logur
logur copied to clipboard
logur
Improve logger performance
Try playing with fields and disabled levels in Zap and Hclog
Original benchmark results
goos: darwin goarch: amd64 pkg: github.com/goph/logur/benchmarks BenchmarkDisabledWithoutFields/zap-12 100000000 12.9 ns/op 32 B/op 2 allocs/op BenchmarkDisabledWithoutFields/hclog-12 50000000 33.1 ns/op 96 B/op 3 allocs/op BenchmarkDisabledWithoutFields/zerolog-12 20000000 61.9 ns/op 96 B/op 3 allocs/op BenchmarkDisabledWithoutFields/kitlog-12 20000000 113 ns/op 288 B/op 8 allocs/op BenchmarkDisabledWithoutFields/logrus-12 100000000 13.9 ns/op 32 B/op 2 allocs/op BenchmarkDisabledAccumulatedContext/logrus-12 100000000 13.7 ns/op 32 B/op 2 allocs/op BenchmarkDisabledAccumulatedContext/zap-12 100000000 16.0 ns/op 32 B/op 2 allocs/op BenchmarkDisabledAccumulatedContext/hclog-12 30000000 39.2 ns/op 96 B/op 3 allocs/op BenchmarkDisabledAccumulatedContext/zerolog-12 20000000 75.1 ns/op 96 B/op 3 allocs/op BenchmarkDisabledAccumulatedContext/kitlog-12 5000000 261 ns/op 1249 B/op 8 allocs/op BenchmarkDisabledAddingFields/logrus-12 3000000 487 ns/op 1580 B/op 15 allocs/op BenchmarkDisabledAddingFields/zap-12 300000 4693 ns/op 8742 B/op 73 allocs/op BenchmarkDisabledAddingFields/hclog-12 1000000 1184 ns/op 3507 B/op 40 allocs/op BenchmarkDisabledAddingFields/zerolog-12 300000 4690 ns/op 7707 B/op 58 allocs/op BenchmarkDisabledAddingFields/kitlog-12 2000000 851 ns/op 2946 B/op 32 allocs/op BenchmarkWithoutFields/zap-12 10000000 207 ns/op 96 B/op 3 allocs/op BenchmarkWithoutFields/hclog-12 2000000 686 ns/op 128 B/op 4 allocs/op BenchmarkWithoutFields/zerolog-12 20000000 71.1 ns/op 96 B/op 3 allocs/op BenchmarkWithoutFields/kitlog-12 3000000 538 ns/op 1058 B/op 21 allocs/op BenchmarkWithoutFields/logrus-12 500000 2936 ns/op 513 B/op 14 allocs/op BenchmarkAccumulatedContext/logrus-12 50000 27717 ns/op 4663 B/op 92 allocs/op BenchmarkAccumulatedContext/zap-12 10000000 214 ns/op 96 B/op 3 allocs/op BenchmarkAccumulatedContext/hclog-12 50000 26948 ns/op 8089 B/op 123 allocs/op BenchmarkAccumulatedContext/zerolog-12 20000000 78.9 ns/op 96 B/op 3 allocs/op BenchmarkAccumulatedContext/kitlog-12 200000 6090 ns/op 6645 B/op 110 allocs/op BenchmarkAddingFields/logrus-12 50000 30147 ns/op 6245 B/op 106 allocs/op BenchmarkAddingFields/zap-12 300000 5333 ns/op 9058 B/op 74 allocs/op BenchmarkAddingFields/hclog-12 50000 31727 ns/op 11503 B/op 160 allocs/op BenchmarkAddingFields/zerolog-12 300000 5021 ns/op 7707 B/op 58 allocs/op BenchmarkAddingFields/kitlog-12 200000 6988 ns/op 8374 B/op 134 allocs/op PASS ok github.com/goph/logur/benchmarks 50.774s
Benchmark results after single message argument
goos: darwin goarch: amd64 pkg: github.com/goph/logur/benchmarks BenchmarkDisabledWithoutFields/logrus-12 200000000 6.94 ns/op 16 B/op 1 allocs/op BenchmarkDisabledWithoutFields/zap-12 200000000 8.82 ns/op 16 B/op 1 allocs/op BenchmarkDisabledWithoutFields/hclog-12 2000000000 1.69 ns/op 0 B/op 0 allocs/op BenchmarkDisabledWithoutFields/zerolog-12 50000000 26.1 ns/op 0 B/op 0 allocs/op BenchmarkDisabledWithoutFields/kitlog-12 20000000 76.5 ns/op 192 B/op 5 allocs/op BenchmarkDisabledAccumulatedContext/logrus-12 200000000 8.26 ns/op 16 B/op 1 allocs/op BenchmarkDisabledAccumulatedContext/zap-12 100000000 10.3 ns/op 16 B/op 1 allocs/op BenchmarkDisabledAccumulatedContext/hclog-12 2000000000 1.77 ns/op 0 B/op 0 allocs/op BenchmarkDisabledAccumulatedContext/zerolog-12 50000000 32.6 ns/op 0 B/op 0 allocs/op BenchmarkDisabledAccumulatedContext/kitlog-12 