.. _development: ########### Development ########### All VyOS source code is hosted on GitHub under the VyOS organization which can be found here: https://github.com/vyos Our code is split into several modules. VyOS is composed of multiple individual packages, some of them are forks of upstream packages and are periodically synced with upstream, so keeping the whole source under a single repository would be very inconvenient and slow. There is now an ongoing effort to consolidate all VyOS-specific framework/config packages into vyos-1x package, but the basic structure is going to stay the same, just with fewer and fewer packages while the base code is rewritten from Perl/BASH into Python using and XML based interface definition for the CLI. The repository that contains all the ISO build scripts is: https://github.com/vyos/vyos-build The README.md file will guide you to use the this top level repository. Submit a Patch ============== Patches are always more then welcome. To have a clean and easy to maintain repository we have some guidelines when working with Git. A clean repository eases the automatic generation of a changelog file. A good approach for writing commit messages is actually to have a look at the file(s) history by invoking ``git log path/to/file.txt``. .. _prepare_commit: Prepare patch/commit -------------------- In a big system, such as VyOS, that is comprised of multiple components, it's impossible to keep track of all the changes and bugs/feature requests in one's head. We use a bugtracker known as Phabricator_ for it ("issue tracker" would be a better term, but this one stuck). The information is used in three ways: * Keep track of the progress (what we've already done in this branch and what we still need to do). * Prepare release notes for upcoming releases * Help future maintainers of VyOS (it could be you!) to find out why certain things have been changed in the codebase or why certain features have been added To make this approach work, every change must be associated with a task number (prefixed with **T**) and a component. If there is no bug report/feature request for the changes you are going to make, you have to create a Phabricator_ task first. Once there is an entry in Phabricator_, you should reference its id in your commit message, as shown below: * ``ddclient: T1030: auto create runtime directories`` * ``Jenkins: add current Git commit ID to build description`` If there is no Phabricator_ reference in the commits of your pull request, we have to ask you to amend the commit message. Otherwise we will have to reject it. Writing good commit messages ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The format should be and is inspired by: https://git-scm.com/book/ch5-2.html * A single, short, summary of the commit (recommended 50 characters or less, not exceeding 80 characters) containing a prefix of the changed component and the corresponding Phabricator_ reference e.g. ``snmp: T1111:`` or ``ethernet: T2222:`` - multiple components could be concatenated as in ``snmp: ethernet: T3333`` * In some contexts, the first line is treated as the subject of an email and the rest of the text as the body. The blank line separating the summary from the body is critical (unless you omit the body entirely); tools like rebase can get confused if you run the two together. * Followed by a message which describes all the details like: * What/why/how something has been changed, makes everyone's life easier when working with `git bisect` * All text of the commit message should be wrapped at 72 characters if possible which makes reading commit logs easier with ``git log`` on a standard terminal (which happens to be 80x25) * If applicable a reference to a previous commit should be made linking those commits nicely when browsing the history: ``After commit abcd12ef ("snmp: this is a headline") a Python import statement is missing, throwing the following exception: ABCDEF`` * Always use the ``-x`` option to the ``git cherry-pick`` command when back or forward porting an individual commit. This automatically appends the line: ``(cherry picked from commit )`` to the original authors commit message making it easier when bisecting problems. * Every change set must be consistent (self containing)! Do not fix multiple bugs in a single commit. If you already worked on multiple fixes in the same file use `git add --patch` to only add the parts related to the one issue into your upcoming commit. Limits: * We only accept bugfixes in packages other than https://github.com/vyos/vyos-1x as no new functionality should use the old style templates (``node.def`` and Perl/BASH code. Use the new style XML/Python interface instead. Please submit your patches using the well-known GitHub pull-request against our repositories found in the VyOS GitHub