Hacker News new | past | comments | ask | show | jobs | submit login
Load testing at scale and lessons learned (kevinlondon.com)
37 points by Kaedon on Jan 20, 2022 | hide | past | favorite | 4 comments



Would have been more interesting article if it described the services affected, what made them slow, what was done to overcome it etc.

Also, the numbers doesn’t seem to reflect a horizontally scalable cluster deployment, the load numbers are low enough for a single server to handle.


Agree. A good structured blog post but it feels like a conceptual scenario more than a real life experience


Very thorough.

> In the face of ambiguity, refuse the temptation to guess.

Great advice, but shouldn't it apply to the test traffic as well? Mimicking traffic patterns and making assumptions has a lot of downsides. You'll always be wrong, though the margin of error will vary. Traffic capture and replay is a great way to ensure load test traffic looks like the real thing.


Traffic capture struggles with speculative testing in general, and can't account for new functionality that doesn't exist yet.

The biggest blocker is that most traffic is highly sensitive to the environment. For example, UUID's and ID's likely don't exist in the development environments where you really want to test ahead of feature release--generally the datasets won't be the same. If they are the same, you'll complicate your life GDPR wise, increase your risk of breaches, and risk sharing PII.

Rather than capture traffic, you can get a similar accuracy, and more flexible setup, by looking at the traffic patterns and allocating your load based on transactions proportionally.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: