Skip to content

Performance

Benchmarks are JMH runs from make benchmark (JDK 24, AverageTime, 1 thread, @Fork(0), 2×1s warmup + 5×1s measurement). Each operation logs 1,000 INFO messages with {} placeholders; SLF4J/Logback writes to /dev/null.

Path log4error Plain SLF4J
Happy path (no error) 5,431 ns per 1,000 lines (~5.4 ns/line) 1,373,980 ns (~1.4 µs/line)
Error path (flush 1,000 + error) 1,357,384 ns 1,379,847 ns
  • Happy path: buffering is about 250× cheaper than writing each line through SLF4J. Appending to a list rarely needs a syscall; writing a log line always does.
  • Error path: about the same cost as logging everything up front. You pay for I/O only when a request fails.

The benchmark backend writes to /dev/null. With real appenders (files, network shippers, JSON encoding), writing costs more, so the happy-path saving is larger.

Raw JMH output

Benchmark                           Mode  Cnt        Score        Error  Units
Log4ErrorBenchmark.errorPathFlush   avgt    5  1357383.947 ± 133696.465  ns/op
Log4ErrorBenchmark.happyPathBuffer  avgt    5     5430.910 ±    455.652  ns/op
SLF4jBenchmark.infoWrite            avgt    5  1373980.078 ±  61640.944  ns/op
SLF4jBenchmark.infoWriteThenError   avgt    5  1379847.224 ±  98604.677  ns/op

Re-run locally with make benchmark. The benchmark sources are in log4error/src/test/java/io/github/parvez3019/benchmarking.

Credit to Christian Hujer for noting that buffering improves the happy path.