Lambda Scaling: Concurrent Executions Insights & Performance #shorts #quarkus #quarkusinsights

Excerpts from Quarkus Insights #227: Going serverless with Quarkus, GraalVM native images and AWS Lambda

View full episode at https://youtube.com/live/iuvLkNlEinY

Transcript: And then concurrent executions this is interesting. So if you look at monitoring and performance some metrics during J-Fall in the first graph you can see well the basically the concurrent invocations So this is per minute so I think at peak we get I don't know like 340 requests per minute which is a couple of requests per second which is still super low right and I think at max we would say that 1800 people vote at exactly the same time but obviously it's still the relative load is quite low I would say So the next graph is duration and what you see there is interesting So the the minimum duration is well 1 point something millisecond so that's complete handling of requests storing them in the database et cetera The average duration was close to 1.36 milliseconds so also super low and the max duration was well around a second or something So this is for a probably a cold start including scaling to a new Lambda that's still something you can wait on if you're you know pressing vote on an app I would say error count this all went went fine It didn't get throttled And then concurrent executions this is interesting So it start with one and then around 8:00 in the morning it scaled to 17 concurrent instances which is a lot if you have I don't know maybe 300 requests per minute right So I was discussing this with somebody from the Lambda team at a conference and he said "Yeah what do you expect?" So the the the scaler has no data at all about what's going on and then at some point the sessions end and a couple of people start voting at the same time For all they know this can be like I don't know Black Friday and it's we should scale to 1,000 concurrent requests per second or something right So what you see here is probably that the autoscaler got scared a little bit Oh no I need to scale up to 17 and then oh wait it was only a couple of requests and then went back again to one.

Видео Lambda Scaling: Concurrent Executions Insights & Performance #shorts #quarkus #quarkusinsights канала Quarkusio