- Arishem是由字节跳动客服平台架构组自研的一款轻量、高性能 的DSL规则引擎。目的是将变更频繁的业务决策从应用程序中剥离出来,使用可视化界面灵活地编写业务决策,提升业务需求的响应速度。
- Arishem采用完全兼容的JSON语法格式来定义规则语法,通过组装、嵌套 的方式可以灵活地表达业务规则,使用Arishem,可以很容易地使用页面将规则进行可视化,使得不具备编程基础的人员也能快速上手。
- Arishem内部从AST解析生成和规则执行都进行了一系列的优化,使得单个复杂规则的执行可在µs级别的时间内完成。
- Arishem支持可自定义的规则执行顺序和并发执行粒度,支持运行时下游数据的并发和预测获取。
- Arishem内部集成了丰富的操作符和内置函数。
Arishem由条件表达式和目的表达式构成,其最大的特点就是其语法完全兼容JSON语法,也能完全兼容IDL(如thrift、protobuf等),在examples部分定义了thrift和protobuf进行规则定义的样例。
arishem在规则应用到可视化场景(如规则配置页面)具有非常显著的优势,在字节跳动的客服平台,有关规则判断的几十个场景都接入了arishem,并实现了产运的可视化规则配置页面,使得规则配置这种原本需要研发去维护的场景,变成产运同学也能进行配置维护,大大解放了研发资源。
在benchmark下执行单个无网络请求的复杂规则的耗时仅在微秒级别!当然没有最好性能的框架,只有最适合的应用场景。
测试环境:
goos: darwin
goarch: amd64
pkg: */arishem/arishem
cpu: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
PASS
测试条件样例,共495个AST节点,包含基本的逻辑操作,包括字符串正则、数组遍历、数组交集等:
// 实时数据
{"username":"Andrew","usernames":["Jack","Mike","Andrew"],"news":"Jack hanged out with Mike last weekend.","number1":100,"numbers":["10",99.9,0]}
// 条件表达式
{"OpLogic":"||","ConditionGroups":[{"OpLogic":"&&","Conditions":[{"Operator":"STRING_START_WITH","Lhs":{"VarExpr":"username"},"Rhs":{"Const":{"StrConst":"Banana"}}},{"Operator":"STRING_END_WITH","Lhs":{"VarExpr":"usernames#1"},"Rhs":{"Const":{"StrConst":"A"}}},{"Operator":"STRING_END_WITH","Lhs":{"VarExpr":"usernames#0"},"Rhs":{"Const":{"StrConst":"hahaha"}}},{"Operator":"CONTAIN_REGULAR","Lhs":{"VarExpr":"usernames#0"},"Rhs":{"Const":{"StrConst":"^M.*"}}}]},{"OpLogic":"and","ConditionGroups":[{"OpLogic":"&&","ConditionGroups":[{"OpLogic":"||","Conditions":[{"Operator":"<=","Lhs":{"Const":{"NumConst":100}},"Rhs":{"Const":{"NumConst":10}}},{"Operator":"<=","Lhs":{"Const":{"NumConst":100}},"Rhs":{"Const":{"NumConst":10}}},{"Operator":"<=","Lhs":{"Const":{"NumConst":100}},"Rhs":{"Const":{"NumConst":10}}},{"Operator":"<=","Lhs":{"Const":{"NumConst":100}},"Rhs":{"Const":{"NumConst":10}}},{"Operator":"<","Lhs":{"Const":{"NumConst":100}},"Rhs":{"Const":{"NumConst":10}}}]}]},{"OpLogic":"&&","Conditions":[{"Operator":"LIST_IN","Lhs":{"VarExpr":"number1"},"Rhs":{"ConstList":[{"NumConst":1},{"NumConst":98},{"NumConst":101},{"NumConst":1.32e-3},{"NumConst":12.234},{"NumConst":-1}]}},{"Operator":"!LIST_IN","Lhs":{"VarExpr":"numbers#0"},"Rhs":{"ConstList":[{"NumConst":1},{"NumConst":99},{"NumConst":999},{"NumConst":1.32e-3},{"NumConst":10},{"NumConst":-1}]}},{"Operator":"LIST_CONTAINS","Lhs":{"VarExpr":"numbers"},"Rhs":{"MathExpr":{"OpMath":"+","ParamList":[{"Const":{"NumConst":6}},{"Const":{"StrConst":"5"}}]}}},{"Operator":"LIST_RETAIN","Lhs":{"VarExpr":"numbers"},"Rhs":{"MathExpr":{"OpMath":"+","ParamList":[{"Const":{