Skip to content
This repository has been archived by the owner on Jan 20, 2024. It is now read-only.

mulesoft-consulting/zipkin-connector

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

73 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

zipkin-connector

Zipkin connector for Mule that allows tracing and logging of requests traversing Mule components. Zipkin is a distributed tracing system. It helps gather timing data needed to troubleshoot latency problems in microservice architectures. It manages both, the collection and lookup of this data. Zipkin server allows for searching and sorting of aggregated spans, and also supports dynamic visualisation of the dependencies graph between logging components. Zipkin’s design is based on the Google Dapper paper.

Overview

Zipkin connector can be used un Mule flows to trace execution of activities, time external callouts, and event logging. Mule tracing can be integrated with tracing in other systems and applications and an end to end performance analysis can be performed. Tracing data can be sent to Zipkin server for analytics and viewing. Zipkin connector integrates with brave Java library to create and manage trace data, and with other Zipkin libraries for connecting to reporting server.

Terminology

For more detailed overview of the philosophy behind Zipkin, refer to Google Dapper paper.

Span

Span is a single measurement. Synchronous (rpc) spans have start timestamp and duration, asynchronous (one-way) spans only have start (point in time) timestamp. Spans have various tags, names, ids, and can link to other spans as parents and children. Parent-child relationship between spans is indicated by parentSpanId = spanId relationship.

Trace

Trace is a collection of spans interlinked between themselves using parentSpanId = spanId relationships. Trace represents an end to end call traced through a number of components and dependencies, each contributing spans to it. Top level span in a trace doesn't have a parentSpanId, but other traces linking to it and each other, have the parentSpanId = spanId relationships.

CLIENT or SERVER?

Traces are tagged differently, depending on whether it is a CLIENT or a SERVER sending or receiving a request. CLIENT indicates first sending a request, and then receiving a response. SERVER indicates the opposite, first, receiving a request, and then, sending the response. When a client sends a request, the trace annotation is timestamped with cs (client send) at the beginning of the trace, and with cr (client receive) at the end of the trace. When a server receives a request at the beginning of the span, it is timestamped with sr (server received) and at the end of the unit of work it is timestamped with ss (server send). The delta between pairs of cs/cr, and sr/ss is calculated as time delta of the span. Note, that asynchronous (one-way) spans only have one stamp - cs or sr, depending on the type of the span being CLIENT or SERVER.

Log message and annotations

Connector can create log messages (log tag) to add logging text to the spans, as well as receive additional tags represented as key/value Map. This allows annotating the spans with custom log information.

Joining external trace

When trace is originated outside of Mule component, it communicates the required parameters needed for Mule spans to be able to join the trace. With HTTP, it will follow the B3 propagation rules, which can be mapped in Mule into the required fields in the connector activity.

Joining internal trace

When Mule creates a span that needs to join with another parent trace also managed by Mule, it is possible to specify the spanId of the active parent span during the creation of the child span. The connector then will retrieve all the required information about the parent span and will join the new child span to it using parentSpanId = spanId relationship.

Synchronous vs. asynchronous spans

Synchronous (rpc) spans need to be started and finished using two different activities in Mule flow. Synchronous spans indicate a time period with start and duration. Asynchronous (one-way) spans only have one timestamp indicating point in time event, and only need to be triggered using one activity.

Use cases and examples

Starting a trace

	<zipkin-logger:create-new-trace
		config-ref="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration"
		logMessage="This is message 1" spanName="myspan1" traceName="mytrace1"
		ServerOrClientSpanType="CLIENT" doc:name="Start sync trace">
	</zipkin-logger:create-new-trace>
	<!-- store SpanData in a flowVar -->
	<set-variable variableName="newSpanId" value="#[payload]"
		doc:name="Save payload with SpanData in a flowVar" />
...Do some stuff...
	<zipkin-logger:finish-span
		config-ref="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration"
		expressionToGetSpanId="#[flowVars.newSpanId.spanId]"
		doc:name="Finish Zipkin span #1" />

Joining externally originated trace

Mapping propagation data from message properties, using message enricher to store SpanData in flowVar and providing additional tags.

