Map & Territorya story of visibility
View Slide
Pierre-Yves@pyrhttps://github.com/pyr
https://exoscale.ch
Visibility
How do we work ?
How do weimprove?
Avoid Shortcuts!
We want lowerdefect rates
We want to makeinformed decisions
DesignBuildLive
Extracting meaningfulstate data fromheterogeneous eventsources, over time
Meaningful(relates to business value)
State Data(structured payload)
Heterogeneous(everyone is involved)
Over time(tracking)
How does it helpmy system'slifecycle ?
Map=/=Territory
Break out of ourmental model
"I'll push thisminor change, itcannot do anyharm"
"I'll just add thisstatic route"
Better lifecycleInformed decisionsBetter maps
Systems are(increasingly)complex
Web Infrastructurecirca 00(2 servers)
Visibility Circa '00
Web Infrastructurecirca '12(27 nodes)
Visibility Circa '12
Q: how is businessdoing today ?A:
Q: how is businessdoing today ?A: based on thesekey metrics we'relooking good
Figure out thosekey metrics
We needappropriate tooling
events across:system,components,software
The event streamapproach
Plenty of smallproducersFew big consumers
Production:Anything thathappens ormoves (logstoo!):Normalize &Stream
Consumption:AggregateCorrelateDecide
Aggregationcompute compoundmetrics (ratios, sums)
Correlation
Decisiontrack, alert, ignore,scale
Implementingon premise, saas or inbetween ?
SaaSloggly, papertrail,librato, datadog, ...
On Premisecollectd, logstash,graphite, statsd,riemann
The path to visibility:Find key metricsFind the right toolsRely on an event streamInvolve everyoneChallenge your mental modelHopefully, improve quality and lowerdefect rates in the process!
Questions ?