Help me understand serverless pricing
So i've been trying the serverless functions to get a understanding on how it works. Been using a cheap one at $0.00016/s which i understand to be execution time + the idle time that you set on the workers. I noticed that while testing my serverless functions, i was seemingly getting billed multiple x's of what i should be based on the execution time in the metrics. Since my tests are only consuming fractions of cents at this point, perhaps theres just a floor im not aware of which causes it to bill more? Maybe some small hidden fees im not aware of? Anywho, based on execution time today i should have spent $0.05 but it currently sits $0.35 in the usage graph, but i have a screenshot from earlier today that shows my im down almost $1.. no clue where the rest of those cents went.
These are just pennies, however this does concern me if i choose to use serverless functions for my business. I want to understand whats consuming the money i've loaded into my account and whether its just penny fees, or whether this magic bill is going to scale with my load. My understanding was that i would only pay for execution time.
Edit: So cold start time is also counted. Still trying to get the math to work out
5 Replies
Yeah so it basically is:
Cold start + execution time + idle time
There is also billing for the first loading of docker image (cause this eats compute across all the workers to download the image) and then this gets cached on Runpod server to prevent redownloading.
If you see notice post-charges that the:
Cold start + execution time + idle time don't line up you can also open a support ticket if you want to deep dive further.
Under here for serverless billing > you can also see what you are charged in compute time
How would i go about reading / understand this cold start time metric? How / where do i see how many minutes were used for cold starts that day. It says total in the top right corner which i assume is just the total cold start for the past 3 days i've been trying runpod?
Also sometimes i get a nice pagianted list of the requests i've executed when i go to the "request" tab, but more often than not its just empty. I dont quite understand where to access the information that allows me to actually double check the billing numbers.
It still doesnt make sense though. I've been billed for 1 hour yesterday, but i had no more than 10 minutes of execution time. The idle time is set to 5 seconds which for 30 spaced out requests is 180 seconds at a price of $0.00016. So roughly $0.15 used in execution time + idle time, which means the last $0.45 would be cold start time? My execution time averaged at 20 seconds, and since this bill is 300% higher, that means the cold start time would have been what.. 1 minute per request?
Surely theres a way for me to see all past requests, their results, the exact cold start time, execution time and idle time no? I have a hard time understanding how to calculate what my business is gonna cost, if i cant access the all the raw numbers for the day, that contributes to the total billing time. I must be missing something here.
@johnson392 thanks for your feedback! I've bought this up to the team, and what you said makes sense.
I'd say for now feel free to contact support and maybe they can give you more insights:
https://contact.runpod.io/hc/en-us/requests/new#new_request_text
Right now it has been raised to the team that under serverless / billing these need to be actual sums and not percentile, and we need to show more transparency under the billing.
A ticket has been raised to the team to work on updating this
Unknown User•3mo ago
Message Not Public
Sign In & Join Server To View
i was just running the serverless sdxl runpod template, trying to make requests to figure out what it was costing when integrating it into my dev build for my business, but after testing for a few days i could not seem to find the stats that was causing the amount i was billed.
Thank you for the help Justin. I dont think i want to rely on support tickets. I will probably continue messing around with runpod and hope for more stats to be made available in the near future. Once i can calculate my own bill from the dashboard, based on the requests that are being made, only then can i consider switching over.
Cold start time seconds
Execution time seconds
Idle delay seconds
+ any other stats thats contributing to serverless billing
I want to be able to see these stats per request in the dashboard. Right now it seems like the requests logs dissapear after a while, ie its only showing the recent requests.