Packages for Google Compute Engine Linux images.
Python Shell C++ Makefile

README.md

Linux Guest Environment for Google Compute Engine

Build Status codecov

This repository stores the collection of packages installed on Google supported Compute Engine images.

Table of Contents

Background

The Linux guest environment denotes the Google provided configuration and tooling inside of a Google Compute Engine (GCE) virtual machine. The metadata server is a communication channel for transferring information from a client into the guest. The Linux guest environment includes a set of scripts and daemons (long-running processes) that read the content of the metadata server to make a virtual machine run properly on our platform.

Guest Overview

The guest environment is made up of the following components:

  • Accounts daemon to setup and manage user accounts, and to enable SSH key based authentication.
  • Clock skew daemon to keep the system clock in sync after VM start and stop events.
  • Disk expand scripts to expand the VM root partition for CentOS 6, CentOS 7, RHEL 6, and RHEL 7 images.
  • Instance setup scripts to execute VM configuration scripts during boot.
  • IP forwarding daemon that integrates network load balancing with forwarding rule changes into the guest.
  • Metadata scripts to run user-provided scripts at VM startup and shutdown.
  • Network setup service to enable multiple network interfaces on boot.

The Linux guest environment is written in Python and is version agnostic between Python 2.6 and 3.5. There is complete unittest coverage for every Python library and script. The design of various guest libraries, daemons, and scripts, are detailed in the sections below.

Common Libraries

The Python libraries are shared with each of the daemons and the instance setup tools.

Metadata Watcher

The guest environment relies upon retrieving content from the metadata server to configure the VM environment. A metadata watching library handles all communication with the metadata server.

The library exposes two functions:

  • GetMetadata immediately retrieves the contents of the metadata server for a given metadata key. The function catches and logs any connection related exceptions. The metadata server content is returned as a deserialized JSON object.
  • WatchMetadata continuously makes a hanging GET, watching for changes to the specified contents of the metadata server. When the request closes, the watcher verifies the etag was updated. In case of an update, the etag is updated and a provided handler function is called with the deserialized JSON metadata content. The WatchMetadata function should never terminate; it catches and logs any connection related exceptions, and catches and logs any exception generated from calling the handler.

Metadata server requests have custom retry logic for metadata server unavailability; by default, any request has one minute to complete before the request is cancelled. In case of a brief network outage where the metadata server is unavailable, there is a short delay between retries.

Logging

The Google added daemons and scripts write to the serial port for added transparency. A common logging library is a thin wrapper around the Python logging module. The library configures appropriate SysLog handlers, sets the logging formatter, and provides a debug options for added logging and console output.

Configuration Management

A configuration file allows users to disable daemons and modify instance setup behaviors from a single location. Guest environment daemons and scripts need a mechanism to integrate user settings into the guest. A configuration management library retrieves and modifies these settings.

The library exposes the following functions:

  • GetOptionString retrieves the value for a configuration option. The type of the value is a string if set.
  • GetOptionBool retrieves the value for a configuration option. The type of the value is a boolean if set.
  • SetOption sets the value of an option in the config file. An overwrite flag specifies whether to replace an existing value.
  • WriteConfig writes the configuration values to a file. The function is responsible for locking the file, preventing concurrent writes, and writing a file header if one is provided.

File Management

Guest environment daemons and scripts use a common library for file management. The library provides the following functions:

  • SetPermissions unifies the logic to set permissions and simplify file creation across the various Linux distributions. The function sets the mode, UID, and GID, of a provided path. On supported OS configurations that user SELinux, the SELinux context is automatically set.
  • LockFile is a context manager that simplifies the process of file locking in Python. The function sets up an flock and releases the lock on exit.

Network Utilities

A network-utilities library retrieves information about a network interface. The library is used for IP forwarding and for setting up an Ethernet interface on boot. The library exposes a GetNetworkInterface function that retrieves the network interface name associated with a MAC address.

Daemons

The guest environment daemons import and use the common libraries described above. Each daemon reads the configuration file before execution. This allows a user to easily disable undesired functionality. Additional daemon behaviors are detailed below.

Accounts

The accounts daemon is responsible for provisioning and deprovisioning user accounts. The daemon grants permissions to user accounts, and updates the list of authorized keys that have access to accounts based on metadata SSH key updates. User account creation is based on adding and remove SSH Keys stored in metadata.

The accounts management daemon has the following behaviors.

  • Administrator permissions are managed with a google-sudoers Linux group.
  • All users provisioned by the account daemon are added to the google-sudoers group.
  • The daemon stores a file in the guest to preserve state for the user accounts managed by Google.
  • The authorized keys file for a Google managed user is deleted when all SSH keys for the user are removed from metadata.
  • User accounts not managed by Google are not modified by the accounts daemon.

Clock Skew

The clock skew daemon is responsible for syncing the software clock with the hypervisor clock after a stop/start event or after a migration. Preventing clock skew may result in system time has changed messages in VM logs.

IP Forwarding

The IP forwarding daemon uses IP forwarding metadata to setup or remove IP routes in the guest.