"NumConst":6}},{"Const":{"StrConst":"5"}}]}}},{"Operator":"LIST_RETAIN","Lhs":{"VarExpr":"numbers"},"Rhs":{"ConstList":[{"StrConst":"-1"},{"BoolConst":true},{"NumConst":-3.1415926}]}},{"Operator":"!LIST_RETAIN","Lhs":{"VarExpr":"numbers"},"Rhs":{"ConstList":[{"StrConst":"-1"},{"BoolConst":true},{"NumConst":-3.1415926}]}}]}]},{"OpLogic":"&&","Conditions":[{"Operator":"==","Lhs":{"Const":{"NumConst":1}},"Rhs":{"Const":{"NumConst":1}}}]}]}
// 目的表达式
{"ActionName":"Greeting2","ParamMap":{"UserAge":{"VarExpr":"user.age"},"TempUsername":{"VarExpr":"user.name"}}}
运行结果:平均不到30µs
goos: darwin
goarch: amd64
pkg: */arishem/arishem
cpu: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
BenchmarkSingleComplexRule-12 205094 28959 ns/op 2227 B/op 69 allocs/op
BenchmarkSingleComplexRule-12 208012 28637 ns/op 2226 B/op 69 allocs/op
BenchmarkSingleComplexRule-12 207696 28614 ns/op 2226 B/op 69 allocs/op
PASS
go get github.com/bytedance/arishem
在使用前必须先调用Initialize方法,否则执行将导致arishem执行异常,一般情况下使用default配置即可,该操作应当在你的init方法中进行。
func init() {
arishem.Initialize(arishem.DefaultConfiguration())
}
- 快速判断一个条件表达式是否符合匹配的预期 arishem语法参考
func main() {
condition := `
{
"OpLogic": "&&",
"Conditions": [
{
"Operator": "==",
"Lhs": {
"Const": {
"NumConst": 1
}
},
"Rhs": {
"Const": {
"NumConst": 1
}
}
}
]
}
`
pass, err := arishem.JudgeCondition(condition)
if err != nil {
// handle error here
// ...
println(err.Error())
}
if pass {
// your business code here
// ...
println("condition passed!")
}
}
输出结果:
condition passed!
- 在规则中通过关键字VarExpr获取实时数据进行判断
func main() {
condition := `
{
"OpLogic": "&&",
"Conditions": [
{
"Operator": ">",
"Lhs": {
"VarExpr": "user.age"
},
"Rhs": {
"VarExpr": "user_ages#1"
}
}
]
}
`
pass, err := arishem.JudgeConditionWithFactMeta(condition, `
{
"user": {
"name": "KJ",
"age": 24
},
"user_ages": [
15,
20,
32
]
}
`)
if err != nil {
// handle error here
return
}
if pass {
println("KJ's age is greater than 20!")
}
}
- 创建一个规则并进行判断,输出规则目的
func main() {
condition := `
{
"OpLogic": "&&",
"Conditions": [
{
"Operator": ">=",
"Lhs": {
"VarExpr": "user.age"
},
"Rhs": {
"VarExpr": "user_ages#1"
}
}
]
}
`
// create an expression aim
aim := `
{
"Const": {
"StrConst": "rule passed!"
}
}
`
rule, err := arishem.NewNoPriorityRule("rule1", condition, aim)
if err != nil {
// handle error here
return
}
dc, err := arishem.DataContext(`
{
"user": {
"name": "KJ",
"age": 24
},
"user_ages": [
20,
18,
32
]
}
`)
if err != nil {
// handle error here
return
}
rr := arishem.ExecuteSingleRule(rule, dc)
if rr.Passed() {
fmt.Printf("%s pass, output=>%s", rr.Identifier(), rr.Aim().AsExpr())
}
}
输出结果
rule passed!
