::steady_clock::now() I think that is the main one

You can also set (rdtsc_resync_interval) value a bit higher like 900 ms (https://github.com/odygrd/quill/blob/master/quill/include/quill/Config.h#L178) although it should make only minor difference