Google user account impersonates as Service Account vs using Google Groups

Hil Liao
4 min readFeb 8, 2020

I have seen the IAM configuration of bastion hosts where the local root user was configured with gcloud auth activate-service-account by passing the — key-file. The project owner then deleted the key file on the host for security reasons; Google user accounts with service account user role were instructed to log into the bastion hosts and execute sudo su - to run gcloud commands as the service account. There are several downsides to the approach which makes it unsuitable for production:

  1. There is no Cloud Audit logging even Identity and Access Management (IAM) API Audit logs are enabled.
  2. The generated service account key is for 1 time use and is usually lost or leaked after deletion.
  3. Users have to run as root on the bastion host defeating the purpose of having multiple individual users on the host. Each user needs to persist their working directories with some arbitrary names e,g, /root/john-smith/gke-deployment.yaml

What’s the alternative solution to this problem? I’d say Google Groups. However, in an enterprise, asking the security team to create a Google Group and maintain members may take weeks and may not even happen. The Google Group approach means assigning Google user accounts to the Group and grant the group IAM roles to perform allowed tasks. Users would inherently have the IAM roles to execute the gcloud commands and enabled audit logs would generate proper logging.

What if Google Groups approach isn’t feasible? The solution is to use service account impersonation which generates the access token for the impersonated service account; The requires IAM roles are Service Account Token Creator role, Service Usage Consumer role, try the following command to run the gcloud command as the compute engine default service account:

gcloud compute zones list --impersonate-service-account=[project number] --project=[project ID]
# The newer method
gcloud config set auth/impersonate_service_account [project number]

My recently testing shows that Service account user role is not required as it’s intended for users with that role to gcloud compute ssh into an instance (of compute engine or AI platform notebooks) or to attach service accounts to resources.

If the command fails with the error below, verify that your Google account has the following roles:

  1. Service Account Token Creator role in the service account’s project or on the service account. You can verify with gcloud iam service-accounts get-iam-policy $GSA
  2. Service Usage Consumer role in the resource project specified by --project=[project ID]
  3. The project specified by --project=[project ID] which the service account is listing zones for has IAM Service Account Credentials API enabled
ERROR: (gcloud.compute.zones.list) Error 403 (Forbidden) - failed to impersonate [project number] Make sure the account that's trying to impersonate it has access to the service account itself and the "roles/iam.serviceAccountTokenCreator" role

If you still encounter error, append --log-http in the gcloud command and inspect the output. If the http response was redacted, you may need to execute gcloud config set log_http_redact_token false first.

"error": {
"code": 403,
"message": "Caller does not have required permission to use project [project ID]. Grant the caller the Owner or Editor role, or a custom role with the permission, by visiting[project ID] and then retry (propagation of new permission may take a few minutes).",

The next step would be to enable Audit logs for [Identity and Access Management (IAM) API]: Admin Read, Data Read, Data Write:

Enable Audit logs for Identity and Access Management (IAM) API

In Stackdriver logging, select the Service Account resource or use filter protoPayload.serviceName="" protoPayload.methodName="GenerateAccessToken" to inspect the logs:

filtered logs of service account impersonation
Expanding the log entry and identify gcloud.compute.zones.list invocation

The downside is you can’t use environment variable GOOGLE_APPLICATION_CREDENTIALS to run python or Java client applications which require the service account key json file. However, you could use gsutil command with -i [service_account] to access Cloud Storage: gsutil -i storagereader@[project] ls gs://path/. Thanks to Google Cloud free tier technical support to give --log-http hint. Many Thanks to Brent Elphick in for providing the logging filter support.

If the Python code runs in Cloud Function or in Cloud Shell as a user, Granting service account token creator IAM role under the impersonated service account for the user or cloud function’s service account would be enough. Here’s a sample code in cloud function where $impersonated_PROJECT_ID has the impersonated service account. Code also works in Cloud Shell for a user to impersonate as a service account.

import google.auth
import google.auth.impersonated_credentials
from import storage

def hello_world(request):
target_scopes = [""]
creds, pid = google.auth.default()
print(f"Obtained default credentials for the project {pid}")

tcreds = google.auth.impersonated_credentials.Credentials(source_credentials=creds, target_principal="hil-tmp-storage-admin@$", target_scopes=target_scopes)
client = storage.Client(credentials=tcreds)
buckets = client.list_buckets(project="$impersonated_PROJECT_ID")
bucket_names = []
for bucket in buckets:

# print(bucket_names)
return str(bucket_names)