Fabric is a Python (2.5 or higher) library and command-line tool for streamlining the use of SSH for application deployment or systems administration tasks.
It provides a basic suite of operations for executing local or remote shell commands (normally or via sudo) and uploading/downloading files, as well as auxiliary functionality such as prompting the running user for input, or aborting execution.
Typical use involves creating a Python module containing one or more functions, then executing them via the fab command-line tool. Below is a small but complete “fabfile” containing a single task:
from fabric.api import run
def host_type():
run('uname -s')
Once a task is defined, it may be run on one or more servers, like so:
$ fab -H localhost,linuxbox host_type
[localhost] run: uname -s
[localhost] out: Darwin
[linuxbox] run: uname -s
[linuxbox] out: Linux
Done.
Disconnecting from localhost... done.
Disconnecting from linuxbox... done.
In addition to use via the fab tool, Fabric’s components may be imported into other Python code, providing a Pythonic interface to the SSH protocol suite at a higher level than that provided by e.g. the ssh library (which Fabric itself uses.)
Stable releases of Fabric are best installed via pip or easy_install; or you may download TGZ or ZIP source archives from a couple of official locations. Detailed instructions and links may be found on the Installation page.
We recommend using the latest stable version of Fabric; releases are made often to prevent any large gaps in functionality between the latest stable release and the development version.
However, if you want to live on the edge, you can pull down the source code from our Git repository, or fork us on Github. The Installation page has details for how to access the source code.
Any hackers interested in improving Fabric (or even users interested in how Fabric is put together or released) please see the Development page. It contains comprehensive info on contributing, repository layout, our release strategy, and more.
Please note that all documentation is currently written with Python 2.5 users in mind, but with an eye for eventual Python 3.x compatibility. This leads to the following patterns that may throw off readers used to Python 2.4 or who have already upgraded to Python 2.6/2.7:
Welcome to Fabric!
This document is a whirlwind tour of Fabric’s features and a quick guide to its use. Additional documentation (which is linked to throughout) can be found in the usage documentation – please make sure to check it out.
As the README says:
Fabric is a Python (2.5 or higher) library and command-line tool for streamlining the use of SSH for application deployment or systems administration tasks.
More specifically, Fabric is:
Naturally, most users combine these two things, using Fabric to write and execute Python functions, or tasks, to automate interactions with remote servers. Let’s take a look.
This wouldn’t be a proper tutorial without “the usual”:
def hello():
print("Hello world!")
Placed in a Python module file named fabfile.py in your current working directory, that hello function can be executed with the fab tool (installed as part of Fabric) and does just what you’d expect:
$ fab hello
Hello world!
Done.
That’s all there is to it. This functionality allows Fabric to be used as a (very) basic build tool even without importing any of its API.
Note
The fab tool simply imports your fabfile and executes the function or functions you instruct it to. There’s nothing magic about it – anything you can do in a normal Python script can be done in a fabfile!
It’s often useful to pass runtime parameters into your tasks, just as you might during regular Python programming. Fabric has basic support for this using a shell-compatible notation: <task name>:<arg>,<kwarg>=<value>,.... It’s contrived, but let’s extend the above example to say hello to you personally:
def hello(name="world"):
print("Hello %s!" % name)
By default, calling fab hello will still behave as it did before; but now we can personalize it:
$ fab hello:name=Jeff
Hello Jeff!
Done.
Those already used to programming in Python might have guessed that this invocation behaves exactly the same way:
$ fab hello:Jeff
Hello Jeff!
Done.
For the time being, your argument values will always show up in Python as strings and may require a bit of string manipulation for complex types such as lists. Future versions may add a typecasting system to make this easier.
See also
As used above, fab only really saves a couple lines of if __name__ == "__main__" boilerplate. It’s mostly designed for use with Fabric’s API, which contains functions (or operations) for executing shell commands, transferring files, and so forth.
Let’s build a hypothetical Web application fabfile. This example scenario is as follows: The Web application is managed via Git on a remote host vcshost. On localhost, we have a local clone of said Web application. When we push changes back to vcshost, we want to be able to immediately install these changes on a remote host my_server in an automated fashion. We will do this by automating the local and remote Git commands.
Fabfiles usually work best at the root of a project:
.
|-- __init__.py
|-- app.wsgi
|-- fabfile.py <-- our fabfile!
|-- manage.py
`-- my_app
|-- __init__.py
|-- models.py
|-- templates
| `-- index.html
|-- tests.py
|-- urls.py
`-- views.py
Note
We’re using a Django application here, but only as an example – Fabric is not tied to any external codebase, save for its SSH library.
For starters, perhaps we want to run our tests and commit to our VCS so we’re ready for a deploy:
from fabric.api import local
def prepare_deploy():
local("./manage.py test my_app")
local("git add -p && git commit")
local("git push")
The output of which might look a bit like this:
$ fab prepare_deploy
[localhost] run: ./manage.py test my_app
Creating test database...
Creating tables
Creating indexes
..........................................
----------------------------------------------------------------------
Ran 42 tests in 9.138s
OK
Destroying test database...
[localhost] run: git add -p && git commit
<interactive Git add / git commit edit message session>
[localhost] run: git push
<git push session, possibly merging conflicts interactively>
Done.
The code itself is straightforward: import a Fabric API function, local, and use it to run and interact with local shell commands. The rest of Fabric’s API is similar – it’s all just Python.
See also
Because Fabric is “just Python” you’re free to organize your fabfile any way you want. For example, it’s often useful to start splitting things up into subtasks:
from fabric.api import local
def test():
local("./manage.py test my_app")
def commit():
local("git add -p && git commit")
def push():
local("git push")
def prepare_deploy():
test()
commit()
push()
The prepare_deploy task can be called just as before, but now you can make a more granular call to one of the sub-tasks, if desired.
Our base case works fine now, but what happens if our tests fail? Chances are we want to put on the brakes and fix them before deploying.