	<enricher target="#[variable:spanId1]" doc:name="Message Enricher">
		<zipkin-logger:join-external-span
			config-ref="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration1"
			logMessage="This is log message" spanName="myspan2"
			doc:name="Join external trace" 
			flags="#[message.inboundProperties.'x-b3-flags']"
			parentSpanId="#[message.inboundProperties.'x-b3-parentspanid']"
			sampled="#[message.inboundProperties.'x-b3-sampled']" 
			spanId="#[message.inboundProperties.'x-b3-spanid']"
			traceId="#[message.inboundProperties.'x-b3-traceid']">
			<zipkin-logger:additional-tags>
				<zipkin-logger:additional-tag key="entry">Work</zipkin-logger:additional-tag>
			</zipkin-logger:additional-tags>
		</zipkin-logger:join-external-span>
	</enricher>
...Do some work...
	<zipkin-logger:finish-span
		config-ref="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration1"
		expressionToGetSpanId="#[flowVars.spanId1.spanId]" doc:name="Finish Zipkin span #2" />

Joining internally originated trace

Passing parentSpanId as a reference to be linked to the current span as a parent.

	<enricher target="#[variable:spanId2]" doc:name="Message Enricher">
		<zipkin-logger:join-span
			config-ref="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration2"
			logMessage="Boom boom boom" parentSpanId="#[flowVars.spanId1.spanId]"
			spanName="myspan3" ServerOrClientSpanType="CLIENT" doc:name="Join internal trace">
			<zipkin-logger:additional-tags>
				<zipkin-logger:additional-tag key="Entry">jjj</zipkin-logger:additional-tag>
			</zipkin-logger:additional-tags>
		</zipkin-logger:join-span>
	</enricher>
...Do some work...
	<zipkin-logger:finish-span
		config-ref="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration2"
		expressionToGetSpanId="#[flowVars.spanId2.spanId]" doc:name="Finish Zipkin Span #3" />

Joining a trace with asynchronous trace

Just one activity without finish-span element. Similarly, it is possible to start new asynchronous trace, and join an externally originated trace (with X-B3 properties mapping).

	<zipkin-logger:join-span-async
		config-ref="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration3"
		logMessage="Async server" parentSpanId="#[flowVars.spanId2.spanId]"
		spanName="myspan4" doc:name="Async server span">
		<zipkin-logger:additional-tags>
			<zipkin-logger:additional-tag key="Entry">call</zipkin-logger:additional-tag>
		</zipkin-logger:additional-tags>

Sending trace propagation data to HTTP endpoint

Note the mapping of spanId into x-b3-parentspanId that indicates top level span in a trace mapping.

	<!-- Map parent span details propagation into X-B3 HTTP headers -->
	<message-properties-transformer doc:name="Populate message headers with span data">
		<add-message-property key="x-b3-sampled" value="#[payload.sampled]" />
		<add-message-property key="x-b3-traceid" value="#[payload.traceId]" />
		<add-message-property key="x-b3-spanid" value="#[payload.spanId]" />
		<add-message-property key="x-b3-parentspanId" value="#[payload.spanId]" />
	</message-properties-transformer>
	<http:request config-ref="HTTP_Request_Configuration1"
		path="/" method="GET" followRedirects="false" parseResponse="false"
		doc:name="HTTP client" />

Configuration

Currently, there are two configuration options available: console and HTTP. Console option is only to be used for debugging purposes and not for production. Configuration also contains service name, so if the same Mule component implements multiple services, create multiple configurations, one per service. Examples:

	<zipkin-logger:console-config
		name="Zipkin_Logger__Zipkin_Console_Logging_Configuration"
		serviceName="my-service-1" doc:name="Zipkin Logger: Zipkin Console Logging Configuration" />
	<zipkin-logger:http-config
		name="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration2" serviceName="my-service-2"
		doc:name="Zipkin Logger: Zipkin HTTP Logging Configuration"
		zipkinUrl="http://127.0.0.1:9411/api/v1/spans"/>
	<zipkin-logger:http-config
		name="Zipkin_Logger__Zipkin_HTTP_Logging_Configuration1" serviceName="my-service-1"
		doc:name="Zipkin Logger: Zipkin HTTP Logging Configuration"
		zipkinUrl="http://127.0.0.1:9411/api/v1/spans"/>

Installation

Mule connector

Download https://github.com/mulesoft-consulting/zipkin-connector/tree/master/zipkin-logger-connector/target/UpdateSite.zip Then, point to the directory from Anypoint Studio, go to Help → Install New Software and add Zipkin Logger Update Site using the Archive button to point at downloaded UpdateSite.zip. This will allow you to find the version to install.

Zipkin server

Described in details here.

More information

TO DO

  • Add support for zipkin-opentracing bridge.
  • Investigate and add support for integration with log aggregators, such as Splunk, Loggly, etc.

About

Zipkin connector for Mule

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages