Should I be getting billed during initialization?
Trying to understand exactly how serverless billing works with respect to workers initialising. From the GUI, behaviour is inconsistent and I can't find an explanation in the docs.
I have an example where workers are pulling a docker image, one of the workers says they're ready despite still pulling the Image while the other two are in the initialising state. The indicator in the bottom right shows the per second pricing for one worker which would make sense if its active, but it clearly isn't ready to accept jobs.
Also, pulling images from Github container registry takes an absolute age, I'd be disappointed about getting charged more because of network congestion.
12 Replies
You are not charged for pulling docker images in serverless, you are only charged for the time the worker is running.
Not sure whats happening with your logs in thje screenshot though, it appears that the worker is running and also pulling a docker image.
Unless its maybe downloading a model and not the docker image, hard to tell since the screenshot has cropped the logs.
Yeah, unfortunately the mobile site crops the log and it jump around with every newline. The worker was marked as running in the UI and my account balance has definitely gone down.
The stage it was in there was extracting the docker image, I am certain it was that and not the model file due to the filesize.
It can be a model, depends on what kind of application you are running but maybe @Justin or @flash-singh need to look into it for you because the worker shoudn't be running and downloading a docker image at the same time.
I've left the endpoint up with 0 max workers so if they want to take a look they can.
Will double check, this is is likely just a UI bug.
Solution
we have seen this happen if you update your container using same tag
Ah, so I should avoid having a
dev tag and pushing a new package while an endpoint has workers then?yes, or use sha tag
Thanks @flash-singh . Is this something you might be looking at changing? I imagine pulling from a common tag is going to be a common trap a lot of users fall into. Plus it makes it difficult to continuously deploy images without adding a step in the workflow to alter the template over the API for example.
its tricky, maybe we build some type of pipeline to detect new tag release and update sha, sha is unique
Fair enough, from a user perspective my expected behaviour would be that if a new image is released on a tag, that an active worker finishe its job and goes back to initialising. It feels like right now there is just a step missing where the worker forgets to mark itself unavailable when it goes to pull a new image.
For the moment though I'm happy to work with specific release tags or sha hashes.
yep its in our backlog to catch this and put some process around it