Fabric checks the return value of programs called via operations and will abort if they didn’t exit cleanly. Let’s see what happens if one of our tests encounters an error:
$ fab prepare_deploy
[localhost] run: ./manage.py test my_app
Creating test database...
Creating tables
Creating indexes
.............E............................
======================================================================
ERROR: testSomething (my_project.my_app.tests.MainTests)
----------------------------------------------------------------------
Traceback (most recent call last):
[...]
----------------------------------------------------------------------
Ran 42 tests in 9.138s
FAILED (errors=1)
Destroying test database...
Fatal error: local() encountered an error (return code 2) while executing './manage.py test my_app'
Aborting.
Great! We didn’t have to do anything ourselves: Fabric detected the failure and aborted, never running the commit task.
But what if we wanted to be flexible and give the user a choice? A setting (or environment variable, usually shortened to env var) called warn_only lets you turn aborts into warnings, allowing flexible error handling to occur.
Let’s flip this setting on for our test function, and then inspect the result of the local call ourselves:
from __future__ import with_statement
from fabric.api import local, settings, abort
from fabric.contrib.console import confirm
def test():
with settings(warn_only=True):
result = local('./manage.py test my_app', capture=True)
if result.failed and not confirm("Tests failed. Continue anyway?"):
abort("Aborting at user request.")
[...]
In adding this new feature we’ve introduced a number of new things:
However, despite the additional complexity, it’s still pretty easy to follow, and is now much more flexible.
See also
Let’s start wrapping up our fabfile by putting in the keystone: a deploy task that is destined to run on one or more remote server(s), and ensures the code is up to date:
def deploy():
code_dir = '/srv/django/myproject'
with cd(code_dir):
run("git pull")
run("touch app.wsgi")
Here again, we introduce a handful of new concepts:
We also need to make sure we import the new functions at the top of our file:
from __future__ import with_statement
from fabric.api import local, settings, abort, run, cd
from fabric.contrib.console import confirm
With these changes in place, let’s deploy:
$ fab deploy
No hosts found. Please specify (single) host string for connection: my_server
[my_server] run: git pull
[my_server] out: Already up-to-date.
[my_server] out:
[my_server] run: touch app.wsgi
Done.
We never specified any connection info in our fabfile, so Fabric doesn’t know on which host(s) the remote command should be executed. When this happens, Fabric prompts us at runtime. Connection definitions use SSH-like “host strings” (e.g. user@host:port) and will use your local username as a default – so in this example, we just had to specify the hostname, my_server.
git pull works fine if you’ve already got a checkout of your source code – but what if this is the first deploy? It’d be nice to handle that case too and do the initial git clone:
def deploy():
code_dir = '/srv/django/myproject'
with settings(warn_only=True):
if run("test -d %s" % code_dir).failed:
run("git clone user@vcshost:/path/to/repo/.git %s" % code_dir)
with cd(code_dir):
run("git pull")
run("touch app.wsgi")
As with our calls to local above, run also lets us construct clean Python-level logic based on executed shell commands. However, the interesting part here is the git clone call: since we’re using Git’s SSH method of accessing the repository on our Git server, this means our remote run call will need to authenticate itself.
Older versions of Fabric (and similar high level SSH libraries) run remote programs in limbo, unable to be touched from the local end. This is problematic when you have a serious need to enter passwords or otherwise interact with the remote program.
Fabric 1.0 and later breaks down this wall and ensures you can always talk to the other side. Let’s see what happens when we run our updated deploy task on a new server with no Git checkout:
$ fab deploy
No hosts found. Please specify (single) host string for connection: my_server
[my_server] run: test -d /srv/django/myproject
Warning: run() encountered an error (return code 1) while executing 'test -d /srv/django/myproject'
[my_server] run: git clone user@vcshost:/path/to/repo/.git /srv/django/myproject
[my_server] out: Cloning into /srv/django/myproject...
[my_server] out: Password: <enter password>
[my_server] out: remote: Counting objects: 6698, done.
[my_server] out: remote: Compressing objects: 100% (2237/2237), done.
[my_server] out: remote: Total 6698 (delta 4633), reused 6414 (delta 4412)
[my_server] out: Receiving objects: 100% (6698/6698), 1.28 MiB, done.
[my_server] out: Resolving deltas: 100% (4633/4633), done.
[my_server] out:
[my_server] run: git pull
[my_server] out: Already up-to-date.
[my_server] out:
[my_server] run: touch app.wsgi
Done.
Notice the Password: prompt – that was our remote git call on our Web server, asking for the password to the Git server. We were able to type it in and the clone continued normally.
See also
Specifying connection info at runtime gets old real fast, so Fabric provides a handful of ways to do it in your fabfile or on the command line. We won’t cover all of them here, but we will show you the most common one: setting the global host list, env.hosts.
env is a global dictionary-like object driving many of Fabric’s settings, and can be written to with attributes as well (in fact, settings, seen above, is simply a wrapper for this.) Thus, we can modify it at module level near the top of our fabfile like so:
from __future__ import with_statement
from fabric.api import *
from fabric.contrib.console import confirm
env.hosts = ['my_server']
def test():
do_test_stuff()
When fab loads up our fabfile, our modification of env will execute, storing our settings change. The end result is exactly as above: our deploy task will run against the my_server server.
This is also how you can tell Fabric to run on multiple remote systems at once: because env.hosts is a list, fab iterates over it, calling the given task once for each connection.