5000000 236 ns/op 1152 B/op 5 allocs/op BenchmarkDisabledAddingFields/logrus-12 3000000 512 ns/op 1564 B/op 14 allocs/op BenchmarkDisabledAddingFields/zap-12 300000 5107 ns/op 8728 B/op 72 allocs/op BenchmarkDisabledAddingFields/hclog-12 1000000 1215 ns/op 3390 B/op 36 allocs/op BenchmarkDisabledAddingFields/zerolog-12 300000 4896 ns/op 7586 B/op 54 allocs/op BenchmarkDisabledAddingFields/kitlog-12 2000000 854 ns/op 2830 B/op 28 allocs/op BenchmarkWithoutFields/logrus-12 500000 2981 ns/op 497 B/op 13 allocs/op BenchmarkWithoutFields/zap-12 10000000 200 ns/op 80 B/op 2 allocs/op BenchmarkWithoutFields/hclog-12 3000000 507 ns/op 32 B/op 1 allocs/op BenchmarkWithoutFields/zerolog-12 100000000 26.2 ns/op 0 B/op 0 allocs/op BenchmarkWithoutFields/kitlog-12 3000000 501 ns/op 961 B/op 18 allocs/op BenchmarkAccumulatedContext/kitlog-12 200000 6371 ns/op 6534 B/op 107 allocs/op BenchmarkAccumulatedContext/logrus-12 50000 28836 ns/op 4646 B/op 91 allocs/op BenchmarkAccumulatedContext/zap-12 10000000 206 ns/op 80 B/op 2 allocs/op BenchmarkAccumulatedContext/hclog-12 50000 27934 ns/op 7993 B/op 120 allocs/op BenchmarkAccumulatedContext/zerolog-12 50000000 31.1 ns/op 0 B/op 0 allocs/op BenchmarkAddingFields/zerolog-12 300000 4693 ns/op 7587 B/op 54 allocs/op BenchmarkAddingFields/kitlog-12 200000 7123 ns/op 8243 B/op 130 allocs/op BenchmarkAddingFields/logrus-12 50000 31767 ns/op 6230 B/op 105 allocs/op BenchmarkAddingFields/zap-12 300000 5263 ns/op 9048 B/op 73 allocs/op BenchmarkAddingFields/hclog-12 50000 32059 ns/op 11391 B/op 156 allocs/op PASS ok github.com/goph/logur/benchmarks 58.244s
Wondering what's the status on this?
"In a nutshell", what's the significance of the performance impact relative to direct usage of any one the abstracted logger libs (or specifically Zap)? And is the performance impact on resource efficiency or operational latency?
Finally bumped into the same problem where I'm unable to create a one-size-fits-all abstraction for logging. Unfortunately most of our logging will be concentrated in the hot path, so as much as I'd like to use this wrapper lib, if the performance is non-negligible, we'd be forced to completely buy into zap.
Most (contextual) logging libraries fall into two categories:
- Ones that accept context as
map[string]interface{} - Ones that accept context as variadic
...interface{}
This abstraction uses map[string]interface{} (see the reasons in the readme), so in case of loggers that accept context as variadic (like zap) the context needs to be transformed to that format. This transformation requires extra allocations, so the "performance hit" is totally up to the size of the context.
You can see some benchmark results in the PR description above. It is actually based on the benchmark suite of zap, so you can directly compare the two. You will find that Logur with a Zap adapter is 5-8 times "slower" (when using a large context) than just using Zap, however we are talking about that magnitude in nanoseconds.
Personally I think that this "performance hit" is mostly negligible, especially if your context does not contain a lot of fields, but you should probably make that decision based on the provided benchmarking data and your use case.
I have plans to add a KVLogger interface implementing the variadic key-value style, which should make Zap-like loggers "faster". Then everyone can decide which style they want to use.
Hope that helps. Let me know what you think or if you decide to use/not use Logur.
