DevopsConf 2015

Devops desde la trinchera

Transcript

1. DevOps desde la trinchera Matías De Carli
2. • +700.000 registered users • 15k/20k monthly users • +150 countries Product • Node.js app, multiple Clients • MongoDB • Microsoft Azure • Lots of sockets Tech
3. Code => Github => Jenkins => Azure BEFORE
4. Problems • Downtime on deploys • Random Downtimes • Latency • Node on Windows
5. Requirements • Azure • No DEIS • No MESOS • Scale Down • Fast & Easy Going • Low Footprint • Zero downtime deploys • FIVE NINes SLA (99.999%) • easy Scaling
7. • Low footprint • No agent needed on host • Visibility trough Tower • Fast • SSH (Password or Key) • YML
8. AFTER • 2 Load Balancers (HA) • HAPROXY ON LB • Internal Network • No routing on hosts • Rolling deployment • Sticky Sessions on LB
9. Host selection: Web Turning off the old container, creating a new one Post-tasks: Smoke test the app Pre-Tasks: Removing host from the HAProxy set
10. Tag deployments on New Relic & Roller Host Selection: LB & Restart HAProxy Putting back the node on the LB set Host Selection: MGMT
11. “Measure what is measurable, and make measurable what is not so.” -Galileo GalileI “You Can’t Improve What You Can’t Measure.” -Agustin Soler
12. Visibility: Ansible Tower
13. Monitoring: New Relic (APM)
14. Monitoring: New Relic (VM)
15. Papertrail (logs) Rollbar (app errors)
16. Matías De Carli @matiasdecarli www.matiasdecarli.com