前一段时间开始搞优化后端 API 性能的事，发现根本无从下手，Kibana 中虽然有记一些简陋且零散的日志，但也瞅不出啥眉目来。总结了下是日志太少了，所以决定先搞搞日志的事，目标是记录 API 请求相关的日志，找出哪个接口慢，最好具体到哪个函数慢。
记日志必然要涉及日志打点，怎么做日志打点是个问题。如果直接在代码中插入日志打点代码不仅侵入性强而且工作量大，也不够灵活，于是考虑如何做智能的自动打点。我们石墨后端使用的 bay 框架（基于 Koa@1 二次开发，基本上你可以认为就是用的 Koa），controller 里用了大量的 yield，所以考虑是不是可以在 yield 表达式前后打点，调研了下觉得可以用 esprima 和 escodegen 做。
奋战几天，koa-yield-breakpoint 诞生了。
koa-yield-breakpoint 就是一个在 controllers 或者 routes 里的 yield 表达式前后打点的工具，不用插入一行日志打点代码，只需要在引入的时候配置一下，可以记录每个请求到来时 yield 表达式前后的现场，如：
yield 表达式所在的行列号（filename）
yield 表达式是执行的第几步（step）
yield 表达式字符串形式（fn）
执行 yield 表达式所花费的时间（take）
执行 yield 表达式的结果（result）
yield 表达式当时的现场（this），通过 this 你也可以添加自定义字段
原理：
重载 Module.prototype._compile，相当于 hack 了 require，如果发现是 require 了配置里指定的的文件，则进行下一步，否则返回原始代码内容，相关源代码如下：
shimmer.wrap(Module.prototype, ‘_compile‘, function (__compile) {
return function koaBreakpointCompile(content, filename) {
if (!_.includes(filenames, filename)) {
return __compile.call(this, content, filename);
}
...
};});用 esprima 解析代码，生成 AST。如：
‘use strict‘;const Mongolass = require(‘mongolass‘);const mongolass = new Mongolass();mongolass.connect(‘mongodb://localhost:27017/test‘);exports.getUsers = function* getUsers() {
yield mongolass.model(‘users‘).create({
name: ‘xx‘,
age: 18
});
const users = yield mongolass.model(‘users‘).find();
this.body = users;};会生成如下 AST：
Program {
type: ‘Program‘,
body:
[ Directive {
type: ‘ExpressionStatement‘,
expression:
Literal {
type: ‘Literal‘,
value: ‘use strict‘,
raw: ‘\‘use strict\‘‘,
loc: { start: { line: 1, column: 0 }, end: { line: 1, column: 12 } } },
directive: ‘use strict‘,
loc: { start: { line: 1, column: 0 }, end: { line: 1, column: 13 } } },
VariableDeclaration {
type: ‘VariableDeclaration‘,
declarations:
[ VariableDeclarator {
type: ‘VariableDeclarator‘,
id:
Identifier {
type: ‘Identifier‘,
name: ‘Mongolass‘,
loc: { start: { line: 3, column: 6 }, end: { line: 3, column: 15 } } },
init:
CallExpression {
type: ‘CallExpression‘,
callee:
Identifier {
type: ‘Identifier‘,
name: ‘require‘,
loc: { start: { line: 3, column: 18 }, end: { line: 3, column: 25 } } },
arguments:
[ Literal {
type: ‘Literal‘,
value: ‘mongolass‘,
raw: ‘\‘mongolass\‘‘,
loc: { start: { line: 3, column: 26 }, end: { line: 3, column: 37 } } } ],
loc: { start: { line: 3, column: 18 }, end: { line: 3, column: 38 } } },
loc: { start: { line: 3, column: 6 }, end: { line: 3, column: 38 } } } ],
kind: ‘const‘,
loc: { start: { line: 3, column: 0 }, end: { line: 3, column: 39 } } },
VariableDeclaration {
type: ‘VariableDeclaration‘,
declarations:
[ VariableDeclarator {
type: ‘VariableDeclarator‘,
id:
Identifier {
type: ‘Identifier‘,
name: ‘mongolass‘,
loc: { start: { line: 4, column: 6 }, end: { line: 4, column: 15 } } },
init:
NewExpression {
type: ‘NewExpression‘,
callee:
Identifier {
type: ‘Identifier‘,
name: ‘Mongolass‘,
loc: { start: { line: 4, column: 22 }, end: { line: 4, column: 31 } } },
arguments: [],
