ZipKin入門(mén)教程
ZipKin介紹
Zipkin是一款開(kāi)源的分布式實(shí)時(shí)數(shù)據(jù)追蹤系統(tǒng)（Distributed Tracking System）瘫镇，基于 Google Dapper的論文設(shè)計(jì)而來(lái)蝶涩，由 Twitter 公司開(kāi)發(fā)貢獻(xiàn)赠摇。其主要功能是聚集來(lái)自各個(gè)異構(gòu)系統(tǒng)的實(shí)時(shí)監(jiān)控?cái)?shù)據(jù)徊都。
ZipKin架構(gòu)
ZipKin可以分為兩部分，一部分是zipkin server静檬，用來(lái)作為數(shù)據(jù)的采集存儲(chǔ)宽档、數(shù)據(jù)分析與展示尉姨；zipkin client是zipkin基于不同的語(yǔ)言及框架封裝的一些列客戶(hù)端工具，這些工具完成了追蹤數(shù)據(jù)的生成與上報(bào)功能吗冤，架構(gòu)如下：
Instrumented client和Instrumented server是分布式系統(tǒng)中的服務(wù)又厉，通過(guò)裝備庫(kù)采集跟蹤信息九府，裝備庫(kù)再調(diào)用Transport，把跟蹤信息發(fā)送給Zipkin收集器(collectors)馋没，這些收集器將跟蹤數(shù)據(jù)保存到存儲(chǔ)（storage）中昔逗。
Zipkin Server主要包括四個(gè)模塊：
（1）Collector 接收或收集各應(yīng)用傳輸?shù)臄?shù)據(jù)
（2）Storage 存儲(chǔ)接收或收集過(guò)來(lái)的數(shù)據(jù)降传，當(dāng)前支持Memory篷朵，MySQL，Cassandra婆排，ElasticSearch等声旺，默認(rèn)存儲(chǔ)在內(nèi)存中。
（3）API（Query） 負(fù)責(zé)查詢(xún)Storage中存儲(chǔ)的數(shù)據(jù)晌柬，提供簡(jiǎn)單的JSON API獲取數(shù)據(jù)辙谜，主要提供給web UI使用
（4）Web 提供簡(jiǎn)單的web界面
服務(wù)追蹤流程如下：
┌─────────────┐ ┌───────────────────────┐ ┌─────────────┐ ┌──────────────────┐
│ User Code │ │ Trace Instrumentation │ │ Http Client │ │ Zipkin Collector │
└─────────────┘ └───────────────────────┘ └─────────────┘ └──────────────────┘
│ │ │ │
┌─────────┐
│ ──┤GET /foo ├─? │ ────┐ │ │
└─────────┘ │ record tags
│ │ ?───┘ │ │
────┐
│ │ │ add trace headers │ │
?───┘
│ │ ────┐ │ │
│ record timestamp
│ │ ?───┘ │ │
┌─────────────────┐
│ │ ──┤GET /foo ├─? │ │
│X-B3-TraceId: aa │ ────┐
│ │ │X-B3-SpanId: 6b │ │ │ │
└─────────────────┘ │ invoke
│ │ │ │ request │
│
│ │ │ │ │
┌────────┐ ?───┘
│ │ ?─────┤200 OK ├─────── │ │
────┐ └────────┘
│ │ │ record duration │ │
┌────────┐ ?───┘
│ ?──┤200 OK ├── │ │ │
└────────┘ ┌────────────────────────────────┐
│ │ ──┤ asynchronously report span ├────? │
│ │
│{ │
│ "traceId": "aa", │
│ "id": "6b", │
│ "name": "get", │
│ "timestamp": 1483945573944000,│
│ "duration": 386000, │
│ "annotations": [ │
│--snip-- │
└────────────────────────────────┘
Instrumented client和server是分別使用了ZipKin Client的服務(wù)胆胰，Zipkin Client會(huì)根據(jù)配置將追蹤數(shù)據(jù)發(fā)送到Zipkin Server中進(jìn)行數(shù)據(jù)存儲(chǔ)、分析和展示澈缺。
ZipKin幾個(gè)概念
traceId：用來(lái)確定一個(gè)追蹤鏈的16字符長(zhǎng)度的字符串，在某個(gè)追蹤鏈中保持不變炕婶。
spanId：區(qū)域Id姐赡，在一個(gè)追蹤鏈中spanId可能存在多個(gè)，每個(gè)spanId用于表明在某個(gè)服務(wù)中的身份柠掂，也是16字符長(zhǎng)度的字符串项滑。
parentId：在跨服務(wù)調(diào)用者的spanId會(huì)傳遞給被調(diào)用者，被調(diào)用者會(huì)將調(diào)用者的spanId作為自己的parentId涯贞，然后自己再生成spanId枪狂。
剛發(fā)起調(diào)用時(shí)traceId和spanId是一致，parentId不存在宋渔。被調(diào)用者的traceId和調(diào)用者的traceId是一致的州疾，被調(diào)用者會(huì)產(chǎn)生自己的spanId，并且被調(diào)用者的parentId是調(diào)用者的spanId皇拣。
Span模型幾乎完全仿造了Dapper中Span模型的設(shè)計(jì)严蓖，Zipkin中的span主要包含三個(gè)數(shù)據(jù)部分：基礎(chǔ)數(shù)據(jù)（包括traceId、spanId审磁、parentId谈飒、name、timestamp和duration态蒂，主要用于跟蹤樹(shù)中節(jié)點(diǎn)的關(guān)聯(lián)和界面展示）杭措、 Annotation（用來(lái)記錄請(qǐng)求特定事件相關(guān)信息）、BinaryAnnotation（提供一些額外信息钾恢，一般以key-value對(duì)出現(xiàn)）手素。
裝備庫(kù)是實(shí)現(xiàn)采集的關(guān)鍵鸳址，針對(duì)不同語(yǔ)言，不同RPC框架泉懦，有不同的裝備庫(kù)實(shí)現(xiàn)稿黍，目前已有實(shí)現(xiàn)列表見(jiàn)zipkin的開(kāi)源社區(qū)，其中Brave是zipkin官方提供的Java的裝備庫(kù)崩哩。一個(gè)裝備庫(kù)的實(shí)現(xiàn)需要考慮如下情況：
? 實(shí)現(xiàn)語(yǔ)言和需要裝備服務(wù)的語(yǔ)言一致
? zipkin需要核心數(shù)據(jù)結(jié)構(gòu)信息記錄巡球，包括tracerid,spanid的生成，延遲時(shí)間的計(jì)算邓嘹，事件記錄酣栈，tag記錄等
