Upgrade to Pro — share decks privately, control downloads, hide ads and more …

DevOps for Startups

Armon Dadgar
January 10, 2017

DevOps for Startups

Most startups are in a race, either against competitors or burn rates to deliver products. The focus on agility of delivery is a perfect fit for the goals of DevOps, which optimizes how quickly an application can be delivered. In this talk, we discuss what DevOps is, and how it affects the people, processes, and tools used to deliver an app. Startups should adopt many of the practices of DevOps, and we will talk about some pragmatic ways to get started and add value without getting caught up with all the shiny new tools.

Armon Dadgar

January 10, 2017
Tweet

More Decks by Armon Dadgar

Other Decks in Technology

Transcript

  1. DevOps Definitions • “DevOps is you have developers do everything”

    • “DevOps is you get rid of operations” • “DevOps is a cultural movement” • “DevOps is …”
  2. Delivering an Application • Software organization is a system like

    any other • Composed of people, processes, and tools • Processes used to organize people • Tools used to support people and process • Output is applications
  3. People  Programming Languages Frameworks Design Patterns Application Architecture …

    Developer  Threat Modeling Cryptography Security Patterns Compliance … Security  Cloud APIs System Administration Infrastructure Architecture Networking … Operator
  4. Unicorn Developers • “Developers should do it all!” • Unicorns

    are in short supply, not a good business decision • Specialization of Knowledge is real • Some knowledge can be outsourced, still exists!
  5. Amdahl’s Law • The theoretical throughput of a system is

    limited by serial latency • Organization is a system that is creating an application • Output is limited by serial coordination • Empowering individuals to work independently improves throughput
  6. Fundamental Steps • Write the application • Test the application

    • Package for staging / production • Provisioning infrastructure resources • Deploying an application to the infrastructure • Monitoring applications and infrastructure • Securing applications and infrastructure
  7. DevOps Defined • Process to fulfill the fundamental steps optimizing

    for throughput • Reduce coordination, empower individuals, focus on delivery time • Use tools to coordinate between steps instead of people • Clean separation of responsibilities
  8. DevOps Process  Developer  Security  Operator Write and

    Test Consume Secrets Deploy Monitor Apps Automate Packaging Provision Infrastructure Provide Deployment Tools Monitor Infrastructure Model Organization Manage Secrets Delegate Access Compliance
  9. Caveats • With great power, comes great responsibility! • Developers

    become owners of application • More discipline around testing • Requires investment in tooling and education
  10. Ask your Doctor • Every process makes assumptions and optimizes

    for different metrics • DevOps optimizes for agility, assumes cost of mistake is low and risk tolerance is high • Waterfall optimizes for risk management, assumes cost of mistake is high and risk tolerance is low • Avionics software poorly suited for DevOps • Very high cost of mistakes, very low risk tolerance, low iteration speed
  11. Scaling down DevOps • That sounded very Enterprise-y • Startups

    have (many) fewer people • Impacts process and tools
  12. Startup Anatomy • Fewer specializations and less teams • Unlikely

    to have dedicated QA, Security, Compliance, etc • In the early days, may not have any operators
  13. Startup Constraints • Burn rate, it’s a race against the

    clock! • Focus on core product, everything else is a cost center • Outsource where possible • High risk tolerance, default is failure
  14. Doing DevOps • Usually happens naturally because of lack of

    specialization! • All developers empowered to deploy in the early days • As you start specializing, stay conscious of the delivery process
  15. Pragmatism • Build for 1x, Design for 10x, Plan for

    100x • You are not Google, nor will you be next year • Business and product may change, reduce sunk costs • Stay flexible to change, without building into a cul-de-sac
  16. Provisioning • Pick a cloud! • Large credits available to

    incentivize usage ($10K-$100K+) • Leverage the expertise of your people • Make it easy to spin up multiple environments (prod, stage, dev) • Terraform, Infrastructure-as-Code, etc
  17. Security • Focus on the low hanging fruit • Enable

    2FA everywhere • Build a network perimeter (private network + bastion host) • Avoid secrets / credentials in code (Vault) • Encrypt sensitive data (Vault) • Use security monitoring (evident.io)
  18. Runtime • Focus on developer productivity • Cost is likely

    a red herring relative to payroll • Assume ~0 operators • Outsource logging (Cloud), metrics (NewRelic, DataDog), exception tracking (Sentry), alerting (PagerDuty) • Leverage platforms like ECS, Nomad, K8S
  19. Note on Schedulers • Schedulers are fantastic, but not silver

    bullets • Complex software has complex failure modes • Keep It Simple Stupid • Dedicated operator almost a requirement for more advanced systems • Ask: Why do we need it?
  20. Starting with Segment Stack • Segment is a streaming analytics

    startup • Published their full AWS stack configuration • https://github.com/segmentio/stack • Leverages Terraform, AWS, Docker, and ECS
  21. Segment Stack Features • Secure networking by default • Basic

    auto-scaling • Deployment handled by ECS • Uses CloudWatch for logging and metrics • Up and running in 10 minutes
  22. Growing into scale • If all goes well, the startup

    will grow • More people eventually forces a specialization of knowledge • Starts to look more like the Enterprise-y process • Allows for more sophistication if done right
  23. Dedicated Operations • Owns the Infrastructure / Security / Runtime

    core • Provide a platform to developers (write, test, deploy, monitor) • Evaluate fancier schedulers (Nomad, K8S, Swarm) • Richer tooling (deployment, observability, tracing, etc) • Performance of the infrastructure • Blue/Green, shadow traffic, enable better testing rigor
  24. Dedicated Security Team • Reduce the surface area of access

    • Locking down SSH access • Mutual TLS for services • Data privacy • Compliance
  25. Splitting Development Teams • Dividing the application into services owned

    by teams • Reduces coordination between teams, increases operational demand • Need better deployment and observability tooling • More disciple around testing required, more moving pieces
  26. DevOps for Startups • DevOps is a process focused on

    agility, aligns with constraints of startups! • Clouds and modern tools give you a huge amount of leverage • Avoid BIY, almost always a cost center, doesn’t add product value • Go forth and build!