loc: { start: { line: 4, column: 18 }, end: { line: 4, column: 33 } } },
loc: { start: { line: 4, column: 6 }, end: { line: 4, column: 33 } } } ],
kind: ‘const‘,
loc: { start: { line: 4, column: 0 }, end: { line: 4, column: 34 } } },
ExpressionStatement {
type: ‘ExpressionStatement‘,
expression:
CallExpression {
type: ‘CallExpression‘,
callee:
StaticMemberExpression {
type: ‘MemberExpression‘,
computed: false,
object:
Identifier {
type: ‘Identifier‘,
name: ‘mongolass‘,
loc: { start: { line: 5, column: 0 }, end: { line: 5, column: 9 } } },
property:
Identifier {
type: ‘Identifier‘,
name: ‘connect‘,
loc: { start: { line: 5, column: 10 }, end: { line: 5, column: 17 } } },
loc: { start: { line: 5, column: 0 }, end: { line: 5, column: 17 } } },
arguments:
[ Literal {
type: ‘Literal‘,
value: ‘mongodb://localhost:27017/test‘,
raw: ‘\‘mongodb://localhost:27017/test\‘‘,
loc: { start: { line: 5, column: 18 }, end: { line: 5, column: 50 } } } ],
loc: { start: { line: 5, column: 0 }, end: { line: 5, column: 51 } } },
loc: { start: { line: 5, column: 0 }, end: { line: 5, column: 52 } } },
ExpressionStatement {
type: ‘ExpressionStatement‘,
expression:
AssignmentExpression {
type: ‘AssignmentExpression‘,
operator: ‘=‘,
left:
StaticMemberExpression {
type: ‘MemberExpression‘,
computed: false,
object:
Identifier {
type: ‘Identifier‘,
name: ‘exports‘,
loc: { start: { line: 7, column: 0 }, end: { line: 7, column: 7 } } },
property:
Identifier {
type: ‘Identifier‘,
name: ‘getUsers‘,
loc: { start: { line: 7, column: 8 }, end: { line: 7, column: 16 } } },
loc: { start: { line: 7, column: 0 }, end: { line: 7, column: 16 } } },
right:
FunctionExpression {
type: ‘FunctionExpression‘,
id:
Identifier {
type: ‘Identifier‘,
name: ‘getUsers‘,
loc: { start: { line: 7, column: 29 }, end: { line: 7, column: 37 } } },
params: [],
body:
BlockStatement {
type: ‘BlockStatement‘,
body:
[ ExpressionStatement {
type: ‘ExpressionStatement‘,
expression:
YieldExpression {
type: ‘YieldExpression‘,
argument:
CallExpression {
type: ‘CallExpression‘,
callee:
StaticMemberExpression {
type: ‘MemberExpression‘,
computed: false,
object: [Object],
property: [Object],
loc: [Object] },
arguments: [ [Object] ],
loc: { start: [Object], end: [Object] } },
delegate: false,
loc: { start: { line: 8, column: 2 }, end: { line: 11, column: 4 } } },
loc: { start: { line: 8, column: 2 }, end: { line: 11, column: 5 } } },
VariableDeclaration {
type: ‘VariableDeclaration‘,
declarations:
[ VariableDeclarator {
type: ‘VariableDeclarator‘,
id: Identifier { type: ‘Identifier‘, name: ‘users‘, loc: [Object] },
init:
YieldExpression {
type: ‘YieldExpression‘,
argument: [Object],
delegate: false,
loc: [Object] },
loc: { start: [Object], end: [Object] } } ],
kind: ‘const‘,
loc:
{ start: { line: 13, column: 2 },
end: { line: 13, column: 54 } } },
ExpressionStatement {
type: ‘ExpressionStatement‘,
expression:
AssignmentExpression {
