From 6951a8bbb92c84f08881eddc537cf2a28b0a42ea Mon Sep 17 00:00:00 2001
From: Aarsh2001 <ac5973@srmist.edu.in>
Date: Fri, 6 Oct 2023 20:27:48 +0100
Subject: [PATCH] minor modifications to gcp_auth

---
 .github/auth/vm_auth.py | 54 ++++++++++++++++++++++++++++++-----------
 1 file changed, 40 insertions(+), 14 deletions(-)

diff --git a/.github/auth/vm_auth.py b/.github/auth/vm_auth.py
index 32742f14..9accf1a8 100644
--- a/.github/auth/vm_auth.py
+++ b/.github/auth/vm_auth.py
@@ -9,12 +9,11 @@
 
 def authenticate_vm(path):
     credentials = Credentials.from_service_account_file(path)
-    return discovery.build('compute', 'v1', credentials=credentials)
+    return discovery.build("compute", "v1", credentials=credentials)
 
 
-def _start_ssh_session(compute, creds, username, passphrase):
-    response = compute.instances().get(project="gpu-insatnce", zone='us-central1-a', instance='demos-tests').execute()
-    external_ip = response['networkInterfaces'][0]['accessConfigs'][0]['natIP']
+def _start_ssh_session(response, creds, username, passphrase):
+    external_ip = response["networkInterfaces"][0]["accessConfigs"][0]["natIP"]
 
     ssh = paramiko.SSHClient()
     ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
@@ -31,35 +30,62 @@ def _start_ssh_session(compute, creds, username, passphrase):
     channel = transport.open_session()
 
     # Execute the command on the instance in the background
-    command = 'cd actions-runner; nohup ./run.sh &'
+    command = "cd actions-runner; nohup ./run.sh &"
     channel.exec_command(command)
 
     # Close the SSH channel and session immediately
-    channel.close()
     ssh.close()
 
 
-
-def start_runner(creds, ssh_creds, ssh_user, key_passphrase, id="gpu-insatnce", zone='us-central1-a',
-                 instance='demos-tests'):
+def start_runner(
+    creds,
+    ssh_creds,
+    ssh_user,
+    key_passphrase,
+    id="gpu-insatnce",
+    zone="us-central1-a",
+    instance="demos-tests",
+):
     compute = authenticate_vm(creds)
     request = compute.instances().start(project=id, zone=zone, instance=instance)
     request.execute()
-    time.sleep(60)
 
-    _start_ssh_session(compute, ssh_creds, ssh_user, key_passphrase)
+    max_wait_time = 600
+    wait_interval = 10
+    waited_time = 0
+    response = None
+
+    while waited_time < max_wait_time:
+        response = (
+            compute.instances().get(project=id, zone=zone, instance=instance).execute()
+        )
+        status = response.get("status")
+
+        if status == "RUNNING":
+            break
+
+        time.sleep(wait_interval)
+        waited_time += wait_interval
+
+    if waited_time >= max_wait_time:
+        raise Exception(f"Instance {instance} did not start within the expected time.")
+
+    # Once the instance is running, start the SSH session
+    _start_ssh_session(response, ssh_creds, ssh_user, key_passphrase)
 
 
 def stop_runner(creds):
     compute = authenticate_vm(creds)
-    request = compute.instances().start(project="gpu-insatnce", zone='us-central1-a', instance='demos-tests')
+    request = compute.instances().start(
+        project="gpu-insatnce", zone="us-central1-a", instance="demos-tests"
+    )
     request.execute()
 
 
 if __name__ == "__main__":
     ssh_user, key_passphrase, stop_vm = sys.argv[1], sys.argv[2], sys.argv[3]
-    gcp_credentials = 'gcp_auth.json'
-    ssh_credentials = '~/.ssh/id_rsa'
+    gcp_credentials = "gcp_auth.json"
+    ssh_credentials = "~/.ssh/id_rsa"
 
     if stop_vm == "true":
         # Stop the instance