Hi all,
I have been playing with Nomad Periodic jobs and they are pretty sweet. My only concern is that each job takes about 3 seconds to register and execute a basic Hello World docker container using NodeJS. I’m doing this on Raspberry Pis, so I imagine if this were running on Graviton2-based EC2s the experience might be a little different, but after going down a rabbit-trail trying to understand how others are doing FaaS-like workloads, I’ve yet to find any concrete recommendations on how to proceed.
It looks like the openfaas nomad driver is not under active development, though there is a nomad-pack (which I am unable to figure out how to properly pass variables to, but that’s another post).
I have also read that others are using the FirecrackerVM driver to bring down that scheduling latency, but can’t find any examples that make it easy to grok how to proceed.
I’ve also not seen what kind of function scheduling performance people are seeing with these strategies. For most of our workloads, the slight increase in scheduling latency in exchange for all the benefits Nomad brings is worth it, but we do have some FaaS-style workloads for which a multi-second turn-around time is too high.
I would love to hear about others’ adventures here; what’s working, what’s not possible, and what the best path for a multi-cloud FaaS strategy currently is (i.e. what solutions are portable, as using the new remote-task driver to schedule Lambdas, when released, doesn’t actually help because we would want to run in non-AWS environments too).
Thanks in advance,
Sam