Cost calculation for serverless

[misty surge]

if this is my request, then how much am i billed: ( suppose i'm using the 24gb GPU )
{
"delayTime": 5981,
"executionTime": 7138,
.....

[misty surge]

is it only billed for the "executionTime": 7138,?

[stable sinew]

Cold start + execution. If your worker performs actions such as download models before the handler is called you will also be billed for this time.

[misty surge]

Oh OK but which is the cold start time

Is it included in the execution time?

[stable sinew]

No, it is prior the execution time. Not something that can easily be reported for a specific request. There should be a cold start time chart on your endpoint page

[misty surge]

Oh

Yaya ok thanks ill check that

hmm is there a guide to estimate my costs

also i've seen my execution time for the first time after the worker is launched is taking way longer like 40 secs ++ than usual why is that

then it became faster like 6 secs

[stable sinew]

May you shareyour handler file/code?

[misty surge]

im using the sd webui, the handler waits for webui service like the ashleyk's one

i modified it abit, to wait sd webui only if using sd's apis because there is some other helpers api to download file and upload file

[stable sinew]

Sounds like it is these initilizations that run on the first request that are taking a while and then once they are ran everything else after is faster. If you can move this step outside of the handler then your execution times should become more consistent.

[misty surge]

so its counted as the "cold start time"?

[stable sinew]

Correct

[misty surge]

ah okay but it doesnt really matter right

ill just keep it this way lol, also thanks for the info

[sudden kernel]

Does the container image fetch onto the host where the pod will run get counted as billable time?