Load testing for v1 about voicefill

[jvehent]

Do we need anything more than a curl loop that sends a few dozen opus blobs and/or random data per second?

That's fine. We can just collect a few POST submissions and then replay them. We should do real blobs instead of random so it actually has to process them.

[pdehaan]

I really like the mozilla-services/screenshots approach of having a JSON file with some valid sample data, and we just choose some random payload from a predefined set.

[pdehaan]

Ref: #53

[andrenatal]

@pdehaan You can use this script to save opus files: #53 (comment)

@milescrabill says our throughput is pretty low, and we'll need to scale horizontally until we solve that. Sounds like the ASR server is the choke point.

For launch, we'll be prepared to scale on servers while we collect data on how often users use this service. Stats will be in datadog, although putting them in GA would be 🍠

[chuckharmston]

putting them in GA would be 🍠

Ref #104

@pdehaan is this closable? :)

[pdehaan]

@wresuolc Everything is closeable... if you believe in yourself! 🦄

But yeah, we've done load tests. We'll continue to do more as stacks change and we do more testing. No point/value in keeping this open though, IMO.