–> go to 总目录
讲述了 zipkin的x

https://raw.githubusercontent.com/openzipkin/b3-propagation/master/README.md

b3-propagation

B3 Propagation is a specification for the header “b3” and those that start with “x-b3-”. These headers are used for trace context propagation across service boundaries.

B3propagation是一堆拥有"x-b3-"前缀的header

Overview

This specification elaborates identifiers used to place an operation in a trace tree. These attributes are propagated in-process, and eventually downstream (often via http headers), to ensure all activity originating from the same root are collected together. A sampling decision is made at the root of the trace, and this indicates if trace details should be collected and reported to the tracing system (usually Zipkin) or not.

规范清晰身份ID通常被放置在trace树的一个操作中。这些属性以in-process的方式传递并且最终顺流而下以确保所有活动（从一个root产生的）能被同时收集。一个采集的决定也是root节点就确认好的，这个策略也指明这个trace是否被收集和发送到追踪系统中去（通常是zipkin）

Overall Process

总体流程

The most common propagation use case is to copy a trace context from a client
sending an RPC request to a server receiving it.

In this case, the same trace IDs are used, which means that both the client and
server side of an operation end up in the same node in the trace tree.

Here’s an example flow using multiple header encoding, assuming an HTTP request carries the propagated trace:

最常见的传递示例是，当一个cliect 发送 RPC request时 将context 发送给一个Server
如下示例中，同样的traceIDs被使用，这意味着一个operation的client和server最终会显示成trace树的一个节点

   Client Tracer                                                  Server Tracer
┌───────────────────────┐                                       ┌───────────────────────┐
│                       │                                       │                       │
│   TraceContext        │          Http Request Headers         │   TraceContext        │
│ ┌───────────────────┐ │         ┌───────────────────┐         │ ┌───────────────────┐ │
│ │ TraceId           │ │         │ X-B3-TraceId      │         │ │ TraceId           │ │
│ │                   │ │         │                   │         │ │                   │ │
│ │ ParentSpanId      │ │ Inject  │ X-B3-ParentSpanId │ Extract │ │ ParentSpanId      │ │
│ │                   ├─┼────────>│                   ├─────-───┼>│                   │ │
│ │ SpanId            │ │         │ X-B3-SpanId       │         │ │ SpanId            │ │
│ │                   │ │         │                   │         │ │                   │ │
│ │ Sampling decision │ │         │ X-B3-Sampled      │         │ │ Sampling decision │ │
│ └───────────────────┘ │         └───────────────────┘         │ └───────────────────┘ │
│                       │                                       │                       │
└───────────────────────┘                                       └───────────────────────┘

Trace identifiers are often sent with a sampling decision. However, it is valid and common practice to send a sampling decision independently. Here’s an example of a proxy forbidding tracing of a /health endpoint. The diagram shows the receiver making a NoOp trace context to ensure minimal overhead.

trace 身份标识通常会携带一个采样策略。然而合法且通用的是独立的去发送这个策略。以下是一个举例：一个代理端禁止/health
端。这个图表显示receiver创造了一个NoOP context(假contet)区保证最小化数据。

                                Server Tracer     ┌───────────────────────┐Health check request         │                       │
┌───────────────────┐         │   TraceContext        │
│ GET /health       │ Extract │ ┌───────────────────┐ │
│ X-B3-Sampled: 0   ├─────────┼>│ NoOp              │ │
└───────────────────┘         │ └───────────────────┘ │└───────────────────────┘

Identifiers

Trace identifiers are 64 or 128-bit, but all span identifiers within a trace are 64-bit. All identifiers are opaque.

Identifiers are almost always sent with Sampling state, but they can be sent alone to implement a “Defer” decision described below.

trace的身份证是64或128，但是trace内span的id都在64位以内。
身份Id 通常会有采样状态。但是他们可以独立的实现一个"defer"策略(如下描述)

TraceId

The TraceId is 64 or 128-bit in length and indicates the overall ID of the trace. Every span in a trace shares this ID.

traceId是总Id,每个span都有

SpanId

The SpanId is 64-bit in length and indicates the position of the current operation in the trace tree. The value should not be interpreted: it may or may not be derived from the value of the TraceId.

spanId可以指示在trace中的位置。跟traceId关系不大

ParentSpanId

The ParentSpanId is 64-bit in length and indicates the position of the parent operation in the trace tree. When the span is the root of the trace tree, there is no ParentSpanId.

