Speed up application testing with azure container instances
- 1. Anton Boyko Microsoft Azure MVP [email protected] Speed up application testing with Azure Container Instances
- 3. ”I`m allergic to manual labor.”
- 4. ”If you know your server name, it’s your pet. But we are dealing with cattle.”
- 5. Problem: it takes eternity to execute all automated tests
- 6. As is • Builds and tests running on a same machine • 8 CPU, 24 GB RAM • 1500 tests • Near 100% CPU load during test run • Average test duration 3 secs • Total tests run time 75 minutes cspkg
- 7. To be cspkg • Builds and tests running on a different machines • Test agents provisioned on demand; tests executed in parallel • 2 CPU, 4 GB RAM • 1500 tests • Near 100% CPU load during build • Average test duration 3 secs • Expected tests run time 10 minutes
- 8. So what? • We need to deliver more business features, because our clients are paying for them. • You can work on optimization later, but now let’s focus on current release. • As soon as we will hire new folk for our team you will have enough time to work on this. Raise your hand when it becomes familiar:
- 9. 0 50 100 150 200 250 300 350 400 fifth fourth third second first Queue Run Total team idle C = 75 + (75 + 75) + (150 + 75) + (225 + 75) + (300 + 75) ~19 hours per week (10%) of 200 max week’s capacity
- 10. Economy • CPU cost $ 0.0007 per minute $ 0.042 per hour $ 30.24 per month • RAM cost $ 0.0003 per minute $ 0.018 per hour $ 12.96 per month • Team member idle $ 0.5 per minute $ 30 per hour 24/7 (as is) C = (8*CPU*1440*7)+(24*RAM*1440*7)+(75*Idle*3*5) C = 56.45 + 72.58 + 562.5 692 per week (100%) 2768 (100%) Working hours (as is+) C = (8*CPU*720*5)+(24*RAM*720*5)+(75*Idle*3*5) C = 20.16 + 25.92 + 562.5 609 per week (88%) 2436 (88%) On demand + working hours (to be) C = (2*CPU*720*5)+(4*RAM*720*5)+(10*Idle*3*5) + 10*((1*CPU*10*3*5)+(2*RAM*10*3*5)) C = 5.04 + 4.32 + 75 + 10 * (0.11 + 0.09) 87 per week (13%) 248 (13%)
- 11. 0 100 200 300 400 500 600 700 800 24/7 Working hours On demand + working hours Idle RAM CPU Economy • CPU cost $ 0.0007 per minute $ 0.042 per hour $ 30.24 per month • RAM cost $ 0.0003 per minute $ 0.018 per hour $ 12.96 per month • Team member idle $ 0.5 per minute $ 30 per hour
- 12. Expectations • Team of 5 engineers • 200 working hours per week • 50 idle hours per week (focus, communication, etc.) • 19 idle hours per week before optimization • 1 idle hour per week after optimization • 160 hours for optimization project Jan Feb Mar Apr May Jun Func 524 1048 1572 2096 2620 3144 Non-func 0 0 0 0 0 0 Idle 276 552 828 1104 1380 1656 Jan Feb Mar Apr May Jun Func 364 960 1556 2152 2748 3344 Non-func 160 0 0 0 0 0 Idle 276 480 684 888 1092 1296 Before (hours capacity) After (hours capacity)
- 13. Goals • Team idle reduced. New features can be delivered faster than before. • Total cost of ownership reduced. More budget for development, less for maintenance.
- 15. Toolset • Docker (distribution) • Azure Container Registry (artifacts) • Azure Container Instances (hosting) • Azure Storage (test results) • Azure Logic Apps (orchestration)
- 17. Let’s take it even further • Solution application should be as much transparent as possible. - No changes should be made within application codebase. - No changes should be made within tests automation codebase. • Solution should be as much technology agnostic as possible. - We should be able to replicate it for .NET, Java, Python, Node, PHP, etc. - We should be able to replicate it for Azure DevOps, Jenkins, TeamCity, etc.
- 23. Questions?