devstack/HACKING.rst

Contributing to DevStack

General

DevStack is written in UNIX shell script. It uses a number of bash-isms and so is limited to Bash (version 4 and up) and compatible shells. Shell script was chosen because it best illustrates the steps used to set up and interact with OpenStack components.

DevStack's official repository is located on git.openstack.org at https://git.openstack.org/openstack-dev/devstack. Besides the master branch that tracks the OpenStack trunk branches a separate branch is maintained for all OpenStack releases starting with Diablo (stable/diablo).

Contributing code to DevStack follows the usual OpenStack process as described in How To Contribute in the OpenStack wiki. DevStack's LaunchPad project contains the usual links for blueprints, bugs, etc.

The Gerrit review queue is used for all commits.

The primary script in DevStack is stack.sh, which performs the bulk of the work for DevStack's use cases. There is a subscript functions that contains generally useful shell functions and is used by a number of the scripts in DevStack.

A number of additional scripts can be found in the tools directory that may be useful in supporting DevStack installations. Of particular note are info.sh to collect and report information about the installed system, and install_prereqs.sh that handles installation of the prerequisite packages for DevStack. It is suitable, for example, to pre-load a system for making a snapshot.

Repo Layout

The DevStack repo generally keeps all of the primary scripts at the root level.

doc - Contains the Sphinx source for the documentation. tools/build_docs.sh is used to generate the HTML versions of the DevStack scripts. A complete doc build can be run with tox -edocs.

exercises - Contains the test scripts used to sanity-check and demonstrate some OpenStack functions. These scripts know how to exit early or skip services that are not enabled.

extras.d - Contains the dispatch scripts called by the hooks in stack.sh, unstack.sh and clean.sh. See the plugins docs <plugins> for more information.

files - Contains a variety of otherwise lost files used in configuring and operating DevStack. This includes templates for configuration files and the system dependency information. This is also where image files are downloaded and expanded if necessary.

lib - Contains the sub-scripts specific to each project. This is where the work of managing a project's services is located. Each top-level project (Keystone, Nova, etc) has a file here. Additionally there are some for system services and project plugins. These variables and functions are also used by related projects, such as Grenade, to manage a DevStack installation.

samples - Contains a sample of the local files not included in the DevStack repo.

tests - the DevStack test suite is rather sparse, mostly consisting of test of specific fragile functions in the functions and functions-common files.

tools - Contains a collection of stand-alone scripts. While these may reference the top-level DevStack configuration they can generally be run alone. There are also some sub-directories to support specific environments such as XenServer.

Scripts

DevStack scripts should generally begin by calling env(1) in the shebang line:

#!/usr/bin/env bash

Sometimes the script needs to know the location of the DevStack install directory. TOP_DIR should always point there, even if the script itself is located in a subdirectory:

# Keep track of the current DevStack directory.
TOP_DIR=$(cd $(dirname "$0") && pwd)

Many scripts will utilize shared functions from the functions file. There are also rc files (stackrc and openrc) that are often included to set the primary configuration of the user environment:

# Keep track of the current DevStack directory.
TOP_DIR=$(cd $(dirname "$0") && pwd)

# Import common functions
source $TOP_DIR/functions

# Import configuration
source $TOP_DIR/openrc

stack.sh is a rather large monolithic script that flows through from beginning to end. It has been broken down into project-specific subscripts (as noted above) located in lib to make stack.sh more manageable and to promote code reuse.

These library sub-scripts have a number of fixed entry points, some of which may just be stubs. These entry points will be called by stack.sh in the following order:

install_XXXX
configure_XXXX
init_XXXX
start_XXXX
stop_XXXX
cleanup_XXXX

There is a sub-script template in lib/templates to be used in creating new service sub-scripts. The comments in <> are meta comments describing how to use the template and should be removed.