Our completed fabfile is still pretty short, as such things go. Here it is in its entirety:
from __future__ import with_statement
from fabric.api import *
from fabric.contrib.console import confirm
env.hosts = ['my_server']
def test():
with settings(warn_only=True):
result = local('./manage.py test my_app', capture=True)
if result.failed and not confirm("Tests failed. Continue anyway?"):
abort("Aborting at user request.")
def commit():
local("git add -p && git commit")
def push():
local("git push")
def prepare_deploy():
test()
commit()
push()
def deploy():
code_dir = '/srv/django/myproject'
with settings(warn_only=True):
if run("test -d %s" % code_dir).failed:
run("git clone user@vcshost:/path/to/repo/.git %s" % code_dir)
with cd(code_dir):
run("git pull")
run("touch app.wsgi")
This fabfile makes use of a large portion of Fabric’s feature set:
However, there’s still a lot more we haven’t covered here! Please make sure you follow the various “see also” links, and check out the documentation table of contents on the main index page.
Thanks for reading!
Fabric is best installed via pip (highly recommended) or easy_install (older, but still works fine), e.g.:
$ pip install fabric
You may also opt to use your operating system’s package manager; the package is typically called fabric or python-fabric. E.g.:
$ sudo apt-get install fabric
Advanced users wanting to install a development version may use pip to grab the latest master branch:
$ pip install fabric==dev
Or, to install an editable version for debugging/hacking, execute pip install -e . (or python setup.py install) inside a downloaded or cloned copy of the source code.
In order for Fabric’s installation to succeed, you will need four primary pieces of software:
and, if using the parallel execution mode:
Please read on for important details on each dependency – there are a few gotchas.
Fabric requires Python version 2.5 or 2.6. Some caveats and notes about other Python versions:
Setuptools comes with some Python installations by default; if yours doesn’t, you’ll need to grab it. In such situations it’s typically packaged as python-setuptools, py25-setuptools or similar. Fabric may drop its setuptools dependency in the future, or include alternative support for the Distribute project, but for now setuptools is required for installation.
PyCrypto provides the low-level (C-based) encryption algorithms used to run SSH, and is thus required by our SSH library. There are a couple gotchas associated with installing PyCrypto: its compatibility with Python’s package tools, and the fact that it is a C-based extension.
We strongly recommend using pip to install Fabric as it is newer and generally better than easy_install. However, a combination of bugs in specific versions of Python, pip and PyCrypto can prevent installation of PyCrypto. Specifically:
When all three criteria are met, you may encounter No such file or directory IOErrors when trying to pip install Fabric or pip install PyCrypto.
The fix is simply to make sure at least one of the above criteria is not met, by doing the following (in order of preference):
Unless you are installing from a precompiled source such as a Debian apt repository or RedHat RPM, or using pypm, you will also need the ability to build Python C-based modules from source in order to install PyCrypto. Users on Unix-based platforms such as Ubuntu or Mac OS X will need the traditional C build toolchain installed (e.g. Developer Tools / XCode Tools on the Mac, or the build-essential package on Ubuntu or Debian Linux – basically, anything with gcc, make and so forth) as well as the Python development libraries, often named python-dev or similar.
For Windows users we recommend using ActivePython and PyPM, installing a C development environment such as Cygwin or obtaining a precompiled Win32 PyCrypto package from voidspace’s Python modules page.
Note
Some Windows users whose Python is 64-bit have found that the PyCrypto dependency winrandom may not install properly, leading to ImportErrors. In this scenario, you’ll probably need to compile winrandom yourself via e.g. MS Visual Studio. See #194 for info.
An optional dependency, the multiprocessing library is included in Python’s standard library in version 2.6 and higher. If you’re using Python 2.5 and want to make use of Fabric’s parallel execution features you’ll need to install it manually; the recommended route, as usual, is via pip. Please see the multiprocessing PyPI page for details.
Warning
Early versions of Python 2.6 (in our testing, 2.6.0 through 2.6.2) ship with a buggy multiprocessing module that appears to cause Fabric to hang at the end of sessions involving large numbers of concurrent hosts. If you encounter this problem, either use env.pool_size / -z to limit the amount of concurrency, or upgrade to Python >=2.6.3.
Python 2.5 is unaffected, as it requires the PyPI version of multiprocessing, which is newer than that shipped with Python <2.6.3.
If you are interested in doing development work on Fabric (or even just running the test suite), you may also need to install some or all of the following packages:
For an up-to-date list of exact testing/development requirements, including version numbers, please see the requirements.txt file included with the source distribution. This file is intended to be used with pip, e.g. pip install -r requirements.txt.
To obtain a tar.gz or zip archive of the Fabric source code, you may visit either of the following locations:
The Fabric developers manage the project’s source code with the Git DVCS. To follow Fabric’s development via Git instead of downloading official releases, you have the following options:
Note
If you’ve obtained the Fabric source via source control and plan on updating your checkout in the future, we highly suggest using python setup.py develop instead – it will use symbolic links instead of file copies, ensuring that imports of the library or use of the command-line tool will always refer to your checkout.
For information on the hows and whys of Fabric development, including which branches may be of interest and how you can help out, please see the Development page.
Windows users who already have ActiveState’s ActivePython distribution installed may find Fabric is best installed with its package manager, PyPM. Below is example output from an installation of Fabric via pypm:
C:\> pypm install fabric
The following packages will be installed into "%APPDATA%\Python" (2.7):
ssh-1.7.8 pycrypto-2.4 fabric-1.3.0
Get: [pypm-free.activestate.com] fabric 1.3.0
Get: [pypm-free.activestate.com] ssh 1.7.8
Get: [pypm-free.activestate.com] pycrypto 2.4
Installing ssh-1.7.8
Installing pycrypto-2.4
Installing fabric-1.3.0
Fixing script %APPDATA%\Python\Scripts\fab-script.py
C:\>
The Fabric development team is headed by Jeff Forcier, aka bitprophet. However, dozens of other developers pitch in by submitting patches and ideas via GitHub, IRC or the mailing list.
Please see the Source code checkouts section of the Installation page for details on how to obtain Fabric’s source code.
There are a number of ways to get involved with Fabric:
While we may not always reply promptly, we do try to make time eventually to inspect all contributions and either incorporate them or explain why we don’t feel the change is a good fit.
