4 ms·
This is an awesome post and sounds like it was an awesome experience. I’d be interested to see an alternate-history version of this, in a universe where it was
by rafram 2y ago
This is an awesome post and sounds like it was an awesome experience.
I’d be interested to see an alternate-history version of this, in a universe where it was built with a serverless architecture. My hunch is that a lot of the late-night hacking sessions wouldn’t have been necessary, but when it went down (due to resource caps) it would’ve been down down, not just slow.
- fragmede 2y agoYou'd have a serverless Lamda hit Elasticache/Redis and then still run into scaling issues because the code you wrote wasn't optimized. Given enough Lamdas and no connection pooling, you'll still exhaust Elasticache and have to do something about it. You'll run into the question about how the bandwidth bill is going to look like in the morning and want to figure out a way to not bankrupt yourself before going to sleep. Worse, AWS is known to gouge for bandwidth, and Lamda doesn't give you the same absolute rate limit control as the author's setup did. If you're just gonna give up if/when you hit eg Redis rate limits, that doesn't inspire confidence in serverless.