Introduction
From v0.12.0, the skipset requires Go version >= 1.18, if your Go version is older, use v0.11.0 instead.
Skipset is a high-performance, scalable, concurrent-safe set based on the skip-list. In the typical pattern (100000 operations, 90%CONTAINS 9%ADD 1%REMOVE, 8C16T), the skipset is up to 15x faster than the built-in sync.Map
.
The main idea behind the skipset is A Simple Optimistic Skiplist Algorithm.
Different from the sync.Map
, the items in the skipset are always sorted, and the Contains
and Range
operations are wait-free (A goroutine is guaranteed to complete an operation as long as it keeps taking steps, regardless of the activity of other goroutines).
The skipset is a set instead of a map, if you need a high-performance full replacement of sync.Map
, see skipmap.
Features
- Scalable, high-performance, concurrent-safe.
- Wait-free
Contains
andRange
operations (wait-free algorithms have stronger guarantees than lock-free). - Sorted items.
When should you use skipset
In most cases, skipset
is better than sync.Map
, especially in these situations:
- Concurrent calls of multiple operations. Such as using both
Range
andAdd
at the same time. In this situation, using skipset can greatly improve the performance. - Memory intensive. The skipset saves at least 50% of the memory in the benchmark.
If only one goroutine accesses the set for most of the time, such to insert a batch of elements and then uses only Contains
or Range
, using built-in map is better.
QuickStart
See Go doc for more information.
package main
import (
"fmt"
"github.com/zhangyunhao116/skipset"
)
func main() {
l := skipset.NewInt()
for _, v := range []int{10, 12, 15} {
if l.Add(v) {
fmt.Println("skipset add", v)
}
}
if l.Contains(10) {
fmt.Println("skipset contains 10")
}
l.Range(func(value int) bool {
fmt.Println("skipset range found ", value)
return true
})
l.Remove(15)
fmt.Printf("skipset contains %d items\r\n", l.Len())
}
From v0.12.0
on, you can use a generic version of APIs.
Note that the generic APIs are always slower than typed APIs, but are more suitable for some scenarios such as functional programming.
e.g.
New[int]
is ~2x slower thanNewInt
, andNewFunc(func(a, b int) bool { return a < b })
is 1~2x slower thanNew[int]
.Performance ranking: NewInt > New[Int] > NewFunc(func(a, b int) bool { return a < b })
package main
import (
"math"
"github.com/zhangyunhao116/skipset"
)
func main() {
x1 := skipset.New[int]()
for _, v := range []int{2, 1, 3} {
x1.Add(v)
}
x1.Range(func(value int) bool {
println(value)
return true
})
x2 := skipset.NewFunc(func(a, b float64) bool {
return a < b || (math.IsNaN(a) && !math.IsNaN(b))
})
for _, v := range []float64{math.NaN(), 3, 1, math.NaN(), 2} {
x2.Add(v)
}
x2.Range(func(value float64) bool {
println(value)
return true
})
}
Benchmark
Based on typed APIs.
Go version: go1.16.2 linux/amd64
CPU: AMD 3700x(8C16T), running at 3.6GHz
OS: ubuntu 18.04
MEMORY: 16G x 2 (3200MHz)
name time/op
Int64/Add/skipset-16 86.6ns ±11%
Int64/Add/sync.Map-16 674ns ± 6%
Int64/Contains50Hits/skipset-16 9.85ns ±12%
Int64/Contains50Hits/sync.Map-16 14.7ns ±30%
Int64/30Add70Contains/skipset-16 38.8ns ±18%
Int64/30Add70Contains/sync.Map-16 586ns ± 5%
Int64/1Remove9Add90Contains/skipset-16 24.9ns ±17%
Int64/1Remove9Add90Contains/sync.Map-16 493ns ± 5%
Int64/1Range9Remove90Add900Contains/skipset-16 25.9ns ±16%
Int64/1Range9Remove90Add900Contains/sync.Map-16 1.00µs ±12%
String/Add/skipset-16 130ns ±14%
String/Add/sync.Map-16 878ns ± 4%
String/Contains50Hits/skipset-16 18.3ns ± 9%
String/Contains50Hits/sync.Map-16 19.2ns ±18%
String/30Add70Contains/skipset-16 61.0ns ±15%
String/30Add70Contains/sync.Map-16 756ns ± 7%
String/1Remove9Add90Contains/skipset-16 31.3ns ±13%
String/1Remove9Add90Contains/sync.Map-16 614ns ± 6%
String/1Range9Remove90Add900Contains/skipset-16 36.2ns ±18%
String/1Range9Remove90Add900Contains/sync.Map-16 1.20µs ±17%
name alloc/op
Int64/Add/skipset-16 65.0B ± 0%
Int64/Add/sync.Map-16 128B ± 1%
Int64/Contains50Hits/skipset-16 0.00B
Int64/Contains50Hits/sync.Map-16 0.00B
Int64/30Add70Contains/skipset-16 19.0B ± 0%
Int64/30Add70Contains/sync.Map-16 77.7B ±16%
Int64/1Remove9Add90Contains/skipset-16 5.00B ± 0%
Int64/1Remove9Add90Contains/sync.Map-16 57.5B ± 4%
Int64/1Range9Remove90Add900Contains/skipset-16 5.00B ± 0%
Int64/1Range9Remove90Add900Contains/sync.Map-16 255B ±22%
String/Add/skipset-16 97.0B ± 0%
String/Add/sync.Map-16 152B ± 0%
String/Contains50Hits/skipset-16 15.0B ± 0%
String/Contains50Hits/sync.Map-16 15.0B ± 0%
String/30Add70Contains/skipset-16 40.0B ± 0%
String/30Add70Contains/sync.Map-16 98.2B ±11%
String/1Remove9Add90Contains/skipset-16 23.0B ± 0%
String/1Remove9Add90Contains/sync.Map-16 73.9B ± 4%
String/1Range9Remove90Add900Contains/skipset-16 23.0B ± 0%
String/1Range9Remove90Add900Contains/sync.Map-16 261B ±18%
name allocs/op
Int64/Add/skipset-16 1.00 ± 0%
Int64/Add/sync.Map-16 4.00 ± 0%
Int64/Contains50Hits/skipset-16 0.00
Int64/Contains50Hits/sync.Map-16 0.00
Int64/30Add70Contains/skipset-16 0.00
Int64/30Add70Contains/sync.Map-16 1.00 ± 0%
Int64/1Remove9Add90Contains/skipset-16 0.00
Int64/1Remove9Add90Contains/sync.Map-16 0.00
Int64/1Range9Remove90Add900Contains/skipset-16 0.00
Int64/1Range9Remove90Add900Contains/sync.Map-16 0.00
String/Add/skipset-16 2.00 ± 0%
String/Add/sync.Map-16 5.00 ± 0%
String/Contains50Hits/skipset-16 1.00 ± 0%
String/Contains50Hits/sync.Map-16 1.00 ± 0%
String/30Add70Contains/skipset-16 1.00 ± 0%
String/30Add70Contains/sync.Map-16 2.00 ± 0%
String/1Remove9Add90Contains/skipset-16 1.00 ± 0%
String/1Remove9Add90Contains/sync.Map-16 1.00 ± 0%
String/1Range9Remove90Add900Contains/skipset-16 1.00 ± 0%
String/1Range9Remove90Add900Contains/sync.Map-16 1.00 ± 0%