type: ‘AssignmentExpression‘,
operator: ‘=‘,
left:
StaticMemberExpression {
type: ‘MemberExpression‘,
computed: false,
object: ThisExpression { type: ‘ThisExpression‘, loc: [Object] },
property: Identifier { type: ‘Identifier‘, name: ‘body‘, loc: [Object] },
loc: { start: [Object], end: [Object] } },
right:
Identifier {
type: ‘Identifier‘,
name: ‘users‘,
loc: { start: [Object], end: [Object] } },
loc:
{ start: { line: 14, column: 2 },
end: { line: 14, column: 19 } } },
loc:
{ start: { line: 14, column: 2 },
end: { line: 14, column: 20 } } } ],
loc: { start: { line: 7, column: 40 }, end: { line: 15, column: 1 } } },
generator: true,
expression: false,
loc: { start: { line: 7, column: 19 }, end: { line: 15, column: 1 } } },
loc: { start: { line: 7, column: 0 }, end: { line: 15, column: 1 } } },
loc: { start: { line: 7, column: 0 }, end: { line: 15, column: 2 } } } ],
sourceType: ‘script‘,
loc: { start: { line: 1, column: 0 }, end: { line: 15, column: 2 } } }相关源代码如下：
let parsedCodes;try {
parsedCodes = esprima.parse(content, { loc: true });} catch (e) {
console.error(‘cannot parse file: %s‘, filename);
console.error(e.stack);
process.exit(1);}遍历找到 YieldExpression 节点，进行以下包装后生成 AST 替换掉原来的节点。
global.logger(
this,
function*(){
return YieldExpression
},
YieldExpressionString,
filename);相关源代码如下：
findYieldAndWrapLogger(parsedCodes);try {
content = escodegen.generate(parsedCodes, {
format: { indent: { style: ‘ ‘ } },
sourceMap: filename,
sourceMapWithCode: true
});} catch (e) {
console.error(‘cannot generate code for file: %s‘, filename);
console.error(e.stack);
process.exit(1);}debug(‘file %s regenerate codes:\n%s‘, filename, content.code);findYieldAndWrapLogger 作用就是遍历 AST 将 YieldExpression 替换成用日志函数包裹后新的 YieldExpression 的 AST。
最后用 escodegen 将 AST 生成代码（支持 soucemap，所以错误栈对应的行数是对的）。
核心：每个请求到来时，生成（项目中本来就有 requestId 的话，可通过配置指定）一个 requestId（uuid 类型）挂载到 this 上，这样就可以通过 requestId 将日志串起来了。
特点：可以记录每个请求的流程（甚至可以查到某个用户某个时间段的请求），及每一次请求的每一步（step）的现场及返回值，方便调试或查 bug，当然只针对 yield 表达式。
app.js
‘use strict‘;const koaYieldBreakpoint = require(‘koa-yield-breakpoint‘)({
files: [‘./routes/*.js‘]});const koa = require(‘koa‘);const routes = require(‘./routes‘);const app = koa();app.use(koaYieldBreakpoint);routes(app);app.listen(3000, () => {
console.log(‘listening on 3000‘);});routes/index.js
‘use strict‘;const route = require(‘koa-route‘);const users = require(‘./users‘);module.exports = function (app) {
app.use(route.get(‘/users‘, users.getUsers));};routes/users.js
‘use strict‘;const Mongolass = require(‘mongolass‘);const mongolass = new Mongolass();mongolass.connect(‘mongodb://localhost:27017/test‘);exports.getUsers = function* getUsers() {
yield mongolass.model(‘users‘).create({
name: ‘xx‘,
age: 18
});
const users = yield mongolass.model(‘users‘).find();
this.body = users;};运行：
DEBUG=koa-yield-breakpoint node app.js
终端打印如下：
