4

The startup-script added to an instance in GCP doesn't seem to run on first boot (but does on subsequent boots)

I'm using the https://www.googleapis.com/compute/v1/projects/centos-cloud/global/images/centos-7-v20160921 base image but had the same issue on a slightly older version of the centos-7 image too.

This happens whether I use the API to create an instance, or the web console.

According to the docs (https://cloud.google.com/compute/docs/startupscript) adding a metadata with the key startup-script and value of a script should cause the script to be run on every boot.

(There is a section about adding the script to a running instance, and it mentions that a reboot is required to get the script to run, but this (apparently) shouldn't apply to new instances.)

It's possible to force the script to run using sudo google_metadata_script_runner --script-type startup (https://cloud.google.com/compute/docs/startupscript#rerunthescript) which implies that the script is loaded, but just not run, so I'm not sure what's going on.

Am I doing something wrong, or have I found a bug in GCP? :)

funkyhat
  • 61
  • 1
  • 8
  • 1
    I was able to reproduce the same behavior on my test account. I would suggest filing a new issue on the public Issue tracker [1] which then will be redirected to Google's engneering team in order to fix it as soon as possible. [1] https://code.google.com/p/google-compute-engine/issues/list – George Oct 04 '16 at 17:36
  • 1
    Thanks, I've opened the issue here: https://code.google.com/p/google-compute-engine/issues/detail?id=433 – funkyhat Oct 07 '16 at 20:27
  • This issue went away by itself, so either google quietly fixed the issue with their image or I was doing something wrong from the start. In any case, I'm no longer able to reproduce it. – funkyhat Feb 06 '17 at 17:16
  • I just ran into this as well. It still seems to be an issue. My google customer account can't seem to view that issue. I've opened a support case with google, but they are suggesting an instance level script, which is a no go for us. We are using a mix of *nix and win environments so the solutions below won't work for us. – nomadic_squirrel Oct 02 '18 at 17:38

2 Answers2

1

This actually worked for me without issues using the Centos 7 image. I created the VM from the console and added this small script

#! /bin/bash
yum update
yum install -y httpd
cat <<EOF > /var/www/html/index.html
<html><body><h1>Hello World</h1>
<p>This page was created from a simple startup script!</p>
</body></html>
EOF

After the VM was created I was able to verify that Apache had been installed in the serial console. 

…
Feb  3 20:36:17 instance-3 startup-script: INFO startup-script: ---> Package httpd.x86_64 0:2.4.6-45.el7.centos will be installed
Feb  3 20:36:17 instance-3 startup-script: INFO startup-script: --> Processing Dependency: httpd-tools = 2.4.6-45.el7.centos for package: httpd-2.4.6-45.el7.centos.x86_64
Feb  3 20:36:17 instance-3 startup-script: INFO startup-script: --> Processing Dependency: /etc/mime.types for package: httpd-2.4.6-45.el7.centos.x86_64
….
Feb  3 20:36:20 instance-3 startup-script: INFO startup-script: Installed:
Feb  3 20:36:20 instance-3 startup-script: INFO startup-script:   httpd.x86_64 0:2.4.6-45.el7.centos
Feb  3 20:36:20 instance-3 startup-script: INFO startup-script: Dependency Installed:
Feb  3 20:36:20 instance-3 startup-script: INFO startup-script:   apr.x86_64 0:1.4.8-3.el7                     apr-util.x86_64 0:1.5.2-6.el7
Feb  3 20:36:20 instance-3 startup-script: INFO startup-script:   httpd-tools.x86_64 0:2.4.6-45.el7.centos     mailcap.noarch 0:2.1.41-2.el7
Feb  3 20:36:20 instance-3 startup-script: INFO startup-script: Complete!

The script did install Apache but the service in the VM was not started. I just had to SSH in the VM to start the service.

Carlos
  • 1,395
  • 9
  • 15
  • That's normal behaviour for RHEL-based distros, my startup script also includes the lines `systemctl enable httpd` `systemctl start httpd` – funkyhat Feb 06 '17 at 17:18
  • Were you able to solve this? Looking at the serial console might provide some hints. – Carlos Feb 06 '17 at 23:21
  • I stopped being able to reproduce the problem. Not sure exactly why and I wasn't keeping particularly good notes when it was occurring. It may have been down to incorrect network setup in the end. – funkyhat Feb 08 '17 at 01:39
1

This is not a bug, it still exists and it is tricky.

Startup scripts are ran after all init.d scripts.

So if you initialise a /etc/environment file for example and try to use it in a init.d script, on the first boot, /etc/environment will not exist when your script will be launched.

On the second boot, there is no more problem, /etc/environment already exists :-)

Potsky
  • 111
  • 1