Somebody using `ARM_64` architecture for Lambda? I...
# general
a
Somebody using
ARM_64
architecture for Lambda? I want to hear some feedback, I might want to try it.
a
Sure. I flipped the switch and noticed no difference. It probably only matters if you are using something tied to the architecture, like Golang runtime or a Lambda layer with x86 native code.
a
Yeah, I was thinking something like that.
c
We're using ARM_64 on Golang, it was a bit of a mission to get to work but we eventually got there. I can't say I've noticed a performance difference, but I also haven't checked. I am just blindly trusting the claims in cost/speed improvement
r
Haven't measured but seemed a no-brainer, typically faster for the same price
l
Likewise, switched with no effort, found it to be faster and cheaper 😅
d
As soon as Node 16 was available, we flipped this switch and haven’t noticed much difference (besides lower cost). Might be a touch faster. Do not recommend with Node 14. Lots of really transient, weird issues.
a
OMI, Derek and I disagree! 😆 I haven't moved anything to Node 16 yet - using Node 14 with ARM just fine.
d
🤯
Our dev machines are all ARM as well, so we get lots of touches. Most things work just fine in Node 14, but since it isnt officially supported (mischaracterized, Adam good to call that out) and we ran into a few issues here and there, we stayed safe. Definitely mostly works though.
a
Oh, I did read that Node 16 added support for Apple Silicon. That can't be true of all ARM chips though - Node has been running on Linux ARM for a long time.
d
Tis true, but my experience was that the disconnect between x64 and ARM was enough to cause issues on occasion, and solving them was a REAL pain since getting ahold of a Linux ARM dev machine is…difficult. 😅 This is compared to exactly zero with Node 16 (although we are ARM in dev and in lambdas now, so we still have no disconnect).
I don’t feel like we are doing any super weird stuff, and our packages are always 100% fully updated to the newest major version, so it actually surprises me a bit that we have divergent experiences there. (as it does anytime we diverge, even slightly)
a
Good to know Derek. My ARM Lambdas aren't under load or production, so 🤷 . Production is still x86 and Node 14. Based on your experience, I'll be sure to switch both at the same time.
d
I’m racking my brain to remember if there was a pattern, only thing that comes to mind is that maybe it was lambda at edge that seemed more common…
r
For what it's worth, we ran a tonne of arm lambdas on node 14 without issue
None of them were edgy though
k
Great discussion, we didn't try arm_64 yet in prod either, but will soon give it a go for sure. On the topic of cot savings - we tried Graviton 2 for our Postgres loads a while ago and these also performed well for our use cases. In theory Aurora is also supported, even though we aren't using it and for people with compute intensive requirements there's now a 3rd option as well.