organisation at https://github.com/vyos Determinine source package -------------------------- Suppose you want to make a change in the webproxy script but yet you do not know which of the many VyOS packages ship this file. You can determine the VyOS package name in question by using Debian's ``dpkg -S`` command of your running VyOS installation. .. code-block:: none vyos@vyos:~ dpkg -S /opt/vyatta/sbin/vyatta-update-webproxy.pl vyatta-webproxy: /opt/vyatta/sbin/vyatta-update-webproxy.pl This means the file in question (``/opt/vyatta/sbin/vyatta-update-webproxy.pl``) is located in the ``vyatta-webproxy`` package which can be found here: https://github.com/vyos/vyatta-webproxy Fork Repository and submit Patch -------------------------------- Forking the repository and submitting a GitHub pull-request is the preferred way of submitting your changes to VyOS. You can fork any VyOS repository to your very own GitHub account by just appending ``/fork`` to any repository's URL on GitHub. To e.g. fork the ``vyos-1x`` repository, open the following URL in your favourite browser: https://github.com/vyos/vyos-1x/fork You then can proceed with cloning your fork or add a new remote to your local repository: * Clone: ``git clone https://github.com//vyos-1x.git`` * Fork: ``git remote add myfork https://github.com//vyos-1x.git`` In order to record you as the author of the fix please identify yourself to Git by setting up your name and email. This can be done local for this one and only repository ``git config`` or globally using ``git config --global``. .. code-block:: none git config --global user.name "J. Random Hacker" git config --global user.email "jrhacker@example.net" Make your changes and save them. Do the following for all changes files to record them in your created Git commit: * Add file to Git index using ``git add myfile``, or for a whole directory: ``git add somedir/*`` * Commit the changes by calling ``git commit``. Please use a meaningful commit headline (read above) and don't forget to reference the Phabricator_ ID. * Submit the patch ``git push`` and create the GitHub pull-request. Attach patch to Phabricator task -------------------------------- Follow the above steps on how to "Fork repository to submit a Patch". Instead of uploading "pushing" your changes to GitHub you can export the patches/ commits and send it to maintainers@vyos.net or attach it directly to the bug (preferred over email) * Export last commit to patch file: ``git format-patch`` or export the last two commits into its appropriate patch files: ``git format-patch -2`` Coding Guidelines ================= Like any other project we have some small guidelines about our source code, too. The rules we have are not there to punish you - the rules are in place to help us all. By having a consistent coding style it becomes very easy for new and also longtime contributors to navigate through the sources and all the implied logic of any one source file.. Python 3 **shall** be used. How long can we keep Python 2 alive anyway? No considerations for Python 2 compatibility **should** be taken at any time. Formatting ---------- * Python: Tabs **shall not** be used. Every indentation level should be 4 spaces * XML: Tabs **shall not** be used. Every indentation level should be 2 spaces .. note:: There are extensions to e.g. VIM (xmllint) which will help you to get your indention levels correct. Add to following to your .vimrc file: ``au FileType xml setlocal equalprg=xmllint\ --format\ --recover\ -\ 2>/dev/null`` now you can call the linter using ``gg=G`` in command mode. Text generation ^^^^^^^^^^^^^^^ Template processor **should** be used for generating config files. Built-in string formatting **may** be used for simple line-oriented formats where every line is self-contained, such as iptables rules. Template processor **must** be used for structured, multi-line formats such as those used by ISC DHCPd. The default template processor for VyOS code is Jinja2_. Summary ------- When modifying the source code, remember these rules of the legacy elimination campaign: * No new features in Perl * No old style command definitions * No code incompatible with Python3 Python ====== The switch to the Python programming language for new code is not merely a change of the language, but a chance to rethink and improve the programming approach. Let's face it: VyOS is full of spaghetti code where logic for reading the VyOS config, generating daemon configs, and restarting processes is all mixed up. Python (or any other language, for that matter) does not provide automatic protection from bad design, so we need to also devise design guidelines and follow them to keep the system extensible and maintainable. But we are here to assist you and want to guide you through how you can become a good VyOS