20170329_kubeCon_1.2.2
20170329_kubeCon_1.2.2
2 - Presentations
2.2 - Reworking your teams for k8s (@Yahoo)
Starting point
- Very large on-prem server infra
- Java, Node, many more
- Minimal containerization
- Some “internal cloud”
- “Traditionnal” ops
Yahoo
- Internet player for a long time
- Big player in the industry
- Contributer
- Hadoop
- NodeJS
- …
but NO CLOUD yet …
Path Forward
Context:
Pod (Yahoo style):
* Apps +
* Auth sidecar
* Logging sidecar
* Metrics sidecar
Questions
- Is it worth it ?
- Do we have the skillset to manage it ?
- How much work will it be to adapt my systems ?
- Is it ready for “mass market” ?
Tech Challenges
- Does the App start fast enough to live under orchestration ?
- Are there IP constraints ?
- Limit Cluster access but still make debugging easy ?
How to
Rule 1: Know thyself
* Which App can be moved first
* Resources needs
* Availability needs
Rule2: Build a Team
* Practical on-call experience
+ Everyone: Team Leads, Fronts, back, SRE, …
* Key experts literature
+ Monitoring
+ Logging
* Brown Bag learning
+ Build an internal community
Bumps in the road
- Packages management issues
- OS compatibility
- Set sane DEFAULTS
=> Tune for real world use cases