或者通过内置的builder函数来构建条件表达式和目的表达式
func main() {
condGroup := arishem.NewConditionsCondGroup(arishem.OpLogicAnd)
cond1 := arishem.NewCondition(operator.Equal)
cond1.Lhs = arishem.NewConstExpr(arishem.NewNumConst(1.0))
cond1.Rhs = arishem.NewConstExpr(arishem.NewNumConst(1.0))
condGroup.AddConditions(cond1)
expr, _ := condGroup.Build()
println(expr)
}
输出结果
{"OpLogic":"&&","Conditions":[{"Operator":"==","Lhs":{"Const":{"NumConst":1}},"Rhs":{"Const":{"NumConst":1}}}]}
更多使用方式请参考详细文档
arishem另一个强大的功能就是支持灵活多变的规则执行顺序,arishem支持优先级规则执行、非优先级规则执行以及优先级和非优先级混合执行。甚至你可以通过自实现arishem规则接口,实现动态的优先级计算。相关使用方式请参考详细文档
arishem支持多达20多种操作符,包括常用的值判断、数组判断,字符串判断等,并且arishem另一个特性就是强大的类型自动转换。 arishem在进行判断时,如果左指和右值的类型不一致,将尝试进行类型统一,在进行类型转换时,以右值的类型为标准进行转换。 更多使用方式请参考详细文档
arishem内部集成了一些常见的功能函数,包括日期、数组、map和字符串函数,使用函数表达式来使用他们。
arishem将实时获取的网络数据定义为一个feature(特征)。arishem在执行规则的时候通过分批的方式进行规则运算,那么在feature获取的时候,也是通过批次去获取的。在使用具有网络数据的场景中,使用FeatureExpr关键字来使用,并实现feature的获取方法。
{
...
"FeatureExpr": {
// user is the feature name, and username is the field path
"FeaturePath": "user.username"
}
}
type MyFeatureFetcher struct{}
...
func (m *MyFeatureFetcher) FetchFeature(feat typedef.FeatureParam, dc typedef.DataCtx) (typedef.MetaType, error) {
println("ready to fetch feature=>%s", feat.FeatureName())
// you code here
return nil, nil
}
func init() {
arishem.Initialize(
arishem.DefaultConfiguration(),
arishem.WithFeatureFetcherFactory(func() typedef.FeatureFetcher {
return &MyFeatureFetcher{}
}),
)
}
每个feature在获取时都是在一个异步协程中执行的,所以你可以没有顾虑地进行网络IO的访问,更多使用方式请参考详细文档。
arishem支持多个配置的自定义,包括规则运算时的缓存实现、批大小的计算方式以及无优先级的最大并大数量等。
在规则运算的场景下,我们小组面临最多的问题便是排查规则为什么通过/没通过?规则运算过程是否有错误?feature获取的时候具体过程是怎么样的,究竟是哪个数据没有获取到? 所以感知规则运算的具体过程是非常有必要的,这将为后续排查规则命中详情提供基础能力的支持。
arishem支持规则匹配过的条件回调和错误回调,并且FeatureFetcher也必须实现observable方法,以便让arishem内部将feature fetch的过程通知给每一个已注册观察者。
// MyObserver implements VisitObserver and FeatureFetchObserver
type MyObserver {}
func (m *MyObserver) OnFeatureFetchStart(feat typedef.FeatureParam) {}
func (m *MyObserver) OnFeatureFetchEnd(featureHash string, featureValue typedef.MetaType, err error) {}
func (m *MyObserver) OnJudgeNodeVisitEnd(info typedef.JudgeNode, vt typedef.VisitTarget) {}
func (m *MyObserver) OnVisitError(node, errMsg string, vt typedef.VisitTarget) {}
func main() {
arishem.ExecuteSingleRule(rule, dataCtx, WithVisitObserver(myObserver), WithFetchObserver(myObserver))
}
有关自定义配置和监听回调的更多使用方式请参考详细文档。
Arishem 基于Apache License 2.0 许可证。
Currently, we only support bug reports, summit your ISSUES here.