contributor. The rules we have are not there to punish you - the rules are in place to help us all. What does it mean? By having a consistent coding style it becomes very easy for new contributors and also longtime contributors to navigate through the sources and all the implied logic of the spaghetti code. Please use the following template as good starting point when developing new modules or even rewrite a whole bunch of code in the new style XML/Pyhon interface. Configuration Script Structure and Behaviour -------------------------------------------- Your configuration script or operation mode script which is also written in Python3 should have a line break on 80 characters. This seems to be a bit odd nowadays but as some people also work remotely or program using vi(m) this is a fair good standard which I hope we can rely on. In addition this also helps when browsing the GitHub codebase on a mobile device if you happen to be a crazy scientist. .. code-block:: python #!/usr/bin/env python3 # # Copyright (C) 2019 VyOS maintainers and contributors # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License version 2 or later as # published by the Free Software Foundation. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program. If not, see . import sys from vyos.config import Config from vyos import ConfigError def get_config(): vc = Config() # Convert the VyOS config to an abstract internal representation config = ... return config def verify(config): # Verify that configuration is valid if invalid: raise ConfigError("Descriptive message") return True def generate(config): # Generate daemon configs pass def apply(config): # Apply the generated configs to the live system pass try: config = get_config() verify(config) except ConfigError as e: print(e) sys.exit(1) The ``get_config()`` function must convert the VyOS config to an abstract, internal representation. No other function is allowed to call the ``vyos.config. Config`` object method directly. The rationale for it is that when config reads are mixed with other logic, it's very hard to change the config syntax since you need to weed out every occurrence of the old syntax. If syntax-specific code is confined to a single function, the rest of the code can be left untouched as long as the internal representation remains compatible. Another advantage is testability of the code. Mocking the entire config subsystem is hard, while constructing an internal representation by hand is way simpler. The ``verify()`` function takes your internal representation of the config and checks if it's valid, otherwise it must raise ``ConfigError`` with an error message that describes the problem and possibly suggests how to fix it. It must not make any changes to the system. The rationale for it is again testability and, in the future when the config backend is ready and every script is rewritten in this fashion, ability to execute commit dry run ("commit test" like in JunOS) and abort commit before making any changes to the system if an error is found in any component. The ``generate()`` function generates config files for system components. The ``apply()`` function applies the generated configuration to the live system. It should use non-disruptive reload whenever possible. It may execute disruptive operations such as daemon process restart if a particular component does not support non-disruptive reload, or when the expected service degradation is minimal (for example, in case of auxiliary services such as LLDPd). In case of high impact services such as VPN daemon and routing protocols, when non- disruptive reload is supported for some but not all types of configuration changes, scripts authors should make effort to determine if a configuration change can be done in a non-disruptive way and only resort to disruptive restart if it cannot be avoided. Unless absolutely necessary, configuration scripts should not modify the active configuration of system components directly. Whenever at all possible, scripts should generate a configuration file or files that can be applied with a single command such as reloading a service through systemd init. Inserting statements one by one is particularly discouraged, for example, when configuring netfilter rules, saving them to a file and loading it with iptables-restore should always be preferred to executing iptables directly. The ``apply()`` and ``generate()`` functions may ``raise ConfigError`` if, for example, the daemon failed to start with the updated config. It shouldn't be a substitute for proper config checking in the ``verify()`` function. All reasonable effort should be made to verify that generated configuration is valid and will be accepted by the daemon, including, when necessary, cross- checks with other VyOS configuration subtrees. Exceptions, including ``VyOSError`` (which is raised by ``vyos.config.Config`` on improper config operations, such as trying to use ``list_nodes()`` on a non-tag node) should not be silenced or caught and re-raised as config error. Sure this will not look pretty on user's screen, but it will make way better bug reports, and help users (and most VyOS users are IT professionals) do their own debugging as well. For easy orientation we suggest you take a look on the ``ntp.py`` or ``interfaces-bonding.py`` (for tag nodes) implementation. Both files can be found in the vyos-1x_ repository. XML (used for CLI definitions) ============================== The bash (or better vbash) completion in VyOS is defined in *templates*. Templates are text files (called ``node.def``) stored in a directory tree. The directory names define the command names, and template files define the command behaviour. Before VyOS 1.2 (crux) this files were created by hand. After a complex redesign process_ the new style template are automatically generated from a XML input file. XML interface definitions for VyOS come with a RelaxNG schema and are located in the vyos-1x_ module. This schema is a slightly modified schema from VyConf_ alias VyOS 2.0 So VyOS 1.2.x interface definitions will be reusable in Nextgen VyOS Versions with very minimal changes. The great thing about schemas is not only that people can know the complete grammar for certain, but also that it can be automatically verified. The `scripts/build-command-templates` script that converts the XML definitions to old style templates also verifies them against the schema, so a bad definition will cause the package build to fail. I do agree that the format is verbose, but there is no other format now that would allow this. Besides, a specialized XML editor can alleviate the issue with verbosity. Example: .. code-block:: xml Task scheduler settings Scheduled task <string> Task name 999 UNIX crontab time specification string Execution interval <minutes> Execution interval in minutes <minutes>m Execution interval in minutes <hours>h Execution interval in hours <days>d Execution interval in days [1-9]([0-9]*)([mhd]{0,1}) Executable path and arguments Path to executable Arguments passed to the executable Command definitions are purely declarative, and cannot contain any logic. All logic for generating config files for target applications, restarting services and so on is implemented in configuration scripts instead. GNU Preprocessor ---------------- XML interface definition files use the `xml.in` file extension which was implemented in :vytask:`T1843`. XML interface definitions tend to have a lot of duplicated code in areas such as: * VIF (incl. VIF-S/VIF-C) * Address * Description * Enabled/Disabled Instead of supplying all those XML nodes multiple times there are now include files with predefined features. Brief overview: * `IPv4, IPv6 and DHCP(v6)`_ address assignment * `IPv4, IPv6`_ address assignment * `VLAN (VIF)`_ definition * `MAC address`_ assignment All interface definition XML input files (.in suffix) will be sent to the GCC preprocess and the output is stored in the `build/interface-definitions` folder. The previously mentioned `scripts/build-command-templates` script operates on the `build/interface-definitions` folder to generate all required CLI nodes. .. code-block:: none $ make interface_definitions install -d -m 0755 build/interface-definitions install -d -m 0755 build/op-mode-definitions Generating build/interface-definitions/intel_qat.xml from interface-definitions/intel_qat.xml.in Generating build/interface-definitions/interfaces-bonding.xml from interface-definitions/interfaces-bonding.xml.in Generating build/interface-definitions/cron.xml from interface-definitions/cron.xml.in Generating build/interface-definitions/pppoe-server.xml from interface-definitions/pppoe-server.xml.in Generating build/interface-definitions/mdns-repeater.xml from interface-definitions/mdns-repeater.xml.in Generating build/interface-definitions/tftp-server.xml from interface-definitions/tftp-server.xml.in [...] Guidelines ---------- Use of numbers ^^^^^^^^^^^^^^^ Use of numbers in command names **should** be avoided unless a number is a part of a protocol name or similar. Thus, ``protocols ospfv3`` is perfectly fine, but something like ``server-1`` is questionable at best. Help String ^^^^^^^^^^^ To ensure uniform look and feel, and improve readability, we should follow a set of guidelines consistently. Capitalization and punctuation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The first word of every help string **must** be capitalized. There **must not** be a period at the end of help strings. Rationale: this seems to be the unwritten standard in network device CLIs, and a good aesthetic compromise. Examples: * Good: "Frobnication