example git:(master) DEBUG=koa-yield-breakpoint node app.jskoa-yield-breakpoint options: {"files":["./routes/*.js"],"nodir":true,"absolute":true,"filter":{"ctx":["state","params"],"request":["method","path","header","query","body"],"response":["status","body"]},"loggerName":"logger","requestIdPath":"requestId"} +0mskoa-yield-breakpoint matched files: ["/Users/nswbmw/node/koa-yield-breakpoint/example/routes/index.js","/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js"] +9mskoa-yield-breakpoint file /Users/nswbmw/node/koa-yield-breakpoint/example/routes/index.js regenerate codes:‘use strict‘;const route = require(‘koa-route‘);const users = require(‘./users‘);module.exports = function (app) {
app.use(route.get(‘/users‘, users.getUsers));}; +73mskoa-yield-breakpoint file /Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js regenerate codes:‘use strict‘;const Mongolass = require(‘mongolass‘);const mongolass = new Mongolass();mongolass.connect(‘mongodb://localhost:27017/test‘);exports.getUsers = function* getUsers() {
yield global.logger(this, function* () {
return mongolass.model(‘users‘).create({
name: ‘xx‘,
age: 18
});
}, ‘mongolass.model(\‘users\‘).create({\n name: \‘xx\‘,\n age: 18\n})‘, ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:8:2‘);
const users = yield global.logger(this, function* () {
return mongolass.model(‘users‘).find();
}, ‘mongolass.model(\‘users\‘).find()‘, ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:13:16‘);
this.body = users;}; +15mslistening on 3000可以看出，routes/users.js 被替换成了：
‘use strict‘;const Mongolass = require(‘mongolass‘);const mongolass = new Mongolass();mongolass.connect(‘mongodb://localhost:27017/test‘);exports.getUsers = function* getUsers() {
yield global.logger(this, function* () {
return mongolass.model(‘users‘).create({
name: ‘xx‘,
age: 18
});
}, ‘mongolass.model(\‘users\‘).create({\n name: \‘xx\‘,\n age: 18\n})‘, ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:8:2‘);
const users = yield global.logger(this, function* () {
return mongolass.model(‘users‘).find();
}, ‘mongolass.model(\‘users\‘).find()‘, ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:13:16‘);
this.body = users;};为什么要将 yield 表达式包裹成这样呢？
global.logger(
this,
function*(){
return YieldExpression
},
YieldExpressionString,
filename);而不是：
global.logger(
this,
function(){
return YieldExpression
},
YieldExpressionString,
filename);或者：
global.logger( this, yield YieldExpression, YieldExpressionString, filename);
因为考虑到 YieldExpression 可能是以下情况导致 esprima 不能解析：
User.find(yield xxx)
User.find(http://this.xxx)
所以综合考虑只能放到 GeneratorFunction 里，另外，不用：
global.logger(
this,
function*(){
return yield YieldExpression
},
YieldExpressionString,
filename
);是因为这样会导致 esprima 解析死循环，所以去掉了里面的 yield，在 logger 里 yield。
访问 localhost:3000/users，终端打印出:
{ requestId: ‘dad593c0-c4a1-4640-a00e-9ba0349cfd2f‘,
timestamp: Wed Oct 26 2016 18:28:49 GMT+0800 (CST),
this:
{ state: {},
request:
{ method: ‘GET‘,
path: ‘/users‘,
header: [Object],
query: [Object] },