If a ticket-tracker ticket exists for a given issue, please keep all communication in that ticket’s comments – for example, when submitting patches via Github, it’s easier for us if you leave a note in the ticket instead of sending a Github pull request.
The core devs receive emails for just about any ticket-tracker activity, so additional notices via Github or other means only serve to slow things down.
Fabric tries hard to honor PEP-8, especially (but not limited to!) the following:
While Fabric’s development methodology isn’t set in stone yet, the following items detail how we currently organize the Git repository and expect to perform merges and so forth. This will be chiefly of interest to those who wish to follow a specific Git branch instead of released versions, or to any contributors.
Fabric tries to follow open-source standards and conventions in its release tagging, including typical version numbers such as 2.0, 1.2.5, or 1.2b1. Each release will be marked as a tag in the Git repositories, and are broken down as follows:
Major releases update the first number, e.g. going from 0.9 to 1.0, and indicate that the software has reached some very large milestone.
For example, the 1.0 release signified a commitment to a medium to long term API and some significant backwards incompatible (compared to the 0.9 series) features. Version 2.0 might indicate a rewrite using a new underlying network technology or an overhaul to be more object-oriented.
Major releases will often be backwards-incompatible with the previous line of development, though this is not a requirement, just a usual happenstance. Users should expect to have to make at least some changes to their fabfiles when switching between major versions.
Minor releases, such as moving from 1.0 to 1.1, typically mean that one or more new, large features has been added. They are also sometimes used to mark off the fact that a lot of bug fixes or small feature modifications have occurred since the previous minor release. (And, naturally, some of them will involve both at the same time.)
These releases are guaranteed to be backwards-compatible with all other releases containing the same major version number, so a fabfile that works with 1.0 should also work fine with 1.1 or even 1.9.
The third and final part of version numbers, such as the ‘3’ in 1.0.3, generally indicate a release containing one or more bugfixes, although minor feature modifications may (rarely) occur.
This third number is sometimes omitted for the first major or minor release in a series, e.g. 1.2 or 2.0, and in these cases it can be considered an implicit zero (e.g. 2.0.0).
Note
The 0.9 series of development included more significant feature work than is typically found in tertiary releases; from 1.0 onwards a more traditional approach, as per the above, is used.
Major and minor releases do not mark the end of the previous line or lines of development:
We hope that this policy will allow us to have a rapid minor release cycle (and thus keep new features coming out frequently) without causing users to feel too much pressure to upgrade right away. At the same time, the backwards compatibility guarantee means that users should still feel comfortable upgrading to the next minor release in order to stay within this sliding support window.
These are some of the most commonly encountered problems or frequently asked questions which we receive from users. They aren’t intended as a substitute for reading the rest of the documentation, especially the usage docs, so please make sure you check those out if your question is not answered here.
Init-style start/stop/restart scripts (e.g. /etc/init.d/apache2 start) sometimes don’t like Fabric’s allocation of a pseudo-tty, which is active by default. In almost all cases, explicitly calling the command in question with pty=False works correctly:
sudo("/etc/init.d/apache2 restart", pty=False)
If you have no need for interactive behavior and run into this problem frequently, you may want to deactivate pty allocation globally by setting env.always_use_pty to False.
While Fabric can be used for many shell-script-like tasks, there’s a slightly unintuitive catch: each run or sudo call has its own distinct shell session. This is required in order for Fabric to reliably figure out, after your command has run, what its standard out/error and return codes were.
Unfortunately, it means that code like the following doesn’t behave as you might assume:
def deploy():
run("cd /path/to/application")
run("./update.sh")
If that were a shell script, the second run call would have executed with a current working directory of /path/to/application/ – but because both commands are run in their own distinct session over SSH, it actually tries to execute $HOME/update.sh instead (since your remote home directory is the default working directory).
A simple workaround is to make use of shell logic operations such as &&, which link multiple expressions together (provided the left hand side executed without error) like so:
def deploy():
run("cd /path/to/application && ./update.sh")
Fabric provides a convenient shortcut for this specific use case, in fact: cd.
Note
You might also get away with an absolute path and skip directory changing altogether:
def deploy():
run("/path/to/application/update.sh")
However, this requires that the command in question makes no assumptions about your current working directory!
This message is typically generated by programs such as biff or mesg lurking within your remote user’s .profile or .bashrc files (or any other such files, including system-wide ones.) Fabric’s default mode of operation involves executing the Bash shell in “login mode”, which causes these files to be executed.
Because Fabric also doesn’t bother asking the remote end for a tty by default (as it’s not usually necessary) programs fired within your startup files, which expect a tty to be present, will complain – and thus, stderr output about “stdin is not a tty” or similar.
There are multiple ways to deal with this problem:
Because Fabric executes a shell on the remote end for each invocation of run or sudo (see also), backgrounding a process via the shell will not work as expected. Backgrounded processes may still prevent the calling shell from exiting until they stop running, and this in turn prevents Fabric from continuing on with its own execution.
The key to fixing this is to ensure that your process’ standard pipes are all disassociated from the calling shell, which may be done in a number of ways:
Use a pre-existing daemonization technique if one exists for the program at hand – for example, calling an init script instead of directly invoking a server binary.
Run the program under nohup and redirect stdin, stdout and stderr to /dev/null (or to your file of choice, if you need the output later):
run("nohup yes >& /dev/null < /dev/null &")
(yes is simply an example of a program that may run for a long time or forever; >&, < and & are Bash syntax for pipe redirection and backgrounding, respectively – see your shell’s man page for details.)
Use tmux, screen or dtach to fully detach the process from the running shell; these tools have the benefit of allowing you to reattach to the process later on if needed (among many other such benefits).
While Fabric is written with bash in mind, it’s not an absolute requirement. Simply change env.shell to call your desired shell, and include an argument similar to bash‘s -c argument, which allows us to build shell commands of the form:
/bin/bash -l -c "<command string here>"
where /bin/bash -l -c is the default value of env.shell.