  • Only IPv4 IP addresses are currently supported.
  • Routes are set on the default Ethernet interface determined dynamically.
  • Google routes are configured, by default, with the routing protocol ID 66. This ID is a namespace for daemon configured IP addresses.

Instance Setup

Instance setup runs during VM boot. The script configures the Linux guest environment by performing the following tasks.

  • Optimize for local SSD.
  • Enable multi-queue on all the virtionet devices.
  • Wait for network availability.
  • Set SSH host keys the first time the instance is booted.
  • Set the boto config for using Google Cloud Storage.
  • Create the defaults configuration file.

The defaults configuration file incorporates any user provided setting in /etc/default/instance_configs.cfg.template and does not override other conflicting settings. This allows package updates without overriding user configuration.

Metadata Scripts

Metadata scripts implement support for running user provided startup scripts and shutdown scripts. The guest support for metadata scripts is implemented in Python with the following design details.

  • Metadata scripts are executed in a shell.
  • If multiple metadata keys are specified (e.g. startup-script and startup-script-url) both are executed.
  • If multiple metadata keys are specified (e.g. startup-script and startup-script-url) a URL is executed first.
  • The exit status of a metadata script is logged after completed execution.

Network Setup

A network setup service runs on boot and enables all associated network interfaces. Network interfaces are specified by MAC address in instance metadata.

Configuration

Users of Google provided images may configure the guest environment behaviors using a configuration file. To make configuration changes, add settings to /etc/default/instance_configs.cfg.template. If you are attempting to change the behavior of a running instance, run /usr/bin/google_instance_setup before reloading the affected daemons.

Linux distributions looking to include their own defaults can specify settings in /etc/default/instance_configs.cfg.distro. These settings will not override /etc/default/instance_configs.cfg.template. This enables distribution settings that do not override user configuration during package update.

The following are valid user configuration options.

Section Option Value
Accounts deprovision_remove true makes deprovisioning a user destructive.
Accounts groups Comma separated list of groups for newly provisioned users.
Accounts useradd_cmd Command string to create a new user.
Accounts userdel_cmd Command string to delete a user.
Accounts usermod_cmd Command string to modify a user's groups.
Accounts groupadd_cmd Command string to create a new group.
Daemons accounts_daemon false disables the accounts daemon.
Daemons clock_skew_daemon false disables the clock skew daemon.
Daemons ip_forwarding_daemon false disables the IP forwarding daemon.
InstanceSetup host_key_types Comma separated list of host key types to generate.
InstanceSetup optimize_local_ssd false prevents optimizing for local SSD.
InstanceSetup network_enabled false skips instance setup functions that require metadata.
InstanceSetup set_boto_config false skips setting up a boto config.
InstanceSetup set_host_keys false skips generating host keys on first boot.
InstanceSetup set_multiqueue false skips multiqueue driver support.
IpForwarding ethernet_proto_id Protocol ID string for daemon added routes.
IpForwarding ip_aliases false disables setting up alias IP routes.
IpForwarding target_instance_ips false disables internal IP address load balancing.
MetadataScripts run_dir String base directory where metadata scripts are executed.
MetadataScripts startup false disables startup script execution.
MetadataScripts shutdown false disables shutdown script execution.
NetworkInterfaces dhclient_script String path to a dhclient script used by dhclient.
NetworkInterfaces dhcp_command String to execute to enable network interfaces.
NetworkInterfaces setup false disables network interface setup.

Setting network_enabled to false will skip setting up host keys and the boto config in the guest. The setting may also prevent startup and shutdown script execution.

Packaging

The guest Python code is packaged as a compliant PyPI Python package that can be used as a library or run independently. In addition to the Python package, deb and rpm packages are created with appropriate init configuration for supported GCE distros. The packages are targeted towards distribution provided Python versions.

Distro Package Type Python Version Init System
Debian 8 deb 2.7 systemd
Debian 9 deb 3.5 or 2.7 systemd
CentOS 6 rpm 2.6 upstart
CentOS 7 rpm 2.7 systemd
RHEL 6 rpm 2.6 upstart
RHEL 7 rpm 2.7 systemd
Ubuntu 14.04 deb 2.7 upstart
Ubuntu 16.04 deb 3.5 or 2.7 systemd
SLES 11 rpm 2.6 sysvinit
SLES 12 rpm 2.7 systemd

We build the following packages for the Linux guest environment.

  • google-compute-engine
    • System init scripts (systemd, upstart, or sysvinit).
    • Includes udev rules, sysctl rules, rsyslog configs, dhcp configs for hostname setting.
    • Entry point scripts created by the Python package located in /usr/bin.
    • Includes bash scripts used by instance_setup.
  • python-google-compute-engine
    • The Python 2 package for Linux daemons and libraries.
  • python3-google-compute-engine
    • The Python 3 package for Linux daemons and libraries.

The package source for Debian and RPM specs for Enterprise Linux 6 and 7 are included in this project. There are also Daisy workflows for spinning up GCE VM's to automatically build the packages for Debian, Red Hat, and CentOS. See the README in the packaging directory for more details.