Sampling State 重点

Sampling is a mechanism to reduce the volume of data that ends up in the tracing system. In B3, sampling applies consistently per-trace: once the sampling decision is made, the same value should be consistently sent downstream. This means you will see all spans sharing a trace ID or none.

采样是一个减少数据量组件。b3中每条trace都有这个标记：一旦这个策略被设定，就会顺延到整个链路中去。这意味着所有span要么有traceId，要么都没有。

Here are the valid sampling states. Note they all applied to the trace ID, not the span ID:

• Defer: aka I don’t know yet!
  • Defer is used when trace identifiers are set by a proxy, but that proxy doesn’t send data to Zipkin. The most common use case for defering a decision is pre-provisioning trace identifiers.
  • In all known encodings, defer is the absence of sampling state.
• Deny: aka don’t sample or don’t record
  • Deny is used to achieve a probabilistic rate or to prevent certain paths (such as health checks) from generating traces. Where possible, instrumentation should should optimize deny such that less overhead occurs.
• Accept: aka sample or record
  • Accept is used to achieve a probabilistic rate or to ensure certain paths (such as low-traffic endpoints) are always traced. When a trace is accepted, spans should be reported to zipkin except in overload scenarios.
• Debug: aka force trace
  • Debug is a production troubleshooting aid used in tools like curl or chrome debug. Debug is an emphasized accept decision that implies accept, additionally reporting Span.debug = true for each span in the trace.

以下是合法的四种状态。都应用于traceId，而不是span ID:

• Defer 状态异常。比如设置了traceId,但是proxy没发送,标记为Defer。这是sampling状态缺失的标记
• Deny 拒绝 。调整速率，或者过滤某些path（如健康监察），这种标记instrumentation应该尽可能少的往header塞入值
• accept 接受。 调整速率，或确保某些具体的paths被追踪到。
• debug 强制接受。用来调测

The most common use of sampling is probablistic: eg, accept 0.01% of traces and deny the rest. Debug is the least common use case.

最常用的采样标识是：accept 0.01%和拒绝rest。debug是最少用的
Sampling state is almost always sent with Identifiers, but it can be sent alone (a predefined decision). This is used for a few use cases:

• Deny: Using a constant entry for a deny decision is more efficient than generating identifiers. Libraries that do this cannot do external ID correlation, though, for example, log correlation.
• Accept: Some proxies send a constant entry for an accept decision as they want a specific path to be traced, but don’t want to interfere with ID generation.
• Debug: Some support guides send debug without IDs as it is easier to do in curl.

Encodings

Below are a number of predefined encodings. The oldest encoding is multiple HTTP headers (2012). Some application notes have been made since, as well a compact, single-header encoding. A binary encoding may be defined as well.

Most users do not need to know the encodings listed here as they are built-in to libraries, frameworks and proxies. While this specification aims to be complete, it will not elaborate all use cases and practice. Please contact the Zipkin gitter if you are confused about something.

id的编码规则，不需要了解很细

Custom Encodings

We recommend re-using existing encodings over developing new ones as there are usually subtleties that even experienced folks overlook. For example, one proxy accidentally chose to use a hyphen to substitute for an absent parent span ID. This caused broken traces, and in some extreme cases, crashed requests. Some types of software have multiple month or longer release cycles, which adds tension to add “spaghetti” to clean it up. Identifying and answering questions about such problems add support load unfairly to others as they weren’t consulted. Even ruling out format mismatches, misunderstanding of the feature set can cause expensive maintenance later.

The best way to avoid costly mistakes is to re-use an encoding. The second-best way is reading this spec completely when designing a bespoke encoding, and getting feedback from the authors on your work. If you are designing a new encoding, please contact the Zipkin gitter so that you have the best chance to avoid mistakes others made in the past.

避免自造，重用编码

Http Encodings

There are two encodings of B3: Single Header and Multiple Header. Multiple header encoding uses an X-B3- prefixed header per item in the trace context. Single header delimits the context into into a single entry named b3. The single-header variant takes precedence over the multiple header one when extracting fields.

http 编解码，分为single Header: b3;Multiple Header ‘X-B3-’。single优先于Multiple

Multiple Headers

B3 attributes are most commonly propagated as multiple http headers. All B3 headers follows the convention of X-B3-${name} with special-casing for flags. When extracting state headers, the first value wins.

Note: Http headers are case-insensitive, but sometimes this encoding is used for other transports. When encoding in case-sensitive transports, prefer lowercase keys or the single header header encoding which is explicitly lowercase.

