Fork me on GitHub
#aws
<
2016-01-18
>
alexmiller13:01:52

those were pretty interesting. what version of Clojure were you using?

alexmiller13:01:57

I'd be curious to see the timings with 1.8.0-RC5 (or soon 1.8.0) with a lambda compiled with direct linking

ricardo13:01:23

@alexmiller: 1.7, didn’t try with 1.8

ricardo13:01:49

Didn’t think to try, actually, I’ll give it a shot.

alexmiller14:01:59

RC-4/5 improve startup time 10-15% is why I ask

alexmiller14:01:23

and AOT compiling your lambda with direct linking may improve it more

ricardo14:01:25

The class is compiled AOT, but will look into RC5.

alexmiller14:01:42

you can enable direct linking with -Dclojure.compiler.direct-linking=true

ricardo14:01:33

@alexmiller: Will need more thorough testing, but right now averages are about the same with 1.8 RC5.

alexmiller14:01:57

thx, probably dwarfed by other costs then

cfleming21:01:54

@martinklepsch: I saw Apex on HN, it looks very nice.

cfleming21:01:19

Tailing CloudWatch logs is worth the price of admission alone.

cfleming21:01:57

@ricardo: You’re measuring total RTT there from the API client, right?

cfleming21:01:49

Is the API client external to AWS, i.e. you’re not invoking from EC2?

cfleming21:01:04

Also, it looks like you’re not calling any other AWS services, correct?

ricardo22:01:48

@cfleming: It’s the billed duration, actually - I should have clarified that. The rest is correct, as far as the test numbers go. It’s also the time for invoking them directly, not through API Gateway. Full roundtrip through API gateway is a bit erratic - adds between 600ms-1200ms.

ricardo22:01:40

Thanks for bringing it up, going to add a note to the article.

cfleming22:01:01

@ricardo: Great, thanks for the details. Just working up my client now.

cfleming22:01:35

@ricardo: How many iterations did you run for the stable times?

cfleming22:01:56

@ricardo: Also, did you get the billed duration from CloudWatch automatically somehow, or by hand from the logs?