? 服務(wù)之間跟蹤信息的傳遞稱(chēng)為植入，不同RPC接口植入的方式不一樣汹押，例如HTTP接口采用B3協(xié)議植入
? 植入的信息包括：Trace Id矿筝、Span Id、Parent Id棚贾、Sampled窖维、Flags
? 可支持采樣率設(shè)置，減少跟蹤導(dǎo)致的系統(tǒng)負(fù)荷
? 可調(diào)用Transport將跟蹤信息傳給zipkin
基于Zipkin提供的裝備庫(kù)妙痹， Zipkin可實(shí)現(xiàn)調(diào)用延時(shí)分析與服務(wù)依賴(lài)關(guān)系分析兩個(gè)基本功能铸史。
調(diào)用鏈分析例子
啟動(dòng)4個(gè)服務(wù)，調(diào)用關(guān)系如下：brave-webmvc-example服務(wù)調(diào)用brave-webmvc-example2细诸，brave-webmvc-example2分別調(diào)用brave-webmvc-example3和brave-webmvc-example4
[
{
"traceId": "a4aa11d855699355",
"id": "a4aa11d855699355",
"name": "get /start",
"timestamp": 1526110753393795,
"duration": 3873359,
"annotations": [
{
"timestamp": 1526110753393795,
"value": "sr",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110757267154,
"value": "ss",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
}
],
"binaryAnnotations": [
{
"key": "ca",
"value": true,
"endpoint": {
"serviceName": "",
"ipv6": "::1",
"port": 64570
}
},
{
"key": "http.method",
"value": "GET",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.path",
"value": "/start",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.class",
"value": "HomeController",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.method",
"value": "start",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
}
]
},
{
"traceId": "a4aa11d855699355",
"id": "cf49951d471ac7c5",
"name": "get /foo",
"parentId": "a4aa11d855699355",
"timestamp": 1526110753583404,
"duration": 3650640,
"annotations": [
{
"timestamp": 1526110753583404,
"value": "cs",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110754327066,
"value": "sr",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110757234044,
"value": "cr",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110757235819,
"value": "ss",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
}
],
"binaryAnnotations": [
{
"key": "ca",
"value": true,
"endpoint": {
"serviceName": "",
"ipv4": "127.0.0.1",
"port": 64578
}
},
{
"key": "http.method",
"value": "GET",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.method",
"value": "GET",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.path",
"value": "/foo",
"endpoint": {
"serviceName": "brave-webmvc-example",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.path",
"value": "/foo",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.class",
"value": "HomeController",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.method",
"value": "foo",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
}
]
},
{
"traceId": "a4aa11d855699355",
"id": "c2c029d693ecc49b",
"name": "get /bar",
"parentId": "cf49951d471ac7c5",
"timestamp": 1526110754397322,
"duration": 1583187,
"annotations": [
{
"timestamp": 1526110754397322,
"value": "cs",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110755367168,
"value": "sr",
"endpoint": {
"serviceName": "brave-webmvc-example3",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110755810759,
"value": "ss",
"endpoint": {
"serviceName": "brave-webmvc-example3",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110755980509,
"value": "cr",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
}
],
"binaryAnnotations": [
{
"key": "ca",
"value": true,