As mentioned earlier, identifiers can be sent with or without sampling state and visa versa. It is important to understand the relationships between these headers. For example, X-B3-TraceId and X-B3-SpanId can be sent alone, or with a sampling header X-B3-Sampled. It is also valid to send sampling state independently, such as a deny decision X-B3-Sampled: 0.

复合headers。注意X-B3-Sampled: 0可以独立发送，其余不能被解析

TraceId

The X-B3-TraceId header is encoded as 32 or 16 lower-hex characters. For example, a 128-bit TraceId header might look like: X-B3-TraceId: 463ac35c9f6413ad48485a3953bb6124. Unless propagating only the Sampling State, the X-B3-TraceId header is required.

SpanId

The X-B3-SpanId header is encoded as 16 lower-hex characters. For example, a SpanId header might look like: X-B3-SpanId: a2fb4a1d1a96d312. Unless propagating only the Sampling State, the X-B3-SpanId header is required.

ParentSpanId

The X-B3-ParentSpanId header may be present on a child span and must be absent on the root span. It is encoded as 16 lower-hex characters. For example, a ParentSpanId header might look like: X-B3-ParentSpanId: 0020000000000001

Sampling State

An accept sampling decision is encoded as X-B3-Sampled: 1 and a deny as X-B3-Sampled: 0. Absent means defer the decision to the receiver of this header. For example, a Sampled header might look like: X-B3-Sampled: 1.

Note: Before this specification was written, some tracers propagated X-B3-Sampled as true or false as opposed to 1 or 0. While you shouldn’t encode X-B3-Sampled as true or false, a lenient implementation may accept them.

Debug Flag

Debug is encoded as X-B3-Flags: 1. Debug implies an accept decision, so don’t also send the X-B3-Sampled header.

Single Header

A single header named b3 standardized in late 2018 for use in JMS and w3c tracestate. Design and rationale are captured here. Check or update our status page for adoption status.

In simplest terms b3 maps propagation fields into a hyphen delimited string.

b3={TraceId}-{SpanId}-{SamplingState}-{ParentSpanId}, where the last two fields are optional.

For example, the following state encoded in multiple headers:

X-B3-TraceId: 80f198ee56343ba864fe8b2a57d3eff7
X-B3-ParentSpanId: 05e3ac9a4f6e3b90
X-B3-SpanId: e457b5a2e4d86bd1
X-B3-Sampled: 1

Becomes one b3 header, for example:

b3: 80f198ee56343ba864fe8b2a57d3eff7-e457b5a2e4d86bd1-1-05e3ac9a4f6e3b90

Note: When only propagating a sampling decision, the header is still named b3, but only contains the sampling state.

A deny decision encodes as:

b3: 0

TraceId

The first position of the b3 value is the 32 or 16 lower-hex character TraceId. Unless propagating
only the Sampling State, the TraceId is required.

For example, a TraceId of 80f198ee56343ba864fe8b2a57d3eff7 encodes in the left-most position:

b3: 80f198ee56343ba864fe8b2a57d3eff7-e457b5a2e4d86bd1-1

SpanId

The second position of the b3 value is the 16 lower-hex character SpanId. Unless propagating only
the Sampling State, the SpanId is required.

For example, a SpanId of e457b5a2e4d86bd1 encodes after the left-most hyphen:

b3: 80f198ee56343ba864fe8b2a57d3eff7-e457b5a2e4d86bd1-1

Sampling State

When the sampling state isn’t Defer, it is encoded as a single hex character in the third position
of the b3 value. Sampling state can also be sent on its own (with no identifiers).

Sampling State is encoded as a single hex character for all states except Defer. Defer is absence of
the sampling field.

• Accept: 1
• Deny: 0
• Debug: d

For example, a debug trace could look like this

b3: 80f198ee56343ba864fe8b2a57d3eff7-e457b5a2e4d86bd1-d

A deny decision may omit identifiers and look like this:

b3: 0

ParentSpanId

When present, ParentSpanId is the 16 lower-hex characters in the final position of b3.

For example, a ParentSpanId of 05e3ac9a4f6e3b90 encodes in the right-most position:

b3: 80f198ee56343ba864fe8b2a57d3eff7-e457b5a2e4d86bd1-1-05e3ac9a4f6e3b90

gRPC Encoding

B3 attributes can also be propagated as ASCII headers in the Custom Metadata of
a request. The encoding is exactly the same as Http headers, except the names are
explicitly or implicitly down-cased.

For example, when using multiple headers, X-B3-ParentSpanId: 0020000000000001 encodes
as an ASCII header x-b3-parentspanid with the same value.

