How can I scale cloud infrastructure without causing downtime?

Just a heads up, scaling your application is one thing, but if your database can't keep up, you're in trouble. I always start by setting up read replicas and implementing connection pooling. Without those strategies, you'll just end up hitting a wall faster.

Hey! A useful resource could be the AWS Cloud Architect Associate exam prep. It covers a range of scaling topics with examples tailored to different business needs, so you can really dive into solutions that fit your challenges. But honestly, it'd help to hear more about the specific issues you're facing that lead to downtime.

Setting CloudWatch alarms at 80% resource usage to trigger auto scaling is a good baseline. Just keep in mind that CPU usage isn’t the only thing to watch; connection pool limits, queue depths, and memory pressure can lead to user-facing issues before CPU load becomes a problem. There's often a delay between when an alarm triggers and new instances are healthy, usually about 3-5 minutes. So, ensure you factor in that delay into your scaling thresholds. Also, ensure your instances are stateless; anything relying on memory or local writes can cause issues when traffic gets distributed.

It's tough to give a one-size-fits-all answer without more details about your setup. However, I recommend focusing on monitoring your shared resources like databases and queues. Make sure you have a scaling plan in place for these, such as moving blob storage to a service like S3 instead of keeping it in a SQL database. Also, adhering to the 12-factor app principles can help make your services more scalable. A lot of it comes down to making the right architectural choices early on.

One thing we found invaluable is having a clear rollback strategy. Using feature flags makes it easier to revert any scaling changes without needing a full redeploy. Once we began validating changes earlier in our pipeline, rollback situations became far less stressful.