Note
The -l argument specifies a login shell and is not absolutely required, merely convenient in many situations. Some shells lack the option entirely and it may be safely omitted in such cases.
A relatively safe baseline is to call /bin/sh, which may call the original sh binary, or (on some systems) csh, and give it the -c argument, like so:
from fabric.api import env
env.shell = "/bin/sh -c"
This has been shown to work on FreeBSD and may work on other systems as well.
Due to a bug of sorts in our SSH layer, it’s not currently possible for Fabric to always accurately detect the type of authentication needed. We have to try and guess whether we’re being asked for a private key passphrase or a remote server password, and in some cases our guess ends up being wrong.
The most common such situation is where you, the local user, appear to have an SSH keychain agent running, but the remote server is not able to honor your SSH key, e.g. you haven’t yet transferred the public key over or are using an incorrect username. In this situation, Fabric will prompt you with “Please enter passphrase for private key”, but the text you enter is actually being sent to the remote end’s password authentication.
We hope to address this in future releases by modifying a fork of the aforementioned SSH library.
Currently, no, it’s not – the present version of Fabric relies heavily on shared state in order to keep the codebase simple. However, there are definite plans to update its internals so that Fabric may be either threaded or otherwise parallelized so your tasks can run on multiple servers concurrently.
This document outlines Fabric’s intended development path. Please make sure you’re reading the latest version of this document!
Warning
This information is subject to change without warning, and should not be used as a basis for any life- or career-altering decisions!
In no particular order, some potential future feature releases:
For new users, and/or for an overview of Fabric’s basic functionality, please see the Overview and Tutorial. The rest of the documentation will assume you’re at least passingly familiar with the material contained within.
The following list contains all major sections of Fabric’s prose (non-API) documentation, which expands upon the concepts outlined in the Overview and Tutorial and also covers advanced topics.
A simple but integral aspect of Fabric is what is known as the “environment”: a Python dictionary subclass which is used as a combination settings registry and shared inter-task data namespace.
The environment dict is currently implemented as a global singleton, fabric.state.env, and is included in fabric.api for convenience. Keys in env are sometimes referred to as “env variables”.
Most of Fabric’s behavior is controllable by modifying env variables, such as env.hosts (as seen in the tutorial). Other commonly-modified env vars include:
There are a number of other env variables; for the full list, see Full list of env vars at the bottom of this document.
In many situations, it’s useful to only temporarily modify env vars so that a given settings change only applies to a block of code. Fabric provides a settings context manager, which takes any numbr of key/value pairs and will use them to modify env within its wrapped block.
For example, there are many situations where setting warn_only (see below) is useful. To apply it to a few lines of code, use settings(warn_only=True), as seen in this simplified version of the contrib exists function:
from fabric.api import settings, run
def exists(path):
with settings(warn_only=True):
return run('test -e %s' % path)
See the Context Managers API documentation for details on settings and other, similar tools.
While it subclasses dict, Fabric’s env has been modified so that its values may be read/written by way of attribute access, as seen in some of the above material. In other words, env.host_string and env['host_string'] are functionally identical. We feel that attribute access can often save a bit of typing and makes the code more readable, so it’s the recommended way to interact with env.
The fact that it’s a dictionary can be useful in other ways, such as with Python’s dict-based string interpolation, which is especially handy if you need to insert multiple env vars into a single string. Using “normal” string interpolation might look like this:
print("Executing on %s as %s" % (env.host, env.user))
Using dict-style interpolation is more readable and slightly shorter:
print("Executing on %(host)s as %(user)s" % env)
Below is a list of all predefined (or defined by Fabric itself during execution) environment variables. While any of them may be manipulated directly, it’s often best to use context_managers, either generally via settings or via specific context managers such as cd.
Note that many of these may be set via fab‘s command-line switches – see fab options and arguments for details. Cross-links will be provided where appropriate.
Default: False
When True, Fabric will run in a non-interactive mode, calling abort anytime it would normally prompt the user for input (such as password prompts, “What host to connect to?” prompts, fabfile invocation of prompt, and so forth.) This allows users to ensure a Fabric session will always terminate cleanly instead of blocking on user input forever when unforeseen circumstances arise.
New in version 1.1.
See also
Default: None
Set by fab to the full host list for the currently executing command. For informational purposes only.
See also
Default: True
When set to False, causes run/sudo to act as if they have been called with pty=False.
The command-line flag --no-pty, if given, will set this env var to False.
New in version 1.0.
Default: True
Causes the SSH layer to merge a remote program’s stdout and stderr streams to avoid becoming meshed together when printed. See Combining stdout and stderr for details on why this is needed and what its effects are.
New in version 1.0.
Default: None
Set by fab to the currently executing command name (e.g. when executed as $ fab task1 task2, env.command will be set to "task1" while task1 is executing, and then to "task2".) For informational purposes only.
See also
Default: []
Modified by prefix, and prepended to commands executed by run/sudo.
New in version 1.0.
Default: False
If True, the SSH layer will skip loading the user’s known-hosts file. Useful for avoiding exceptions in situations where a “known host” changing its host key is actually valid (e.g. cloud servers such as EC2.)
See also
Default: []
Specifies a list of host strings to be skipped over during fab execution. Typically set via --exclude-hosts/-x.
New in version 1.1.
Default: fabfile.py
Filename pattern which fab searches for when loading fabfiles. To indicate a specific file, use the full path to the file. Obviously, it doesn’t make sense to set this in a fabfile, but it may be specified in a .fabricrc file or on the command line.
See also
Default: None
Defines the current user/host/port which Fabric will connect to when executing run, put and so forth. This is set by fab when iterating over a previously set host list, and may also be manually set when using Fabric as a library.
See also
Default: None
Set to the hostname part of env.host_string by fab. For informational purposes only.
Default: 0 (i.e. no keepalive)
An integer specifying an SSH keepalive interval to use; basically maps to the SSH config option ClientAliveInterval. Useful if you find connections are timing out due to meddlesome network hardware or what have you.