JMS Encoding

JMS (Java Message Service) has constraints that disallow X-B3- prefixed headers.
As such, only the single header format b3 should be used with JMS. As messaging spans
never share a SpanId, it is encouraged to omit the ParentSpanId field when propagating.

Refer to our design
documentation for details.

Frequently Asked Questions 重点

常见问题

Why is ParentSpanId propagated?

In B3, the trace context is extracted from incoming headers. Timing and
metadata for the client and server side of an operation are recorded with
the same context. The ParentSpanId is the ID of the operation that caused
the current RPC. For example, it could be the ID of another server request
or a scheduled job. ParentSpanId is propagated so that when data is reported
to Zipkin, it can be placed in the correct spot in the trace tree.

Here’s an example of a B3 library extracting a trace context from incoming http
request headers:

                           ┌───────────────────┐Incoming Headers          │   TraceContext    │
┌───────────────────┐      │ ┌───────────────┐ │
│ X-B3-TraceId      │──────┼─┼> TraceId      │ │
│                   │      │ │               │ │
│ X-B3-ParentSpanId │──────┼─┼> ParentSpanId │ │
│                   │      │ │               │ │
│ X-B3-SpanId       │──────┼─┼> SpanId       │ │
│                   │      │ └───────────────┘ │
└───────────────────┘      └───────────────────┘

Some propagation formats look similar to B3, but don’t propagate a field named
parent. Instead, they propagate a span ID field which serves the same purpose
as ParentSpanId. Unlike B3, these systems use a different span ID for the
client and server side of an RPC. When a server reads headers like this, it is
expected to provision a new span ID for itself, and use the one it extracted as
its parent.

Here’s an example of an alternate library composing a trace context with
incoming http request headers and an ID generator:

                           ┌───────────────────┐Incoming Headers          │   TraceContext    │
┌───────────────────┐      │ ┌───────────────┐ │
│ XXXX-TraceId      │──────┼─┼> TraceId      │ │
│                   │      │ │               │ │
│ XXXX-SpanId       │──────┼─┼> ParentSpanId │ │
└───────────────────┘      │ │               │ │      ┌──────────────┐│ │  SpanId      <┼─┼──────│ ID Generator ││ └───────────────┘ │      └──────────────┘└───────────────────┘

In both B3 and the above example, incoming headers contain the parent’s span ID,
and three IDs (trace, parent, span) end up in the trace context. The difference
is that B3 uses the same span ID for the client and server side of an RPC, where
the latter does not.

Why propagate a reject sampling decision?

It may not be obvious why you’d send a reject sampling decision to the next hop.
Imagine a service decides not to trace an operation and makes 2 out-going calls,
and these branched out further. If reject (“don’t trace”) isn’t propagated, the
system might receive only parts of the operation, confusing users.

Another reason to reject sampling is to prevent overwhelming the collector with
high throughput. If the reject-sampling decision is not propagated then there’s
no way to tell recipients of outbound calls that they should not send span info
to the collector.

Why send trace IDs with a reject sampling decision?

While valid to propagate only a reject sampling decision (X-B3-Sampled: 0),
if trace identifiers are established, they should be propagated, too. The
tracing system is often not the only consumer of trace identifiers. Services may
still want to do things with the tracing info even if they aren’t sending span
data to the collector, e.g. tag logs with trace IDs.

Why defer a sampling decision?

Deferring the sampling decision is special-case. The only known use-cases are
the following:

• Debug trace: A debug trace is implicitly an accepted trace
• Externally provisioned IDs: When you want to control IDs, but not sampling policy

Unless it is a debug trace, leaving out a sampling decision is typically ID
correlation. For example, someone re-uses a global identifier from another
system, or correlating in logs. In these cases, the caller knows the ID they
want, but allows the next hop to decide if it will be traced or not. The caller
should not report a span to the tracing system using this ID unless they
propagate an accept sampling decision.

Why is Debug encoded as X-B3-Flags: 1?

The first tracer to use B3 was Finagle. In Finagle, thrift-rpc was a dominant
protocol, and a fixed-width binary encoding held B3 attributes. This binary
encoding used a bit field to encode the sampling decision and the debug flag.

Http encoding chose to use a separate header for the sampling decision, but
kept the flags field for debug. In hind sight, it would have made more sense to
use a separate header such as X-B3-Debug: 1 as no other flags were added and
no implementation treats X-B3-Flags as a bit field. That said, renaming the
field would cause more damage than good, so it was left alone.

b b3-propagation协议相关推荐

1. 链路追踪——【Brave】第一遍小结

   前言 微服务链路追踪系列博客,后续可能会涉及到Brave.Zipkin.Sleuth内容的梳理. Brave 何为Brave? github地址:https://github.com/openzipk ...

