Automated tooling for a cloud dev environment
- Bring up a dev machine with custom installlation scripts
- SSH to the image
- Auto-Shutdown the image to save money
- Maintain your data across shutdown and resume
- Resume your instance with restored data
- Mount a fast local SSD to /mnt/disks/ssd
./just.sh resume
./just.sh will download and run just if it doesn't exist. You can place just in your PATH and then run just directly.
This command may fail for you: see the Bootstrap and Configuration sections
just ssh
The SSH firewall is set to allow access only to your current ip address and is automatically updated when running commands.
The resume command will also ssh to the instance.
Assumed Dependencies:
- gcloud
- jq
Logged in via gcloud
gcloud auth login
Store Pulumi state in GCP is resilient and works across machines.
pulumi login gs://$(whoami)-pulumi-dev-machine-state
Initialize pulumi state. Storing in the local fs is faster
pulumi login file://state/dev-machine-state
When you first run this, it will complain about missing configurations. You need to set them with
just set KEY VALUE
To see all the configurations, run:
just get config
Save yourself some money when you are done by running
just shutdown
This will preserve your cloud disk. It will copy your local SSD to the cloud disk (this may take some time). When you are ready to work again:
just resume
Currently the system disk is not preserved, but resume will re-run all the installer scripts.
If you don't need your disks, run the following to destroy all infrastructure:
just destroy
There is a cron job that will automatically shut down the instance if it isn't used for an hour (no processes running for your user other than idle sessions).
Don't want to wait so long for Rust to compile things?
just set gcp_machine_type n2-standard-8
just shutdown # if you already have a machine running
just resume
If you are using a bigger instance, try to shut things down when you are finished instead of waiting for auto-shutdown.
just ssh will use Eternal Terminal rather than ssh if et is installed on your machine.
just --list
The machine uses a small boot disk that for now is lost on every shutdown (the resume command will performa re-install at startup). Your $HOME directory is a cloud disk that is saved across shutdowns so your work will persist.
Local SSD is mounted at /mnt/disks/ssd.
/tmp is a symlink to /mnt/disks/ssd/tmp.
So any data written to /tmp will get the benefit of fast local SSD.
However, this data will be lost on shutdown.
To save data, write it to a different folder in /mnt/disks/ssd/
For example:
mkdir -p /mnt/disks/ssd/tikv/target
ln -s /mnt/disks/ssd/tikv/target
just shutdown will run the script shutdown.sh which will move the data to $HOME/ssd/save.
just resume will move the data back to the ssd drive.
TODO: store in GCS
You can login as a user with sudo with:
just ssh-gcloud bash -l
From there you can set the password for your normal user with:
sudo passwd <USER>
Your normal user has sudo access.
Auto-shutdown when not in use