Use gcp service-account file to create a GPU instance

[godgunman]

No description provided.

[gary9630]

@godgunman 我查了一下，使用 gcloud compute 的指令，好像不能透過指定 credential 的方式來使用這個 API，必須要在 create VM 的時候就先設定好了，你可以幫我看看我有沒有搞錯這部分

https://stackoverflow.com/questions/35928534/403-request-had-insufficient-authentication-scopes-during-gcloud-container-clu/35932921

https://acloud.guru/forums/gcp-certified-associate-cloud-engineer/discussion/-Lh3ET0aNrv3FwNbNvh6/Insufficient%20Permission:%20Request%20had%20insufficient%20authentication%20scopes.

https://serverfault.com/questions/974012/how-to-start-an-instance-from-another-instance-in-the-same-project-gcp

只要一開始 VM 有開啟 compute engine API，就可以使用了，感覺不需要特別做什麼

[godgunman]

我試過用 gcloud 可以直接開 gpu instance 所以應該是權限足夠的，正名一下好了，好像應該要叫做 service-account

$ gcloud compute instances create ggm-test \
>     --machine-type n1-standard-2 --zone asia-east1-a \
>     --accelerator type=nvidia-tesla-t4,count=1 \
>     --image-family pytorch-latest-gpu --image-project deeplearning-platform-release \
>     --maintenance-policy TERMINATE --restart-on-failure
Created [https://www.googleapis.com/compute/v1/projects/godgunman/zones/asia-east1-a/instances/ggm-test].
NAME      ZONE          MACHINE_TYPE   PREEMPTIBLE  INTERNAL_IP  EXTERNAL_IP   STATUS
ggm-test  asia-east1-a  n1-standard-2               10.140.0.2   a.b.c.d  RUNNING

[gary9630]

Done.

[gary9630]

Refer to: https://github.com/cofacts/rumors-ai-bert/blob/master/CPU_host/run-scripts/main.py#L176