2. day_44_Django

   day44 1.内容回顾 12月18号开班 python基础语法 网络编程 并发编程 前端(HTML.css.js) 数据库(mysql) 2.今日概要 1.了解web应用程序的本质 2.Django ...

3. 车载无线自组织网络的介质访问控制协议研究

   摘  要 车载无线自组织网络技术(Vehicular Ad Hoc Network, VANET),作为未来智能交通系统的基础之一,通过车与车.车与路边节点的相互通信来构成无线通信网络,用于传递辅助驾 ...

4. 微服务全链路跟踪：jaeger集成istio，并兼容uber-trace-id与b3

   微服务全链路跟踪:grpc集成zipkin 微服务全链路跟踪:grpc集成jaeger 微服务全链路跟踪:springcloud集成jaeger 微服务全链路跟踪:jaeger集成istio,并兼容u ...

5. 分布式一致性协议paxos

   Paxos协议/算法是分布式系统中比较重要的协议,它有多重要呢? <分布式系统的事务处理>: Google Chubby的作者Mike Burrows说过这个世界上只有一种一致性算法,那就 ...

6. tcp协议头窗口，滑动窗口，流控制，拥塞控制关系

   tcp协议头窗口,滑动窗口,流控制,拥塞控制关系 参考文章 TCP 的那些事儿(下) http://coolshell.cn/articles/11609.html tcp/ip详解--拥塞控制 &a ...

7. Django--django安装和HTTP协议

   HTTP协议 http协议超文本传输协议(应用层的,基于tcp\ip协议的)特点:无状态,无连接(短连接)请求信息格式GET / HTTP/1.1 请求行 Host: 127.0.0.1:8003 请 ...

8. 【Paper】2015_异构无人机群鲁棒一致性协议设计_孙长银

   原文地址:[1]孙长银,余瑶,张兰.异构无人机群鲁棒一致性协议设计[J].中国科学:技术科学,2015,45(06):573-582. 2015_异构无人机群鲁棒一致性协议设计_孙长银 4 分布式鲁棒 ...

9. TCP/IP之大明内阁---协议的制定

   个人感言:真正的知识是深入浅出的,码农翻身" 公共号将苦涩难懂的计算机知识,用形象有趣的生活中实例呈现给我们,让我们更好地理解.感谢"码农翻身" 公共号,感谢你们的成果, ...

最新文章

1. 2021年春季学期-信号与系统-第五次作业参考答案-第四小题
2. javascript 变量提前
3. pandas 中有关isin（）函数的介绍,python中del解释
4. find_cmd函数分析
5. mix2线刷开发板救砖_小米Mix 2（安卓8.0）一键救砖教程，轻松刷回官方系统
6. Java注解学习一：注解术语
7. mysql的socket文件_修改socket文件， MySQL启动报错
8. 解决Failed to connect to raw.githubusercontent.com port 443的办法
9. Shell脚本学习-阶段四-mysqladmin
10. intellij idea 程序包不可见问题
11. C语言字符型数据scanf,scanf()函数如何输入字符型数据？
12. keil教程——解压缩BCD码
13. Devc++- 源文件未编译”
14. 计算机学业水平测试模拟题,信息技术学业水平测试模拟题库.doc
15. 存储卡的相关标准.CF.SD.SDHC.MMC.TF
16. Java-好玩的猜数小游戏
17. ArcGIS教程：最小值和最大值条形图
18. CY系列菁染料CY3、CY5、CY5.5、CY7、CY7.5修饰标记泛素Ubiquitin Rhodamine（Ub）
19. 计算机的标准输入法,计算机操作系统标准教程 第4章 五笔字型输入法.pdf
20. 【uniapp引入阿里字体图标】

热门文章

1. 您的技术面试是否时间太长了？
2. 关于用硬盘安装工具Win6Ins_v1.2.0.62安装时提示缺少hildr.mbr文件的解决办法
3. 有效的互联网推广有哪些？
4. 怎么实现好看的自定义进度条
5. font-size：625%；
6. mysql truncate 条件_MySQL的TRUNCATE关键字
7. TCL一个雷鸟，或是互联网电视的最后一声惊雷
8. 股市分析软件开发界面设计02
9. 外用药祛除痤疮，对皮肤的影响。
10. 对“判断内六角规格项目”一些改进与完善