In order to show the dependencies and conditions under which project functions are executed the top-level conditional testing for things like is_service_enabled should be done in stack.sh. There may be nested conditionals that need to be in the sub-script, such as testing for keystone being enabled in configure_swift().

stackrc

stackrc is the global configuration file for DevStack. It is responsible for calling local.conf (or localrc if it exists) so local user configuration is recognized.

The criteria for what belongs in stackrc can be vaguely summarized as follows:

• All project repositories and branches handled directly in stack.sh
• Global configuration that may be referenced in local.conf, i.e. DEST, DATA_DIR
• Global service configuration like ENABLED_SERVICES
• Variables used by multiple services that do not have a clear owner, i.e. VOLUME_BACKING_FILE_SIZE (nova-compute, nova-volumes and cinder) or PUBLIC_NETWORK_NAME (nova-network and neutron)
• Variables that can not be cleanly declared in a project file due to dependency ordering, i.e. the order of sourcing the project files can not be changed for other reasons but the earlier file needs to dereference a variable set in the later file. This should be rare.

Also, variable declarations in stackrc before local.conf is sourced do NOT allow overriding (the form FOO=${FOO:-baz}); if they did then they can already be changed in local.conf and can stay in the project file.

Documentation

The DevStack repo now contains all of the static pages of devstack.org in the doc/source directory. The OpenStack CI system rebuilds the docs after every commit and updates devstack.org (now a redirect to docs.openstack.org/developer/devstack).

All of the scripts are processed with shocco to render them with the comments as text describing the script below. For this reason we tend to be a little verbose in the comments _ABOVE the code they pertain to. Shocco also supports Markdown formatting in the comments; use it sparingly. Specifically, stack.sh uses Markdown headers to divide the script into logical sections.

The script used to drive <code>shocco</code> is <code>tools/build_docs.sh</code>. The complete docs build is also handled with <code>tox -edocs</code> per the OpenStack project standard.

Exercises

The scripts in the exercises directory are meant to 1) perform basic operational checks on certain aspects of OpenStack; and b) document the use of the OpenStack command-line clients.

In addition to the guidelines above, exercise scripts MUST follow the structure outlined here. swift.sh is perhaps the clearest example of these guidelines. These scripts are executed serially by exercise.sh in testing situations.

• Begin and end with a banner that stands out in a sea of script logs to aid in debugging failures, particularly in automated testing situations. If the end banner is not displayed, the script ended prematurely and can be assumed to have failed.

  echo "**************************************************"
  echo "Begin DevStack Exercise: $0"
  echo "**************************************************"
  ...
  set +o xtrace
  echo "**************************************************"
  echo "End DevStack Exercise: $0"
  echo "**************************************************"

• The scripts will generally have the shell xtrace attribute set to display the actual commands being executed, and the errexit attribute set to exit the script on non-zero exit codes:

  # This script exits on an error so that errors don't compound and you see
  # only the first error that occurred.
  set -o errexit
  
  # Print the commands being run so that we can see the command that triggers
  # an error.  It is also useful for following as the install occurs.
  set -o xtrace

• Settings and configuration are stored in exerciserc, which must be sourced after openrc or stackrc:

  # Import exercise configuration
  source $TOP_DIR/exerciserc

• There are a couple of helper functions in the common functions sub-script that will check for non-zero exit codes and unset environment variables and print a message and exit the script. These should be called after most client commands that are not otherwise checked to short-circuit long timeouts (instance boot failure, for example):

  swift post $CONTAINER
  die_if_error "Failure creating container $CONTAINER"
  
  FLOATING_IP=`euca-allocate-address | cut -f2`
  die_if_not_set FLOATING_IP "Failure allocating floating IP"

• If you want an exercise to be skipped when for example a service wasn't enabled for the exercise to be run, you can exit your exercise with the special exitcode 55 and it will be detected as skipped.

• The exercise scripts should only use the various OpenStack client binaries to interact with OpenStack. This specifically excludes any *-manage tools as those assume direct access to configuration and databases, as well as direct database access from the exercise itself.

• If specific configuration needs to be present for the exercise to complete, it should be staged in stack.sh, or called from stack.sh.

