2. Nike
Digital
Engineering
Nike
Digital
Engineering
ADAM
NUTT
✦ Engineering Manager, Nike Digital
✦ 15+ years experience, including government, video
games, and startups
✦ 3 years with Nike, all with AWS infrastructure
DEMOND
JACKSON
✦ Lead SRE, Nike Digital
✦ 18+ years experience, mostly large enterprise w/
some consulting
✦ 3 years with Nike, primarily AWS infrastructure
9. Nike
Digital
Engineering
OUR NEEDS
We need to know there is an error quickly.
Seconds matter during brand events
SPEED
RESOLUTION
Our volume means minute issues only
become discoverable when we have all
the data
SCALE
We have a large, ephemeral
infrastructure. We need to monitor all of
it, easily
ANALYTICS
Metrics become significantly more powerful
when the same metric can be used in
different ways by applying live
transformations
13. Nike
Digital
Engineering
WHAT DO THESE NUMBERS MEAN?
ARE THEY GOOD OR BAD? DO WE
HAVE A PROBLEM?
AVERAGES CAN SMOOTH OUT ERRORS.
THEY CAN ALSO MAKE AN INSTANT
EVENT SEEM LONG LASTING
KNOWING IS BETTER THAN
NOT KNOWING
METRICS = WINNING
GAVE CONFIDENCE
DIFFICULT TO PARSE
CRAWL-WALK-RUN
CRAWL
15. Nike
Digital
Engineering
LEARN WHAT MATTERS. THEN
MEASURE THAT
HAVING THE PERTINENT PARTS OF
THE USER REQUEST BROKEN DOWN
IN SENSIBLE WAYS ALLOWS US TO
TRIAGE IN REAL TIME
OBSERVABILITY MEANS FINDING
ANSWERS TO QUESTIONS YOU DIDN’T
KNOW TO ASK
NEW QUESTIONS CAN
BE ASKED
EASILY UNDERSTOOD
KEY QUESTIONS CAN
BE ANSWERED
CRAWL-WALK-RUN
WALK
18. Nike
Digital
Engineering
INSPIRED BY MICHELLE AND
TWITTER
INFORMS OUR FUTURE
BUSINESS DECISIONS.
ALSO FUN TO WATCH IN REAL
TIME
SHOES* PER
SECOND IS A NIKE
LAUNCH METRIC
MEASURES
WHAT MATTERS
CUSTOM KPIS
IMPORTANT FOR
OUR LINE OF
BUSINESS
CRAWL-WALK-RUN
RUN
22. Nike
Digital
Engineering
EASY TO IMPLEMENT. BIG WIN
FOR THE DEV TEAM
NIKE WORKED CLOSELY WITH
SIGNALFX DURING BOTH THE
ALPHA AND THE BETA
SIMPLIFY YOUR DEVOPS
ALIGNMENT ACROSS MULTIPLE
TEAMS
SMART GATEWAY
WITH TAIL-BASED
SAMPLING
MAP OUT YOUR
GEOMETRY,
INCLUDING
SERVERLESS
QUICKLY
INVESTIGATE ERRORS
AND DETERMINE
THEIR TRUE SOURCE
MICROSERVICES APPLICATION PERFORMANCE MANAGEMENT
FLY
23. Nike
Digital
Engineering
BE CAREFUL NOT TO
OVER INSTRUMENT
1
One service reported a single
success or failure metric and used
tags to qualify the success or failure
✦ Only report what you care about
✦ Use metric buckets
HAVING CUSTOM METRICS
LEADS TO BETTER METRICS
Our initial efforts concentrated on
reporting numbers we thought were
important. After observing the initial
metrics we were able to derive a
better set of metrics
✦ Start with what you think you know
✦ Use that knowledge to build better
metrics
THE RIGHT METRICS ARE
EASY TO IMPLEMENT
Most of our metrics are single lines of
code that expose values that already
exist. Writing many lines of code for a
single metric can signal trouble
✦ Metrics should be easy to expose
✦ Use SignalFx’s domain knowledge
to build better metrics
2 3
LESSONS LEARNED