Blazing fast, structured, leveled logging in Go.
go get -u go.uber.org/zap
In contexts where performance is nice, but not critical, use the
SugaredLogger
. It's 4-10x faster than than other structured logging libraries
and includes both structured and printf
-style APIs.
logger, _ := NewProduction()
sugar := logger.Sugar()
sugar.Infow("Failed to fetch URL.",
// Structured context as loosely-typed key-value pairs.
"url", url,
"attempt", retryNum,
"backoff", time.Second,
)
sugar.Infof("Failed to fetch URL: %s", url)
When performance and type safety are critical, use the Logger
. It's even faster than
the SugaredLogger
and allocates far less, but it only supports structured logging.
logger, _ := NewProduction()
logger.Info("Failed to fetch URL.",
// Structured context as strongly-typed Field values.
zap.String("url", url),
zap.Int("attempt", tryNum),
zap.Duration("backoff", time.Second),
)
For applications that log in the hot path, reflection-based serialization and
string formatting are prohibitively expensive — they're CPU-intensive and
make many small allocations. Put differently, using encoding/json
and
fmt.Fprintf
to log tons of interface{}
s makes your application slow.
Zap takes a different approach. It includes a reflection-free, zero-allocation
JSON encoder, and the base Logger
strives to avoid serialization overhead and
allocations wherever possible. By building the high-level SugaredLogger
on
that foundation, zap lets users choose when they need to count every
allocation and when they'd prefer a more familiar, loosely-typed API.
As measured by its own benchmarking suite, not only is zap more performant than comparable structured logging libraries — it's also faster than the standard library. Like all benchmarks, take these with a grain of salt.1
Log a message and 10 fields:
Library | Time | Bytes Allocated | Objects Allocated |
---|---|---|---|
⚡ zap | 1436 ns/op | 705 B/op | 2 allocs/op |
⚡ zap (sugared) | 2436 ns/op | 1931 B/op | 21 allocs/op |
logrus | 9393 ns/op | 5783 B/op | 77 allocs/op |
go-kit | 6929 ns/op | 3119 B/op | 65 allocs/op |
log15 | 25004 ns/op | 5535 B/op | 91 allocs/op |
apex/log | 18450 ns/op | 4025 B/op | 64 allocs/op |
Log a message with a logger that already has 10 fields of context:
Library | Time | Bytes Allocated | Objects Allocated |
---|---|---|---|
⚡ zap | 368 ns/op | 0 B/op | 0 allocs/op |
⚡ zap (sugared) | 388 ns/op | 0 B/op | 0 allocs/op |
logrus | 8420 ns/op | 3967 B/op | 61 allocs/op |
go-kit | 7288 ns/op | 2950 B/op | 50 allocs/op |
log15 | 17678 ns/op | 2546 B/op | 42 allocs/op |
apex/log | 16126 ns/op | 2801 B/op | 49 allocs/op |
Log a static string, without any context or printf
-style templating:
Library | Time | Bytes Allocated | Objects Allocated |
---|---|---|---|
⚡ zap | 398 ns/op | 0 B/op | 0 allocs/op |
⚡ zap (sugared) | 400 ns/op | 80 B/op | 2 allocs/op |
standard library | 678 ns/op | 80 B/op | 2 allocs/op |
logrus | 2778 ns/op | 1409 B/op | 25 allocs/op |
go-kit | 1318 ns/op | 656 B/op | 13 allocs/op |
log15 | 5720 ns/op | 1496 B/op | 24 allocs/op |
apex/log | 3282 ns/op | 584 B/op | 11 allocs/op |
The current release is v1.0.0-rc.2
. No further breaking changes are planned
unless wider use reveals critical bugs or usability issues, but users who need
absolute stability should wait for the 1.0.0 release.
Released under the [MIT License](LICENSE.txt).
1 In particular, keep in mind that we may be benchmarking against slightly older versions of other libraries. Versions are pinned in zap's glide.lock file. ↩