• The OS_* environment variables should be the only ones used for all authentication to OpenStack clients as documented in the CLIAuth wiki page.

• The exercise MUST clean up after itself if successful. If it is not successful, it is assumed that state will be left behind; this allows a chance for developers to look around and attempt to debug the problem. The exercise SHOULD clean up or graciously handle possible artifacts left over from previous runs if executed again. It is acceptable to require a reboot or even a re-install of DevStack to restore a clean test environment.

Bash Style Guidelines

DevStack defines a bash set of best practices for maintaining large collections of bash scripts. These should be considered as part of the review process.

DevStack uses the bashate style checker to enforce basic guidelines, similar to pep8 and flake8 tools for Python. The list below is not complete for what bashate checks, nor is it all checked by bashate. So many lines of code, so little time.

Whitespace Rules

• lines should not include trailing whitespace
• there should be no hard tabs in the file
• indents are 4 spaces, and all indentation should be some multiple of them

Control Structure Rules

• then should be on the same line as the if
• do should be on the same line as the for

Example:

if [[ -r $TOP_DIR/local.conf ]]; then
    LRC=$(get_meta_section_files $TOP_DIR/local.conf local)
    for lfile in $LRC; do
        if [[ "$lfile" == "localrc" ]]; then
            if [[ -r $TOP_DIR/localrc ]]; then
                warn $LINENO "localrc and local.conf:[[local]] both exist, using localrc"
            else
                echo "# Generated file, do not edit" >$TOP_DIR/.localrc.auto
                get_meta_section $TOP_DIR/local.conf local $lfile >>$TOP_DIR/.localrc.auto
            fi
        fi
    done
fi

Variables and Functions

• functions should be used whenever possible for clarity
• functions should use local variables as much as possible to ensure they are isolated from the rest of the environment
• local variables should be lower case, global variables should be upper case
• function names should_have_underscores, NotCamelCase.
• functions should be declared as per the regex ^function foo {$ with code starting on the next line

Review Criteria

There are some broad criteria that will be followed when reviewing your change

• Is it passing tests -- your change will not be reviewed thoroughly unless the official CI has run successfully against it.

• Does this belong in DevStack -- DevStack reviewers have a default position of "no" but are ready to be convinced by your change.

  For very large changes, you should consider the plugins system <plugins> to see if your code is better abstracted from the main repository.

  For smaller changes, you should always consider if the change can be encapsulated by per-user settings in local.conf. A common example is adding a simple config-option to an ini file. Specific flags are not usually required for this, although adding documentation about how to achieve a larger goal (which might include turning on various settings, etc) is always welcome.

• Work-arounds -- often things get broken and DevStack can be in a position to fix them. Work-arounds are fine, but should be presented in the context of fixing the root-cause of the problem. This means it is well-commented in the code and the change-log and mostly likely includes links to changes or bugs that fix the underlying problem.

• Should this be upstream -- DevStack generally does not override default choices provided by projects and attempts to not unexpectedly modify behavior.

• Context in commit messages -- DevStack touches many different areas and reviewers need context around changes to make good decisions. We also always want it to be clear to someone -- perhaps even years from now -- why we were motivated to make a change at the time.

• Reviewers -- please see MAINTAINERS.rst for a list of people that should be added to reviews of various sub-systems.

Making Changes, Testing, and CI

Changes to Devstack are tested by automated continuous integration jobs that run on a variety of Linux Distros using a handful of common configurations. What this means is that every change to Devstack is self testing. One major benefit of this is that developers do not typically need to add new non voting test jobs to add features to Devstack. Instead the features can be added, then if testing passes with the feature enabled the change is ready to merge (pending code review).

A concrete example of this was the switch from screen based service management to systemd based service management. No new jobs were created for this. Instead the features were added to devstack, tested locally and in CI using a change that enabled the feature, then once the enabling change was passing and the new behavior communicated and documented it was merged.

Using this process has been proven to be effective and leads to quicker implementation of desired features.