- Newest
- Most votes
- Most comments
You are likely encountering a lag with auto scaling. Your Auto Scaling Group (ASG) is detecting sustained load above your CPU target and begins to "warm up" a new instance to add to your ASG. Depending on a number of factors your load test may be expanding the load faster than your current scaling policy can service that extra load.
You may need to enable and adjust DefaultInstanceWarmup (a setting that lets an instance warm up all the way ready before it is available to serve requests).
Another thing to look at is how your instances are created. If you're doing a lot in the User Data script your instances may not be warming up quickly enough.
Finally, you could also look at your CPU targets. Either trigger scaling earlier in the load profile, or even question your assumption that CPU load is the correct metric to scale on. If your app is IO bound for instance then CPU isn't a good proxy for load, or perhaps number of requests is a better proxy etc.
Once you've got your scaling policy tuned you should see more consistent results. Attaining 100% success guaranteed may need to be traded off against the extra expense of having semi-idle resources available.
Relevant content
- AWS OFFICIALUpdated 9 months ago
- AWS OFFICIALUpdated 2 years ago