algorithm" * Bad: "frobnication algorithm" * Bad: "Frobnication algorithm." * Horrible: "frobnication algorithm." Use of abbreviations and acronyms ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Abbreviations and acronyms **must** be capitalized. Examples: * Good: "TCP connection timeout" * Bad: "tcp connection timeout" * Horrible: "Tcp connection timeout" Acronyms also **must** be capitalized to visually distinguish them from normal words: Examples: * Good: RADIUS (as in remote authentication for dial-in user services) * Bad: radius (unless it's about the distance between a center of a circle and any of its points) Some abbreviations are traditionally written in mixed case. Generally, if it contains words "over" or "version", the letter **should** be lowercase. If there's an accepted spelling (especially if defined by an RFC or another standard), it **must** be followed. Examples: * Good: PPPoE, IPsec * Bad: PPPOE, IPSEC * Bad: pppoe, ipsec Use of verbs ~~~~~~~~~~~~ Verbs **should** be avoided. If a verb can be omitted, omit it. Examples: * Good: "TCP connection timeout" * Bad: "Set TCP connection timeout" If a verb is essential, keep it. For example, in the help text of ``set system ipv6 disable-forwarding``, "Disable IPv6 forwarding on all interfaces" is a perfectly justified wording. Prefer infinitives ~~~~~~~~~~~~~~~~~~ Verbs, when they are necessary, **should** be in their infinitive form. Examples: * Good: "Disable IPv6 forwarding" * Bad: "Disables IPv6 forwarding" Migrating old CLI ----------------- .. list-table:: :widths: 25 25 50 :header-rows: 1 * - Old concept/syntax - New syntax - Notes * - mynode/node.def - - Leaf nodes (nodes with values) use tag instead * - mynode/node.tag , tag: - - "${vyos_libexecdir}/validators/foo bar $VAR(@)" will be executed, will be displayed on failure * - syntax:expression: (arithmetic expression) - None - External arithmetic validator may be added if there's demand, complex validation is better left to commit-time scripts * - priority: 999 - 999 - Please leave a comment explaining why the priority was chosen (e.g. "after interfaces are configured") * - multi: - - Only applicable to leaf nodes * - allowed: echo foo bar - foo bar - * - allowed: cli-shell-api listNodes vpn ipsec esp-group - vpn ipsec esp-group ... - * - allowed: /path/to/script - ... - * - default: - None - Move default values to scripts * - commit:expression: - None - All commit time checks should be in the verify() function of the script * - begin:/create:/delete: - None - All logic should be in the scripts Debugging ========= There are two flags available to aid in debugging configuration scripts. Since configuration loading issues will manifest during boot, the flags are passed as kernel boot parameters. Kernel boot parameters ---------------------- * ``vyos-debug`` - Adding the parameter to the linux boot line will produce timing results for the execution of scripts during commit. If one is seeing an unexpected delay during manual or boot commit, this may be useful in identifying bottlenecks. The internal flag is ``VYOS_DEBUG``, and is found in vyatta-cfg_. Output is directed to ``/var/log/vyatta/cfg-stdout.log``. * ``vyos-config-debug`` - During development, coding errors can lead to a commit failure on boot, possibly resulting in a failed initialization of the CLI. In this circumstance, the kernel boot parameter ``vyos-config-debug`` will ensure access to the system as user ``vyos``, and will log a Python stack trace to the file ``/tmp/boot-config-trace``. Debugging features ------------------ A number of flags can be set up to change the behaviour of VyOS at runtime. These flags can be toggled using either environment variables or creating files. For each feature, a file called ``vyos.feature.debug`` can be created to toggle the feature on. If a parameter is required it can be placed inside the file as its first line. The file can be placed in ``/tmp`` for one time debugging (as the file will be removed on reboot) or placed in '/config' to stay permanently. For example, ``/tmp/vyos.ifconfig.debug`` can be created to enable interface debugging. It is also possible to set up the debugging using environment variables. In that case, the name will be (in uppercase) VYOS_FEATURE_DEBUG. for example running, ``export VYOS_IFCONFIG_DEBUG=""`` on your vash, will have the same effect as ``touch /tmp/vyos.ifconfig.debug``. * ``ifconfig`` - Once set, all commands used, and their responses received from the OS, will be presented on the screen for inspection. * ``command`` - Once set, all commands used, and their responses received from the OS, will be presented on the screen for inspection. * ``developer`` - Should a command fail, instead of printing a message to the user explaining how to report issues, the python interpreter will