See also
New in version 1.1.
Default: None
May be a string or list of strings, referencing file paths to SSH key files to try when connecting. Passed through directly to the SSH layer. May be set/appended to with -i.
Default: False
Forces buffering by line instead of by character/byte, typically when running in parallel mode. May be activated via --linewise. This option is implied by env.parallel – even if linewise is False, if parallel is True then linewise behavior will occur.
See also
New in version 1.3.
A read-only value containing the local system username. This is the same value as user‘s initial value, but whereas user may be altered by CLI arguments, Python code or specific host strings, local_user will always contain the same value.
Default: False
If True, will tell the SSH layer not to seek out running SSH agents when using key-based authentication.
New in version 0.9.1.
Default: False
If True, will tell the SSH layer not to load any private key files from one’s $HOME/.ssh/ folder. (Key files explicitly loaded via fab -i will still be used, of course.)
New in version 0.9.1.
Default: None
The default password used by the SSH layer when connecting to remote hosts, and/or when answering sudo prompts.
See also
See also
Default: {}
This dictionary is largely for internal use, and is filled automatically as a per-host-string password cache. Keys are full host strings and values are passwords (strings).
See also
New in version 1.0.
Default: ''
Used to set the $PATH shell environment variable when executing commands in run/sudo/local. It is recommended to use the path context manager for managing this value instead of setting it directly.
New in version 1.0.
Default: 0
Sets the number of concurrent processes to use when executing tasks in parallel.
New in version 1.3.
See also
Default: None
Set to the port part of env.host_string by fab when iterating over a host list. For informational purposes only.
Default: None
Set by fab with the path to the fabfile it has loaded up, if it got that far. For informational purposes only.
See also
Default: False
If True, the SSH layer will raise an exception when connecting to hosts not listed in the user’s known-hosts file.
See also
Default: False
When True, forces all tasks to run in parallel. Implies env.linewise.
New in version 1.3.
See also
Default: /bin/bash -l -c
Value used as shell wrapper when executing commands with e.g. run. Must be able to exist in the form <env.shell> "<command goes here>" – e.g. the default uses Bash’s -c option which takes a command string as its value.
See also
Default: "sudo -S -p '%(sudo_prompt)s' " % env
The actual sudo command prefixed onto sudo calls’ command strings. Users who do not have sudo on their default remote $PATH, or who need to make other changes (such as removing the -p when passwordless sudo is in effect) may find changing this useful.
See also
The sudo operation; env.sudo_prompt
Default: "sudo password:"
Passed to the sudo program on remote systems so that Fabric may correctly identify its password prompt.
See also
The sudo operation; env.sudo_prefix
Default: True
Global setting which acts like the use_shell argument to run/sudo: if it is set to False, operations will not wrap executed commands in env.shell.
Default: User’s local username
The username used by the SSH layer when connecting to remote hosts. May be set globally, and will be used when not otherwise explicitly set in host strings. However, when explicitly given in such a manner, this variable will be temporarily overwritten with the current value – i.e. it will always display the user currently being connected as.
To illustrate this, a fabfile:
from fabric.api import env, run
env.user = 'implicit_user'
env.hosts = ['host1', 'explicit_user@host2', 'host3']
def print_user():
with hide('running'):
run('echo "%(user)s"' % env)
and its use:
$ fab print_user
[host1] out: implicit_user
[explicit_user@host2] out: explicit_user
[host3] out: implicit_user
Done.
Disconnecting from host1... done.
Disconnecting from host2... done.
Disconnecting from host3... done.
As you can see, during execution on host2, env.user was set to "explicit_user", but was restored to its previous value ("implicit_user") afterwards.
Note
env.user is currently somewhat confusing (it’s used for configuration and informational purposes) so expect this to change in the future – the informational aspect will likely be broken out into a separate env variable.
See also
Default: current Fabric version string
Mostly for informational purposes. Modification is not recommended, but probably won’t break anything either.
If you’ve read the Overview and Tutorial, you should already be familiar with how Fabric operates in the base case (a single task on a single host.) However, in many situations you’ll find yourself wanting to execute multiple tasks and/or on multiple hosts. Perhaps you want to split a big task into smaller reusable parts, or crawl a collection of servers looking for an old user to remove. Such a scenario requires specific rules for when and how tasks are executed.
This document explores Fabric’s execution model, including the main execution loop, how to define host lists, how connections are made, and so forth.
Fabric defaults to a single, serial execution method, though there is an alternative parallel mode available as of Fabric 1.3 (see Parallel execution). This default behavior is as follows:
Thus, given the following fabfile:
from fabric.api import run, env
env.hosts = ['host1', 'host2']
def taskA():
run('ls')
def taskB():
run('whoami')
and the following invocation:
$ fab taskA taskB
you will see that Fabric performs the following:
While this approach is simplistic, it allows for a straightforward composition of task functions, and (unlike tools which push the multi-host functionality down to the individual function calls) enables shell script-like logic where you may introspect the output or return code of a given command and decide what to do next.
For details on what constitutes a Fabric task and how to organize them, please see Defining tasks.
Unless you’re using Fabric as a simple build system (which is possible, but not the primary use-case) having tasks won’t do you any good without the ability to specify remote hosts on which to execute them. There are a number of ways to do so, with scopes varying from global to per-task, and it’s possible mix and match as needed.
Hosts, in this context, refer to what are also called “host strings”: Python strings specifying a username, hostname and port combination, in the form of username@hostname:port. User and/or port (and the associated @ or :) may be omitted, and will be filled by the executing user’s local username, and/or port 22, respectively. Thus, admin@foo.com:222, deploy@website and nameserver1 could all be valid host strings.
Note
The user/hostname split occurs at the last @ found, so e.g. email address usernames are valid and will be parsed correctly.