response: { status: 404, body: undefined } },
type: ‘start‘,
step: 1,
take: 0 }{ requestId: ‘dad593c0-c4a1-4640-a00e-9ba0349cfd2f‘,
step: 2,
filename: ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:8:2‘,
timestamp: Wed Oct 26 2016 18:28:49 GMT+0800 (CST),
this:
{ state: {},
request:
{ method: ‘GET‘,
path: ‘/users‘,
header: [Object],
query: [Object] },
response: { status: 404, body: undefined } },
type: ‘before‘,
fn: ‘mongolass.model(\‘users\‘).create({\n name: \‘xx\‘,\n age: 18\n})‘,
result: undefined,
take: 1 }{ requestId: ‘dad593c0-c4a1-4640-a00e-9ba0349cfd2f‘,
step: 3,
filename: ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:8:2‘,
timestamp: Wed Oct 26 2016 18:28:49 GMT+0800 (CST),
this:
{ state: {},
request:
{ method: ‘GET‘,
path: ‘/users‘,
header: [Object],
query: [Object] },
response: { status: 404, body: undefined } },
type: ‘after‘,
fn: ‘mongolass.model(\‘users\‘).create({\n name: \‘xx\‘,\n age: 18\n})‘,
result:
{ result: { ok: 1, n: 1 },
ops: [ [Object] ],
insertedCount: 1,
insertedIds: [ undefined, 5810856182d0eea2f12030fd ] },
take: 7 }{ requestId: ‘dad593c0-c4a1-4640-a00e-9ba0349cfd2f‘,
step: 4,
filename: ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:13:16‘,
timestamp: Wed Oct 26 2016 18:28:49 GMT+0800 (CST),
this:
{ state: {},
request:
{ method: ‘GET‘,
path: ‘/users‘,
header: [Object],
query: [Object] },
response: { status: 404, body: undefined } },
type: ‘before‘,
fn: ‘mongolass.model(\‘users\‘).find()‘,
result: undefined,
take: 1 }{ requestId: ‘dad593c0-c4a1-4640-a00e-9ba0349cfd2f‘,
step: 5,
filename: ‘/Users/nswbmw/node/koa-yield-breakpoint/example/routes/users.js:13:16‘,
timestamp: Wed Oct 26 2016 18:28:49 GMT+0800 (CST),
this:
{ state: {},
request:
{ method: ‘GET‘,
path: ‘/users‘,
header: [Object],
query: [Object] },
response: { status: 404, body: undefined } },
type: ‘after‘,
fn: ‘mongolass.model(\‘users\‘).find()‘,
result: [ { _id: 5810856182d0eea2f12030fd, name: ‘xx‘, age: 18 } ],
take: 4 }{ requestId: ‘dad593c0-c4a1-4640-a00e-9ba0349cfd2f‘,
timestamp: Wed Oct 26 2016 18:28:49 GMT+0800 (CST),
this:
{ state: {},
request:
{ method: ‘GET‘,
path: ‘/users‘,
header: [Object],
query: [Object] },
response: { status: 200, body: [Object] } },
type: ‘end‘,
step: 6,
take: 1 }注意：type 是 [‘start‘, ‘beforeYield‘, ‘afterYield‘, ‘error‘, ‘end‘] 其中一种，take 单位是 ms。
start: 请求到来时第一次打点
beforeYield: 上一个 yieldExpression 之后到这一个 yieldExpression 之前
afterYield: 这个 yieldExpression 开始到结束
error: 错误日志，koa-yield-breakpoint 会记录下来，然后再 throw 出去
end: 请求结束时打点
const koaYieldBreakpoint = require(‘koa-yield-breakpoint‘)({
files: [‘./controllers/*.js‘, ‘./middlewares/*.js‘],
store: require(‘./lib/koa_yield_breakpoint_logstash‘),
loggerName: ‘__logger‘,
requestIdPath: ‘header.reqid‘,
yieldCondition: (filename, yieldExpression) => {
// yield next 不包裹 logger
if (yieldExpression === ‘next‘) {
return {
wrapYield: false,
deep: false
};
}