start a PBD post-mortem session to allow the developer to debug the issue. As the debugger will wait from input from the developer, it has the capacity to prevent a router to boot and therefore should only be permanently set up on production if you are ready to see the OS fail to boot. * ``log`` - In some rare cases, it may be useful to see what the OS is doing, including during boot. This option sends all commands used by VyOS to a file. The default file is ``/tmp/full-log`` but it can be changed. Config Migration ---------------- When writing a new configuration migrator it may happen that you see an error when you try to invoke it manually on a development system. This error will look like: .. code-block:: none vyos@vyos:~$ /opt/vyatta/etc/config-migrate/migrate/ssh/0-to-1 /tmp/config.boot Traceback (most recent call last): File "/opt/vyatta/etc/config-migrate/migrate/ssh/0-to-1", line 31, in config = ConfigTree(config_file) File "/usr/lib/python3/dist-packages/vyos/configtree.py", line 134, in __init__ raise ValueError("Failed to parse config: {0}".format(msg)) ValueError: Failed to parse config: Syntax error on line 240, character 1: Invalid syntax. The reason is that the configuration migration backend is rewritten and uses a new form of "magic string" which is applied on demand when real config migration is run on boot. When runnint individual migrators for testing, you need to convert the "magic string" on your own by: .. code-block:: none vyos@vyos:~$ /usr/libexec/vyos/run-config-migration.py --virtual --set-vintage vyos /tmp/config.boot Boot Timing ----------- During the migration and extensive rewrite of functionality from Perl into Python a significant increase in the overall system boottime was noticed. The system boot time can be analysed and a graph can be generated in the end which shows in detail who called whom during the system startup phase. This is done by utilizing the ``systemd-bootchart`` package which is now installed by default on the VyOS 1.3 (equuleus) branch. The configuration is also versioned so we get comparable results. ``systemd-bootchart`` is configured using this file: bootchart.conf_ To enable boot time graphing change the Kernel commandline and add the folowing string: ``init=/usr/lib/systemd/systemd-bootchart`` This can also be done permanently by changing ``/boot/grub/grub.cfg``. Priorities ========== VyOS CLI is all about priorities. Every CLI node has a corresponding ``node.def`` file and possibly an attached script that is executed when the node is present. Nodes can have a priority, and on system bootup - or any other ``commit`` to the config all scripts are executed from lowest to higest priority. This is good as this gives a deterministic behavior. To debug issues in priorities or to see what's going on in the background you can use the ``/opt/vyatta/sbin/priority.pl`` script which lists to you the execution order of the scripts. Continuous Integration ====================== VyOS makes use of Jenkins_ as our Continuous Integration (CI) service. Our CI server is publicly accessible here: https://ci.vyos.net. You can get a brief overview of all required components shipped in a VyOS ISO. To build our modules we utilize a CI/CD Pipeline script. Each and every VyOS component comes with it's own ``Jenkinsfile`` which is (more or less) a copy. The Pipeline utilizes the Docker container from the :ref:`build_iso` section - but instead of building it from source on every run, we rather always fetch a fresh copy (if needed) from Dockerhub_. Each module is build on demand if a new commit on the branch in question is found. After a successful run the resulting Debian Package(s) will be deployed to our Debian repository which is used during build time. It is located here: http://dev.packages.vyos.net/repositories/. .. _process: https://blog.vyos.io/vyos-development-digest-10 .. _VyConf: https://github.com/vyos/vyconf/tree/master/data/schemata .. _vyos-1x: https://github.com/vyos/vyos-1x/tree/current/schema .. _Jinja2: https://jinja.palletsprojects.com/ .. _Jenkins: https://jenkins.io/ .. _Dockerhub: https://hub.docker.com/u/vyos/ .. _`IPv4, IPv6 and DHCP(v6)`: https://github.com/vyos/vyos-1x/tree/current/interface-definitions/include/address-ipv4-ipv6-dhcp.xml.i .. _`IPv4, IPv6`: https://github.com/vyos/vyos-1x/tree/current/interface-definitions/include/address-ipv4-ipv6.xml.i .. _`VLAN (VIF)`: https://github.com/vyos/vyos-1x/tree/current/interface-definitions/include/vif.xml.i .. _`MAC address`: https://github.com/vyos/vyos-1x/tree/current/interface-definitions/include/interface-mac.xml.i .. _vyatta-cfg: https://github.com/vyos/vyatta-cfg .. _bootchart.conf: https://github.com/vyos/vyos-build/blob/current/data/live-build-config/includes.chroot/etc/systemd/bootchart.conf .. include:: ../common-references.rst