Scaling Data Engineering

Talk at Big Data Spain, Madrid.

https://www.bigdataspain.org/program/thu-scaling-data-engineering.html

Transcript

1. © 2016 Mesosphere, Inc. All Rights Reserved. SCALING DATA ENGINEERING 1 Michael Hausenblas | 2016-11-17 | Big Data Spain, Madrid
2. © 2015 Mesosphere, Inc. All Rights Reserved. MOTIVATION 2
3. © 2015 Mesosphere, Inc. All Rights Reserved. AIRLINES 3
4. © 2015 Mesosphere, Inc. All Rights Reserved. LOGISTICS 4
5. © 2015 Mesosphere, Inc. All Rights Reserved. HEALTH
 CARE 5
6. © 2015 Mesosphere, Inc. All Rights Reserved. TRADERS 6
7. © 2015 Mesosphere, Inc. All Rights Reserved. FARMERS 7
8. © 2015 Mesosphere, Inc. All Rights Reserved. CITIES 8 © 2014, Wired magazine
9. © 2015 Mesosphere, Inc. All Rights Reserved. YOU 9
10. © 2015 Mesosphere, Inc. All Rights Reserved. TRENDS,
 OPPORTUNITIES, AND CHALLENGES 10
11. © 2016 Mesosphere, Inc. All Rights Reserved. ACTIONABLE INSIGHTS,
 ANYONE? 11
12. © 2016 Mesosphere, Inc. All Rights Reserved. IMPEDANCE MISMATCH 12
13. © 2016 Mesosphere, Inc. All Rights Reserved. TOWARDS 100% 
 SELF SERVICE 13
14. © 2016 Mesosphere, Inc. All Rights Reserved. CLOUD, CONTAINERS
 AND DEVOPS 14
15. © 2015 Mesosphere, Inc. All Rights Reserved. FAST AND BIG DATA … 15 *) kudos to Timothy St. Clair, @timothysc batch streaming PaaS MapReduce
16. © 2015 Mesosphere, Inc. All Rights Reserved. CHALLENGES 16 • Set up and operation of data pipeline components • Dealing with back-pressure: elasticity (static vs. dynamic partitioning) • Efficient usage of resources (utilization/TCO)
17. © 2015 Mesosphere, Inc. All Rights Reserved. TOOLING 17
18. © 2015 Mesosphere, Inc. All Rights Reserved. • Apache Kafka • ØMQ, RabbitMQ, Disque (Redis-based), etc. • fluentd, Logstash, Flume • Akka streams • cloud-only: AWS SQS, Google Cloud Pub/Sub • see also queues.io MESSAGE QUEUES & ROUTERS 18
19. © 2015 Mesosphere, Inc. All Rights Reserved. STREAM PROCESSING PLATFORMS 19 • Apache Storm • Apache Spark • Apache Samza • Apache Flink • Concord • cloud-only: AWS Kinesis, Google Cloud Dataflow • see also my webinar on stream processing
20. © 2015 Mesosphere, Inc. All Rights Reserved. TIME SERIES DATASTORES 20 • InfluxDB • OpenTSDB • KairosDB • Prometheus
21. © 2016 Mesosphere, Inc. All Rights Reserved. TIME FOR A NEW KIND OF OPERATING SYSTEM 21
22. © 2016 Mesosphere, Inc. All Rights Reserved. SINGLE MACHINE APPLICATION 22 hardware OS app
23. © 2016 Mesosphere, Inc. All Rights Reserved. DISTRIBUTED APPLICATION 23 hardware OS app hardware OS hardware OS hardware OS hardware OS hardware OS hardware OS
24. © 2016 Mesosphere, Inc. All Rights Reserved. DISTRIBUTED OS + DISTRIBUTED APP 24 hardware OS app hardware OS hardware OS hardware OS hardware OS hardware OS hardware OS distributed OS
25. © 2016 Mesosphere, Inc. All Rights Reserved. LOCAL OS
 VS
 DISTRIBUTED OS 25
26. © 2016 Mesosphere, Inc. All Rights Reserved. DC/OS ARCHITECTURE 26
27. © 2016 Mesosphere, Inc. All Rights Reserved. DC/OS BENEFITS 27 • One cluster for • stateless services such as Web servers & app servers • stateful services like PostgreSQL, MemSQL, Kafka, Cassandra, etc. • elastic data processing via Spark, Storm/Heron, Akka, etc. • CI/CD, for example Jenkins/Marathon • Dynamic partitioning of your cluster, depending on your needs • Increased utilization (10% → 80%+)
28. © 2016 Mesosphere, Inc. All Rights Reserved. BUT WHAT ABOUT HUMANS? 28
29. © 2016 Mesosphere, Inc. All Rights Reserved. 29 appops The person who writes an app is also the person responsible for operating the app in prod.
30. © 2016 Mesosphere, Inc. All Rights Reserved. 30 It's not about provisioning
 a VM or installing a DC/OS cluster or replacing a faulty HDD …
 
 … this would be on the infrastructure team. appops
31. © 2016 Mesosphere, Inc. All Rights Reserved. 31 human fault tolerance UX matters! protect people from themselves
32. © 2016 Mesosphere, Inc. All Rights Reserved. DEMO TIME! 32
33. © 2015 Mesosphere, Inc. All Rights Reserved. 33 A SIMPLE DATA PIPELINE https://api.github.com/orgs/$ORG/events
34. © 2015 Mesosphere, Inc. All Rights Reserved. 34 A SIMPLE DATA PIPELINE $ dcos package install marathon-lb $ dcos package install --options=config.json influxdb $ dcos package install grafana $ dcos marathon app add fetcher.json $ curl fetcher.marathon.l4lb.thisdcos.directory:80/start
35. © 2016 Mesosphere, Inc. All Rights Reserved. TAKE HOME MESSAGES 35 • Try to have short feedback loops • Containers and 'The Cloud' make deployment easy, leverage it! • Technology is the simple part of the solution:
 big data technologies won't fix your broken culture
36. © 2016 Mesosphere, Inc. All Rights Reserved. Q & A 36 • @mhausenblas • mhausenblas.info • michael@dcos.io https://dcos.io