if (yieldExpression.match(/^db\.sequelize\.transaction/)) {
return {
wrapYield: true,
deep: false
};
}
}});...app.use(koaYieldBreakpoint);koa-yield-breakpoint 主要用在 Koa 的 routes、controllers 或 middlewares 文件。co-yield-breakpoint 主要用来给普通 yield 表达式（如 lib 或 models 下的文件）打点（没有 requestId，所以 log 是零散的，这里只用来计算平均耗时），实现原理跟 koa-yield-breakpoint 类似，这里不再赘述。
foo.js
‘use strict‘;const co = require(‘co‘);const Mongolass = require(‘mongolass‘);const mongolass = new Mongolass();mongolass.connect(‘mongodb://localhost:27017/test‘);co(function* () {
yield mongolass.model(‘users‘).create({
name: ‘xx‘,
age: 18
});
const users = yield mongolass.model(‘users‘).find();
console.log(‘users: %j‘, users);}).catch(e => console.error(e.stack));终端打印如下：
{
"filename": "/Users/nswbmw/node/co-yield-breakpoint/example/foo.js:9:2",
"timestamp": "2016-11-10T13:53:33.053Z",
"fn": "mongolass.model(‘users‘).create({\n name: ‘xx‘,\n age: 18\n})",
"result": {
"result": {
"ok": 1,
"n": 1
},
"ops": [{
"name": "xx",
"age": 18,
"_id": "58247bdd3fb205dbad5418a0"
}],
"insertedCount": 1,
"insertedIds": [null, "58247bdd3fb205dbad5418a0"]
},
"take": 55}{
"filename": "/Users/nswbmw/node/co-yield-breakpoint/example/foo.js:14:16",
"timestamp": "2016-11-10T13:53:33.060Z",
"fn": "mongolass.model(‘users‘).find()",
"result": [{
"_id": "58247bdd3fb205dbad5418a0",
"name": "xx",
"age": 18
}],
"take": 7}users: [{
"_id": "58247bdd3fb205dbad5418a0",
"name": "xx",
"age": 18}]真实用例：
require(‘co-yield-breakpoint‘)({
files: [‘./lib/**/*.js‘, ‘./models/**/*.js‘],
store: require(‘./lib/koa_yield_breakpoint_logstash‘),
loggerName: ‘__coLogger‘});store 参数最好自己定义，可以是一个 object 可以是一个实例，只需要有一个 save 方法即可。在 save 方法内可做一些自定义，比如我在 koa_yield_breakpoint_logstash.js 里添加了一些字段（如：url），同时对正常请求日志过滤了一些字段，对错误请求日志添加了一些字段方便 bug 追踪。如下：
if (ctx.request) {
record.request = _.pick(ctx.request, ‘headers‘, ‘body‘, ‘query‘, ‘ips‘);}线上 API 在上了 koa-yield-breakpoint + co-yield-breakpoint 这个组合后，我又让运维在 Kibana 作了以下 3 种图表：
fn 的平均耗时，由高到低（根据 filename 聚合, 因为 filename 中包含行列数，唯一）
每个接口的平均耗时，由高到低（根据 url 聚合）
所有 requestId 耗时排行（对应这个 requestId 所有日志 take 值累加），由高到低
看起来效果良好，但后来运维告诉我日志打的太多了，想想也是，一个逻辑复杂的 api 的请求从开始到结束，可能执行了几十次 yield。。讨论了下，决定只采样 10% 日志。如下：
exports.save = function save(record, ctx) {
...
if (Math.random() > 0.1 && !record.error) {// co-yield-breakpoint 的 record.error 总为 undefined
return;
}
...};注意：这里 koa-yield-breakpoint 和 co-yield-breakpoint 的日志都随机取样了 10%， co-yield-breakpoint 的日志倒无所谓，但 koa-yield-breakpoint 的日志则被打散了，也就是说你基本上不可能通过一个 requestId 查到完整的请求日志链了（step 是连续的）。另一种方案是在 koa-yield-breakpoint 中间件之前给 this 添加一个属性：
function *(next) {
this.shouldSaveLog = Math.random() < 0.1;
yield next;};然后再 save 方法内判断一下 shouldSaveLog 决定是否记录这条日志，上面代码改为：
exports.save = function save(record, ctx) {
...
if (!ctx.shouldSaveLog && !record.error) {
return;
}
...};这样也实现了只记录 10% 的请求日志，但这种方案是：这个请求的所有日志要么都记，要么都不记（错误日志除外，错误日志一定记）。
我们正在招聘！
[北京/武汉] 石墨文档 做最美产品 - 寻找中国最有才华的工程师加入
原文地址：http://12405360.blog.51cto.com/12395360/1882747