Version Updates

There are several places where package versions have to be updated and must match to successfully release an update.

  • setup.py Update the version string with the Python package version. Used for entry points through the Python egg and PyPI.
  • specs/google-compute-engine.spec Update the version of the google-compute-engine package for EL6 and EL7.
  • specs/python-google-compute-engine.spec Update the version string of the python-google-compute-engine package for EL6 and EL7.
  • debian/changelog Update google-compute-image-packages (VERSION) stable, the version of the Debian packages.
  • Update the variable package_version when invoking the package build workflows.

Package Distribution

The deb and rpm packages used in some GCE images are published to Google Cloud repositories. Debian 8 and 9, CentOS 6 and 7, and RHEL 6 and 7 use these repositories to install and update the google-compute-engine, and python-google-compute-engine (and python3-google-compute-engine for Python 3) packages. If you are creating a custom image, you can also use these repositories in your image.

For Debian 8, run the following commands as root:

Add the public repo key to your system:

curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add -

Add a source list file /etc/apt/sources.list.d/google-cloud.list:

tee /etc/apt/sources.list.d/google-cloud.list << EOM
deb http://packages.cloud.google.com/apt google-compute-engine-jessie-stable main
deb http://packages.cloud.google.com/apt google-cloud-packages-archive-keyring-jessie main
EOM

Install the packages to maintain the public key over time:

apt-get update; apt-get install google-cloud-packages-archive-keyring

Install the google-compute-engine and python-google-compute-engine packages:

apt-get update; apt-get install -y google-compute-engine python-google-compute-engine

For Debian 9, run the following commands as root:

Add the public repo key to your system:

curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add -

Add a source list file /etc/apt/sources.list.d/google-cloud.list:

tee /etc/apt/sources.list.d/google-cloud.list << EOM
deb http://packages.cloud.google.com/apt google-compute-engine-stretch-stable main
deb http://packages.cloud.google.com/apt google-cloud-packages-archive-keyring-stretch main
EOM

Install the packages to maintain the public key over time:

apt-get update; apt-get install google-cloud-packages-archive-keyring

Install the google-compute-engine and python-google-compute-engine packages:

apt-get update; apt-get install -y google-compute-engine python-google-compute-engine

For EL6 and EL7 based distributions, run the following commands as root:

Add the yum repo to a repo file /etc/yum.repos.d/google-cloud.repo for either EL6 or EL7. Change DIST to either 6 or 7 respectively:

DIST=7
tee /etc/yum.repos.d/google-cloud.repo << EOM
[google-cloud-compute]
name=Google Cloud Compute
baseurl=https://packages.cloud.google.com/yum/repos/google-cloud-compute-el${DIST}-x86_64
enabled=1
gpgcheck=1
repo_gpgcheck=1
gpgkey=https://packages.cloud.google.com/yum/doc/yum-key.gpg
       https://packages.cloud.google.com/yum/doc/rpm-package-key.gpg
EOM

Install the google-compute-engine, python-google-compute-engine packages:

yum install -y google-compute-engine python-google-compute-engine

Troubleshooting

Deprecated Packages

Deprecated Package Replacement
google-compute-engine-jessie google-compute-engine and python-google-compute-engine
google-compute-engine-stretch google-compute-engine and python-google-compute-engine
google-compute-engine-init google-compute-engine
google-compute-engine-init-jessie google-compute-engine
google-compute-engine-init-stretch google-compute-engine
google-config google-compute-engine
google-config-jessie google-compute-engine
google-config-stretch google-compute-engine
google-compute-daemon python-google-compute-engine
google-startup-scripts google-compute-engine

An old CentOS 6 image fails to install the packages with an error on SCL

CentOS 6 images prior to v20160526 may fail to install the package with the error:

http://mirror.centos.org/centos/6/SCL/x86_64/repodata/repomd.xml: [Errno 14] PYCURL ERROR 22 - "The requested URL returned error: 404 Not Found"

Remove the stale repository file: sudo rm -f /etc/yum.repos.d/CentOS-SCL.repo

On some CentOS or RHEL 6 systems, extraneous python egg directories can cause the python daemons to fail.

In /usr/lib/python2.6/site-packages look for google_compute_engine-2.4.1-py27.egg-info directories and google_compute_engine-2.5.2.egg-info directories and delete them if you run into this problem.

Using boto with virtualenv

Specific to running boto inside of a Python virtualenv, virtual environments are isolated from system site-packages. This includes the installed Linux guest environment libraries that are used to configure boto credentials. There are two recommended solutions:

  • Create a virtual environment with virtualenv venv --system-site-packages.
  • Install boto via the Linux guest environment PyPI package using pip install google-compute-engine.

Contributing

Have a patch that will benefit this project? Awesome! Follow these steps to have it accepted.

  1. Please sign our Contributor License Agreement.
  2. Fork this Git repository and make your changes.
  3. Create a Pull Request against the development branch.
  4. Incorporate review feedback to your changes.
  5. Accepted!

License

All files in this repository are under the Apache License, Version 2.0 unless noted otherwise.