组内 zipkin 分享

06 Sep 2013

在组内做了 zipkin 的分享,下面是分享摘要

Zipkin

Menghan

@menghan412

Outline

  1. Google’s Dapper

  2. Twitter’s Zipkin

  3. Douban’s ???

Dapper

Why Google create launch Dapper?

What is Dapper like?

black-box V.S. annotation-based

Dapper

  1. Low overhead

  2. Application-level transparency

  3. Scalability

  4. Analyse quickly

Dapper design

  1. data structure: trace, span, annotation, key-value annotation

  2. instrumentation points

  3. sample

  4. collection: out of band

Dapper deploy

  1. Dapper daemon

  2. overhead calculation

  3. adaptive sampling

  4. adaptive collection

  5. D-API

Dapper learned

  1. A sample of just one out of thousands of requests provides sufficient information for many common uses of tracing data

  2. integration with exception monitoring

  3. coalescing effects

  4. offline workload

  5. find root cause

  6. record kernel level infomation

Zipkin

http://www.infoq.com/presentations/Zipkin

Zipkin V.S. Dapper

  1. scribe vs daemon

  2. initiative vs passive

  3. cassandra vs BigTable

Zipkin

usage

Douban

???

END

comments powered by Disqus