Async logging is not a performance optimization

Eduard Mishkurov — Thu, 09 Apr 2026 06:28:53 +0000

Async logging is not a performance optimization.

Most developers assume that moving logging to another thread makes it cheaper.

It doesn’t.

Logging has a cost, and that cost doesn’t disappear just because you enqueue work somewhere else.

In real systems, logging overhead typically comes from:

formatting (often the dominant part)
memory allocations and copying
synchronization between threads
actual I/O (file, console, network)

Async logging changes where these costs are paid, but not whether they exist.

If your system produces logs faster than they can be processed, async logging will not fix the problem — it will only hide it temporarily behind a queue.

Eventually, one of these happens:

the queue grows and increases memory usage
producers get blocked
logs get dropped

Another subtle issue is data lifetime.

If you defer formatting, you must ensure that all referenced data is still valid in the logging thread. That often means copying data earlier — which again shifts cost, but does not remove it.

In practice, async logging is mostly a latency optimization:
it can reduce the time spent in the calling thread.

But total work done by the system often stays the same — or even increases.

Understanding this distinction is important when designing logging systems and interpreting benchmarks.

If you're interested, I wrote a deeper breakdown with examples and real-world observations:
https://medium.com/@emishkurov/async-logging-is-not-a-silver-bullet-what-actually-limits-performance-90cd45f23c7b

DEV Community: Eduard Mishkurov

Async logging is not a performance optimization