flexi_logger Performance

[rohitjoshi]

I switch from slog_async to flexi_logger and lost ~20% throughput. The log level is set to info and less than 10 lines are logged per second. Any reason for this performance drop?

I suspect few things but please help me here.

1. For slog, I used features to remove debug statements at compile time:

   slog = { version = "2.5.2", features = ["max_level_trace", "release_max_level_info"] }

   but this is not supported by flexi_logger. I did use log level features as below.

   
   log = { version = "0.4.11", features = ["max_level_trace", "release_max_level_info"] }

2. slog uses async sink with channel size as ~64K but don't see similar config for flexi_logger. Any way to use async / buffer logging using flexi_logger?

[emabee]

Ad 1: Compile time filters are provided by the log crate into which flexi_logger 'plugs' in. See https://docs.rs/log/0.4.11/log/#compile-time-filters .

Ad 2: I added an option use_buffering (see https://docs.rs/flexi_logger/0.17.0/flexi_logger/struct.Logger.html#method.use_buffering) with version 0.17; would be great if you could try it out and let me know if this helps already far enough. I also started to experiment with crossbeam channels, but didn't see a big enough difference yet (but this may be platform specific).

[rohitjoshi]

I will try it out and let you know. Thank you 🙏

[rohitjoshi]

With buffering, it would be good to implement auto flush timer so buffered output will be written to log. Eg. Nginx allows configurable buffer size and flush interval http://nginx.org/en/docs/stream/ngx_stream_log_module.html#access_log

[emabee]

Is implemented with 0.17.1; see https://docs.rs/flexi_logger/0.17.1/flexi_logger/struct.Logger.html#method.buffer_and_flush or https://docs.rs/flexi_logger/0.17.1/flexi_logger/struct.Logger.html#method.buffer_and_flush_with

Thanks for the proposal!!