"endpoint": {
"serviceName": "",
"ipv4": "127.0.0.1",
"port": 64583
}
},
{
"key": "http.method",
"value": "GET",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.method",
"value": "GET",
"endpoint": {
"serviceName": "brave-webmvc-example3",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.path",
"value": "/bar",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.path",
"value": "/bar",
"endpoint": {
"serviceName": "brave-webmvc-example3",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.class",
"value": "HomeController",
"endpoint": {
"serviceName": "brave-webmvc-example3",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.method",
"value": "bar",
"endpoint": {
"serviceName": "brave-webmvc-example3",
"ipv4": "192.168.1.101"
}
}
]
},
{
"traceId": "a4aa11d855699355",
"id": "e3968cec8747ce95",
"name": "get /tar",
"parentId": "cf49951d471ac7c5",
"timestamp": 1526110756017988,
"duration": 1194871,
"annotations": [
{
"timestamp": 1526110756017988,
"value": "cs",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110757081683,
"value": "sr",
"endpoint": {
"serviceName": "brave-webmvc-example4",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110757212859,
"value": "cr",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"timestamp": 1526110757222145,
"value": "ss",
"endpoint": {
"serviceName": "brave-webmvc-example4",
"ipv4": "192.168.1.101"
}
}
],
"binaryAnnotations": [
{
"key": "ca",
"value": true,
"endpoint": {
"serviceName": "",
"ipv4": "127.0.0.1",
"port": 64584
}
},
{
"key": "http.method",
"value": "GET",
"endpoint": {
"serviceName": "brave-webmvc-example4",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.method",
"value": "GET",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.path",
"value": "/tar",
"endpoint": {
"serviceName": "brave-webmvc-example4",
"ipv4": "192.168.1.101"
}
},
{
"key": "http.path",
"value": "/tar",
"endpoint": {
"serviceName": "brave-webmvc-example2",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.class",
"value": "HomeController",
"endpoint": {
"serviceName": "brave-webmvc-example4",
"ipv4": "192.168.1.101"
}
},
{
"key": "mvc.controller.method",
"value": "tar",
"endpoint": {
"serviceName": "brave-webmvc-example4",
"ipv4": "192.168.1.101"
}
}
]
}
]
總結(jié)
通過(guò)上面的分析沛贪，有如下總結(jié)：
第一，在原理上震贵，在每次業(yè)務(wù)調(diào)用發(fā)生時(shí)利赋，在源頭請(qǐng)求中產(chǎn)生一個(gè)全局唯一的TraceId，通過(guò)網(wǎng)絡(luò)依次將TraceId在調(diào)用過(guò)程中透?jìng)餍上担恳粋€(gè)調(diào)用環(huán)節(jié)都用將信息記錄到Span日志中形成上下文記錄媚送，最后通過(guò)TraceId將散落在分布式系統(tǒng)上的“孤立”上下文記錄聯(lián)系在一起，重組還原出調(diào)用鏈過(guò)程寇甸，最后根據(jù)業(yè)務(wù)的需要對(duì)調(diào)用數(shù)據(jù)進(jìn)行分析塘偎。
第二，埋點(diǎn)與數(shù)據(jù)生成上拿霉，以上的實(shí)踐均采用了基于標(biāo)注的方案吟秩，該方式的缺點(diǎn)是需要代碼植入。無(wú)處不在的埋點(diǎn)以為這無(wú)處不在的代碼植入绽淘，如何做到對(duì)業(yè)務(wù)系統(tǒng)的透明化涵防，是個(gè)巨大的挑戰(zhàn)。對(duì)于這個(gè)問(wèn)題沪铭，大多數(shù)的互聯(lián)網(wǎng)公司得益于自身的高度統(tǒng)一的架構(gòu)設(shè)計(jì)壮池，代碼的植入只需要通過(guò)在中間件植入便可偏瓤，很好地解決了代碼植入對(duì)業(yè)務(wù)部件不透明的問(wèn)題。
第三椰憋，日志的收集與存儲(chǔ)涉及到數(shù)據(jù)及時(shí)性及系統(tǒng)性能需求等問(wèn)題厅克，根據(jù)業(yè)務(wù)需求，選擇最優(yōu)的存儲(chǔ)的方案橙依。
第四证舟，調(diào)用鏈跟蹤數(shù)據(jù)規(guī)范定義與可視化，調(diào)用跟蹤數(shù)據(jù)規(guī)范為數(shù)據(jù)結(jié)構(gòu)上的規(guī)范定義票编，通過(guò)規(guī)范的定義保證生成的日志最終可以被調(diào)用跟蹤系統(tǒng)進(jìn)行分析和可視化褪储。
