Videos / DevOps

Episode 5 of 14 · DevOps, end to end

Autoscaling groups: capacity that follows demand

Chapters

1. One number, wrong twice0:00
2. A thermostat, for machines0:16
3. The launch template0:33
4. Minimum, maximum, desired0:51
5. Read, compare, act1:08
6. Capacity is a number1:26
7. The dead bulb1:42
8. Two health checks1:59
9. Liveness and readiness2:19
10. The boot loop2:38
11. Terminate first, or not2:56
12. Replacement is not repair3:12
13. The dial that came with it3:31
14. Waiting is not idle3:48
15. Does it double, or not4:06
16. The honest backlog4:22
17. Latency is a trailing signal4:39
18. Three policies4:56
19. The shower tap5:13
20. What a launch costs5:29
21. The metric is old5:47
22. Thrash6:03
23. Warm-up and cool-down6:19
24. Out fast, in slow6:36
25. Planks of a bridge6:54
26. Change it, nothing happens7:12
27. Surge, or a hole7:30
28. A bad batch stops7:48
29. Rollback is boring8:06
30. Let it drain8:24
31. One chair is bolted8:42
32. Connections that stick9:01
33. A slow boot9:18
34. Straight into the connection cap9:35
35. The loop feeds the fire9:53
36. What it is not10:10
37. The whole thing10:27

Autoscaling groups: capacity that follows demand. Traffic is not flat. It has a morning ramp, an evening peak, and a quiet middle of the night. Fixed capacity gives you two bad choices. Buy for the peak and pay for idle machines a

Mark as watchedSaved in this browser only.