During execution, Fabric normalizes the host strings given and then stores each part (username/hostname/port) in the environment dictionary, for both its use and for tasks to reference if the need arises. See The environment dictionary, env for details.
Host strings map to single hosts, but sometimes it’s useful to arrange hosts in groups. Perhaps you have a number of Web servers behind a load balancer and want to update all of them, or want to run a task on “all client servers”. Roles provide a way of defining strings which correspond to lists of host strings, and can then be specified instead of writing out the entire list every time.
This mapping is defined as a dictionary, env.roledefs, which must be modified by a fabfile in order to be used. A simple example:
from fabric.api import env
env.roledefs['webservers'] = ['www1', 'www2', 'www3']
Since env.roledefs is naturally empty by default, you may also opt to re-assign to it without fear of losing any information (provided you aren’t loading other fabfiles which also modify it, of course):
from fabric.api import env
env.roledefs = {
'web': ['www1', 'www2', 'www3'],
'dns': ['ns1', 'ns2']
}
In addition to list/iterable object types, the values in env.roledefs may be callables, and will thus be called when looked up when tasks are run instead of at module load time. (For example, you could connect to remote servers to obtain role definitions, and not worry about causing delays at fabfile load time when calling e.g. fab --list.)
Use of roles is not required in any way – it’s simply a convenience in situations where you have common groupings of servers.
Changed in version 0.9.2: Added ability to use callables as roledefs values.
There are a number of ways to specify host lists, either globally or per-task, and generally these methods override one another instead of merging together (though this may change in future releases.) Each such method is typically split into two parts, one for hosts and one for roles.
The most common method of setting hosts or roles is by modifying two key-value pairs in the environment dictionary, env: hosts and roles. The value of these variables is checked at runtime, while constructing each tasks’s host list.
Thus, they may be set at module level, which will take effect when the fabfile is imported:
from fabric.api import env, run
env.hosts = ['host1', 'host2']
def mytask():
run('ls /var/www')
Such a fabfile, run simply as fab mytask, will run mytask on host1 followed by host2.
Since the env vars are checked for each task, this means that if you have the need, you can actually modify env in one task and it will affect all following tasks:
from fabric.api import env, run
def set_hosts():
env.hosts = ['host1', 'host2']
def mytask():
run('ls /var/www')
When run as fab set_hosts mytask, set_hosts is a “local” task – its own host list is empty – but mytask will again run on the two hosts given.
Note
This technique used to be a common way of creating fake “roles”, but is less necessary now that roles are fully implemented. It may still be useful in some situations, however.
Alongside env.hosts is env.roles (not to be confused with env.roledefs!) which, if given, will be taken as a list of role names to look up in env.roledefs.
In addition to modifying env.hosts, env.roles, and env.exclude_hosts at the module level, you may define them by passing comma-separated string arguments to the command-line switches --hosts/-H and --roles/-R, e.g.:
$ fab -H host1,host2 mytask
Such an invocation is directly equivalent to env.hosts = ['host1', 'host2'] – the argument parser knows to look for these arguments and will modify env at parse time.
Note
It’s possible, and in fact common, to use these switches to set only a single host or role. Fabric simply calls string.split(',') on the given string, so a string with no commas turns into a single-item list.
It is important to know that these command-line switches are interpreted before your fabfile is loaded: any reassignment to env.hosts or env.roles in your fabfile will overwrite them.
If you wish to nondestructively merge the command-line hosts with your fabfile-defined ones, make sure your fabfile uses env.hosts.extend() instead:
from fabric.api import env, run
env.hosts.extend(['host3', 'host4'])
def mytask():
run('ls /var/www')
When this fabfile is run as fab -H host1,host2 mytask, env.hosts will then contain ['host1', 'host2', 'host3', 'host4'] at the time that mytask is executed.
Note
env.hosts is simply a Python list object – so you may use env.hosts.append() or any other such method you wish.
Globally setting host lists only works if you want all your tasks to run on the same host list all the time. This isn’t always true, so Fabric provides a few ways to be more granular and specify host lists which apply to a single task only. The first of these uses task arguments.
As outlined in fab options and arguments, it’s possible to specify per-task arguments via a special command-line syntax. In addition to naming actual arguments to your task function, this may be used to set the host, hosts, role or roles “arguments”, which are interpreted by Fabric when building host lists (and removed from the arguments passed to the task itself.)
Note
Since commas are already used to separate task arguments from one another, semicolons must be used in the hosts or roles arguments to delineate individual host strings or role names. Furthermore, the argument must be quoted to prevent your shell from interpreting the semicolons.
Take the below fabfile, which is the same one we’ve been using, but which doesn’t define any host info at all:
from fabric.api import run
def mytask():
run('ls /var/www')
To specify per-task hosts for mytask, execute it like so:
$ fab mytask:hosts="host1;host2"
This will override any other host list and ensure mytask always runs on just those two hosts.
If a given task should always run on a predetermined host list, you may wish to specify this in your fabfile itself. This can be done by decorating a task function with the hosts or roles decorators. These decorators take a variable argument list, like so:
from fabric.api import hosts, run
@hosts('host1', 'host2')
def mytask():
run('ls /var/www')
They will also take an single iterable argument, e.g.:
my_hosts = ('host1', 'host2')
@hosts(my_hosts)
def mytask():
# ...
When used, these decorators override any checks of env for that particular task’s host list (though env is not modified in any way – it is simply ignored.) Thus, even if the above fabfile had defined env.hosts or the call to fab uses --hosts/-H, mytask would still run on a host list of ['host1', 'host2'].
However, decorator host lists do not override per-task command-line arguments, as given in the previous section.
We’ve been pointing out which methods of setting host lists trump the others, as we’ve gone along. However, to make things clearer, here’s a quick breakdown:
This logic may change slightly in the future to be more consistent (e.g. having --hosts somehow take precedence over env.hosts in the same way that command-line per-task lists trump in-code ones) but only in a backwards-incompatible release.
