Let me try now.
Yep. I got the same resource not found error on us-central1-c.
There is an incident, but does Kubernetes relate to what I am doing? https://status.cloud.google.com//incident/container-engine/18005
Hmm. Seems not.
But anyway, there might be some cascading issues.
I found luck with us-west1-b with p100.
We are not alone: https://www.reddit.com/r/googlecloud/comments/9vae2g/not_enough_resources_available_to_fulfill_the/?utm_source=ifttt
1 Like
Success!
nice tool. thanks!
and if you need more than 50GB for storage? can you connect google cloud storage buckets to selected instance?
Well, actually create worked but now I cannot start instancesā¦ 
Not even nogpusā¦ This is strange. Is this to be expected form time to time?
Oh! I hope Google will fix this pretty soon.
Yeah. Thanks for the script though!
1 Like
Thatās a one way to do it.
But Iād edit the script: https://github.com/arunoda/fastai-shell/blob/master/fastai.sh#L165
You can edit your local copy after the installation at ~/.fastai/bin/fastai
1 Like
the ssd provisioned storage is quite expensive if you need more storage. i think you pay flat montlhy cost regardless if you use 10 or 50GB ?
from what i understand for cloud storge bucket you pay only for actual data stored and at much lower prices. there could be additional cost of transferring the data if you connect to different zone, say vm in europe and bucket in north america.
not sure though if data access speed will be affected and if it will slow down training etc if you use buckets?
1 Like
The speed will be much lower than SSD or standard disk types.
But you try out different options.
Nice work.
How is that differ from editing your instance in the browser?
I could stop my instance - > edit and change the cpu core counts, remove or add gpu and many other changes- > run the instance
This way is still keeping my data without issues.
This is workflow to make it simple and less error prone. All of these can be done via GUI.
In addition to that, this is optimized for preemptible instances. (Create an initial boot disk and create preemptible instances after that when needed)
Also added helper commands like changing availability zone.
Can I ask again? I get that GPU instances can be limited, but shouldnāt I expect to be able to spawn a nogpu instance?
Is it really possible that the region/ zone has not enough instances of nogpu servers?
We are not alone here. See our original GCP thread as well.
us-west2-c has some resources. Use that.
I added a new command to test the availability of the zone.
See: https://github.com/arunoda/fastai-shell/blob/master/fastai.sh#L425
1 Like
Will do. Thanks