Things Zipkin isn’t

10m first run of a new talk I'm working on which is the opposite of the usual. Instead of passing cups of koolaid, this will focus on the things Zipkin doesn't do well out of the box, relies on other systems for, or things tracing is generally not good for.

Firstly, this was presented a triple wellington meetup https://www.meetup.com/Cloud-Native-CNCF-Wellington/events/263154909/

Transcript

1. Things Zipkin isn’t
   common misconceptions about distributed tracing
   @adrianfcole
   works at Pivotal
   works on Zipkin
   

   View Slide

2. Zipkin is neat!
   https://twitter.com/zipkinproject
   https://github.com/openzipkin/zipkin
   

   View Slide

3. We use it to solve our own latency problems
   This dotted line is how much latency we took of worst
   request performance!
   

   View Slide

4. Zipkin is not a good one-stop shop!
   @adrianfcole
   #zipkin
   this applies to all non-commercial products I’m aware of
   

   View Slide

5. Zipkin isn’t a replacement for your database
   All the things around Zipkin are built to be lossy. Access to
   data is not managed as well as transactional systems.
   Mixing low-level telemetry apis in business code can be messy.
   Use agents or annotations to track business functions.
   

   View Slide

6. Zipkin isn’t logging
   Imagine if you had to rewrite code to use a niche api in order for
   logs to come out!
   Even if they would, trace data is usually not indexed for full text
   search (oh my!)
   Using correlation fields, you can combine tracing and logging.
   

   View Slide

7. Zipkin isn't metrics
   The core facts of the main Zipkin model are timestamp
   and duration, not statistical significance, not time in
   queue.
   Other things eat Zipkin data though, ex haystack trends
   https://github.com/ExpediaDotCom/haystack-trends
   

   View Slide

8. Zipkin isn't artificial intelligence
   Some commercial products analyse trace data and push
   advice like minify this web app, or change loading order.
   Zipkin can neither surface anomalies nor intuit solutions.
   Some projects are progressing anomaly detection. ex
   adaptive alerting
   https://github.com/ExpediaDotCom/adaptive-alerting
   

   View Slide

9. Zipkin isn't profiling
   Profilers act at stack frame level, a line of code abstraction.
   Most profilers aren’t safe to use in production and/or make
   huge data.
   Zipkin acts on operation or service abstraction, its data
   should be intuitive to those unfamiliar with the code.
   

   View Slide

10. Zipkin can’t do your dishes
    Zipkin focuses on request behaviour. It is good for focusing
    other tools, but implies integration for fancier things.
    Zipkin is not a golden hammer, nor will it ever wash your
    dishes…
    

    View Slide

11. But your dishwasher might use zipkin.
    

    View Slide

12. Thank you!
    @adrianfcole
    #zipkin
    gitter.im/openzipkin/zipkin
    github.com/openzipkin/zipkin
    

    View Slide