I've had this email exchange with GCP billing support for almost two months now and getting nowhere.
My two projects are built on app engine standard, using VPC access connector to connect our services
to a compute engine instance running Redis. This works well.
When experimenting with CE instances in December, I created and killed some instances in both
projects. I noticed later on the billing statement that I kept accruing costs for these killed instances that
were now gone and out of reach.
After lots of emailing back and forth with billing support, I finally get this answer:
I have a feedback from the specialized team and they advised that using of "VPC access connectors" in managed
instance group which makes some of the instances not visible to your end which is the reason why you are unable
to see the instance that is accruing charge to your account.
This is also related to the mechanism of hiding resources that are managed by some Google system.
I hope this information helps. Should you have additional questions on this matter, feel free to reply, and I'll be more
than happy to address your concerns at that time.
The implication of this is that using the officially supported VPC access connectors can turn killed instances to zombies
that are out of your reach since you already killed them, but they linger deep down in GCP and will show up on your
billing statement every month.
Doesn't sound right, and doesn't sound like Google.
Anyone else experienced this?
Haavard