#I don't know my serverless balance goes down

1 messages · Page 1 of 1 (latest)

unkempt crown
#

Hi, I recently made some changes to my platform. It analyzes videos using 3 different computer vision models.

I have a serverless endpoint for each. I think that somewhere I am making some requests that I should not be making, or that some endpoints are active when they should not.

For example, this happened from 14:20CET to 15:15CET today. I had $85.443 and when I came back I had $83.934. I was doing other tests on my app during this time, but I wasn't calling any endpoint here.

I've lost quite a bit of money because of this now. Could someone help me find or debug this? I am not able to replicate this balance drop.

dry valve
#

Hey, Maybe check on billings > serverless

unkempt crown
# dry valve Hey, Maybe check on billings > serverless

This is the endpoint that has abnormally high usage costs in the last 2 days is: t94jq90tvw8v0u (image A)

These are its usage metrics, they seem normal (Image B)

These are the logs during the time where I got abnormally high costs, nothing shows (image C)

I noticed that a worker kept running for a long time after a test run that I recently did. It showed an error message somewhere, related to some daemon thing.

Maybe that's what makes these costs high, but I don't know since nothing else appears on these logs and I didn't get the error again.

What can I do? I enabled execution timeout but I don't know if it will solve it: Here are the endpoint settings (Image D)

image_A.png image_B.png image_C.png image_D.png
dry valve
#

Okay what error Is it can you copy paste that one and maybe if it's from your code search your code for relevant thing

#

Or maybe let's open a ticket so staffs can check

slim coralBOT
dry valve
#

Send your endpoint ID(s)

unkempt crown
# dry valve Send your endpoint ID(s)

Hi, I opened the issue on zendesk too and I think that we solved it. But I have another question.

I want to process a video using an AI model. In order to make this quicker, I separate the video in different clips and send each clip to a worker to process the clips in parallel

So, for example, if I separate the video in 10 clips and send them to 10 workers, is it more expensive than sending it to 5 workers?

dry valve
#
  1. More Workers (10 workers):

Advantage: Processes all clips in parallel, potentially faster if workers are ready.

Disadvantage: Higher costs due to model loading time charged for each worker.

  1. Fewer Workers (5 workers):

Advantage: Workers process clips sequentially but reuse the loaded model, minimizing setup costs.

Disadvantage: Slightly slower because fewer workers handle the clips.

  1. Constant Requests Scenario:

If workers remain active (model stays loaded), using more workers is faster but not significantly more expensive. This is efficient for high throughput.

Recommendation:

For occasional jobs: Use fewer workers to reduce setup costs.

For frequent jobs: Use more workers to maximize speed and maintain efficiency.
@unkempt crown

weary talon
#

@unkempt crown how did u end up fixing the issue

#

I’m getting essentially the same thing

dry valve
#

Then it's best to open a support ticket too

#

Check your thread