How to automatically start a GCP preemptible VM

How to automatically start a GCP preemptible VM

GCP has a VM called Preemptible VM that stops in up to 24 hours at the cost of a low price. https://cloud.google.com/compute/docs/instances/preemptible?hl=ja

VMs that are stopped after receiving the preempt notification will not start automatically. The following is a description of the tools that enable automatic startup using shutdown scripts and Cloud Functions.

Tool overview

--Preemptible VM is a tool to start because it does not start automatically after stopping. --It is designed to start only when you receive a preempt notification so that it will not start if you stop it manually. --The execution environment is python3.7.

Tool behavior

1. Set shutdown-script on the VM. It runs when the VM is stopped, checks if it has received a preempt notification, writes TRUE or FALSE to a file and uploads it to GCS.
2. Run Cloud Functions triggered by a GCS file update.
3. Function starts the VM only if the preempt notification is TRUE.

Setup steps

1. Download the source code

   git clone [email protected]:cloud-mainte/auto-boot-preemtible-vm.git
   
   or 
   
   git clone https://gitlab.com/cloud-mainte/auto-boot-preemtible-vm.git
   

2. Create a GCS bucket --Create a bucket to place the shutdown script and place the shutdown script. --Rewrite the BUCKET variable in the script as appropriate.

   $ gsutil mb -l ASIA-NORTHEAST1 gs://<backet name>
   $ gsutil cp shutdown-script.sh gs://<backet name>/
   $ gsutil ls -l gs://<backet name>
   

--Create a bucket to store the status of whether or not you received a preempt notification.

```
$ gsutil mb -l ASIA-NORTHEAST1 gs://<backet name>
```

3. Set shutdown-script on the target instance

   gcloud compute instances add-metadata <instance name> \
       --metadata shutdown-script-url=gs://<backet name>/shutdown-script.sh
   

4. Set the instance launcher in Cloud Functions --Rewrite bucket in main.py as appropriate.

   gcloud functions deploy boot_preemptible_instance --region=asia-northeast1 -- 
   entry-point boot_instance --runtime python37 --trigger-event 
   google.storage.object.finalize --trigger-resource gs://<backet name>
   

Operation check

1. Try stopping the instance manually.

$ gcloud compute instances stop $ gcloud compute instances list $ gsutil cat gs:///preempted_status.out {"preempted_status":"FALSE", "instance_name":"", "zone":"asia-northeast1-a", "project":""} ```

If stopped manually, it will be " preempted_status ":" FALSE ".
Check the log.

```
$ gcloud logging read "resource.type=\"cloud_function\" AND resource.labels.function_name=\"boot_preemptible_instance\" AND resource.labels.region=\"asia-northeast1\"" --format="table(timestamp, textPayload)" |head -10
TIMESTAMP                       TEXT_PAYLOAD
2020-08-11T04:38:26.982711475Z  Function execution took 1894 ms, finished with status: 'ok'
2020-08-11T04:38:26.979Z        Preempted status is FALSE.
 :
2020-08-11T04:38:25.090798409Z  Function execution started
```

2. Try to imitate the Preempt notification.

$ cat preempted_status.out
{"preempted_status":"TRUE", "instance_name":"<instance name>", "zone":"asia-northeast1-a", "project":"<project name>"}
$ gsutil cp preempted_status.out gs://<backet name>/
$ gsutil cat gs://<backet name>/preempted_status.out
{"preempted_status":"TRUE", "instance_name":"<instance name>", "zone":"asia-northeast1-a", "project":"<project name>"}

Check the log.

$ gcloud logging read "resource.type=\"cloud_function\" AND resource.labels.function_name=\"boot_preemptible_instance\" AND resource.labels.region=\"asia-northeast1\"" --format="table(timestamp, textPayload)" |head -12
TIMESTAMP                       TEXT_PAYLOAD
2020-08-11T05:02:51.640069088Z  Function execution took 15066 ms, finished with status: 'ok'
2020-08-11T05:02:51.637Z        Instance started.
2020-08-11T05:02:51.637Z        Instance starting.. status = RUNNING
2020-08-11T05:02:46.246Z        Instance starting.. status = STAGING
2020-08-11T05:02:40.862Z        Instance starting.. status = TERMINATED
2020-08-11T05:02:39.373Z        Instance status is now TERMINATED
2020-08-11T05:02:38.959Z        Preempted status is TRUE. Boot up the instance.
 :
2020-08-11T05:02:36.577339221Z  Function execution started

The instance has started.

$ gcloud compute instances list
NAME                         ZONE               MACHINE_TYPE   PREEMPTIBLE  INTERNAL_IP     EXTERNAL_IP    STATUS
<instance name>  asia-northeast1-a  g1-small       true         XX.XX.XX.XX   XX.XX.XX.XXX  RUNNING