There is no “unionizing” of hosts between the various sources mentioned in How host lists are constructed. If env.hosts is set to ['host1', 'host2', 'host3'], and a per-function (e.g. via hosts) host list is set to just ['host2', 'host3'], that function will not execute on host1, because the per-task decorator host list takes precedence.
However, for each given source, if both roles and hosts are specified, they will be merged together into a single host list. Take, for example, this fabfile where both of the decorators are used:
from fabric.api import env, hosts, roles, run
env.roledefs = {'role1': ['b', 'c']}
@hosts('a', 'b')
@roles('role1')
def mytask():
run('ls /var/www')
Assuming no command-line hosts or roles are given when mytask is executed, this fabfile will call mytask on a host list of ['a', 'b', 'c'] – the union of role1 and the contents of the hosts call.
At times, it is useful to exclude one or more specific hosts, e.g. to override a few bad or otherwise undesirable hosts which are pulled in from a role or an autogenerated host list. This may be accomplished globally with --exclude-hosts/-x:
$ fab -R myrole -x host2,host5 mytask
If myrole was defined as ['host1', 'host2', ..., 'host15'], the above invocation would run with an effective host list of ['host1', 'host3', 'host4', 'host6', ..., 'host15'].
Note
Using this option does not modify env.hosts – it only causes the main execution loop to skip the requested hosts.
Exclusions may be specified per-task by using an extra exclude_hosts kwarg, which is implemented similarly to the abovementioned hosts and roles per-task kwargs, in that it is stripped from the actual task invocation. This example would have the same result as the global exclude above:
$ fab mytask:roles=myrole,exclude_hosts="host2;host5"
Note that the host list is semicolon-separated, just as with the hosts per-task argument.
Host exclusion lists, like host lists themselves, are not merged together across the different “levels” they can be declared in. For example, a global -x option will not affect a per-task host list set with a decorator or keyword argument, nor will per-task exclude_hosts keyword arguments affect a global -H list.
There is one minor exception to this rule, namely that CLI-level keyword arguments (mytask:exclude_hosts=x,y) will be taken into account when examining host lists set via @hosts or @roles. Thus a task function decorated with @hosts('host1', 'host2') executed as fab taskname:exclude_hosts=host2 will only run on host1.
As with the host list merging, this functionality is currently limited (partly to keep the implementation simple) and may be expanded in future releases.
New in version 1.3.
Most of the information here involves “top level” tasks executed via fab, such as the first example where we called fab taskA taskB. However, it’s often convenient to wrap up multi-task invocations like this into their own, “meta” tasks.
Prior to Fabric 1.3, this had to be done by hand, as outlined in Library Use. Fabric’s design eschews magical behavior, so simply calling a task function does not take into account decorators such as roles.
New in Fabric 1.3 is the execute helper function, which takes a task object or name as its first argument. Using it is effectively the same as calling the given task from the command line: all the rules given above in How host lists are constructed apply. (The hosts and roles keyword arguments to execute are analogous to CLI per-task arguments, including how they override all other host/role-setting methods.)
As an example, here’s a fabfile defining two stand-alone tasks for deploying a Web application:
from fabric.api import run, roles
env.roledefs = {
'db': ['db1', 'db2'],
'web': ['web1', 'web2', 'web3'],
}
@roles('db')
def migrate():
# Database stuff here.
pass
@roles('web')
def update():
# Code updates here.
pass
In Fabric <=1.2, the only way to ensure that migrate runs on the DB servers and that update runs on the Web servers (short of manual env.host_string manipulation) was to call both as top level tasks:
$ fab migrate update
Fabric >=1.3 can use execute to set up a meta-task. Update the import line like so:
from fabric.api import run, roles, execute
and append this to the bottom of the file:
def deploy():
execute(migrate)
execute(update)
That’s all there is to it; the roles decorators will be honored as expected, resulting in the following execution sequence:
Warning
This technique works because tasks that themselves have no host list (this includes the global host list settings) only run one time. If used inside a “regular” task that is going to run on multiple hosts, calls to execute will also run multiple times, resulting in multiplicative numbers of subtask calls – be careful!
See also
Once the task list has been constructed, Fabric will start executing them as outlined in Execution strategy, until all tasks have been run on the entirety of their host lists. However, Fabric defaults to a “fail-fast” behavior pattern: if anything goes wrong, such as a remote program returning a nonzero return value or your fabfile’s Python code encountering an exception, execution will halt immediately.
This is typically the desired behavior, but there are many exceptions to the rule, so Fabric provides env.warn_only, a Boolean setting. It defaults to False, meaning an error condition will result in the program aborting immediately. However, if env.warn_only is set to True at the time of failure – with, say, the settings context manager – Fabric will emit a warning message but continue executing.
fab itself doesn’t actually make any connections to remote hosts. Instead, it simply ensures that for each distinct run of a task on one of its hosts, the env var env.host_string is set to the right value. Users wanting to leverage Fabric as a library may do so manually to achieve similar effects.
env.host_string is (as the name implies) the “current” host string, and is what Fabric uses to determine what connections to make (or re-use) when network-aware functions are run. Operations like run or put use env.host_string as a lookup key in a shared dictionary which maps host strings to SSH connection objects.
Note
The connections dictionary (currently located at fabric.state.connections) acts as a cache, opting to return previously created connections if possible in order to save some overhead, and creating new ones otherwise.
Because connections are driven by the individual operations, Fabric will not actually make connections until they’re necessary. Take for example this task which does some local housekeeping prior to interacting with the remote server:
from fabric.api import *
@hosts('host1')
def clean_and_upload():
local('find assets/ -name "*.DS_Store" -exec rm '{}' \;')
local('tar czf /tmp/assets.tgz assets/')
put('/tmp/assets.tgz', '/tmp/assets.tgz')
with cd('/var/www/myapp/'):
run('tar xzf /tmp/assets.tgz')
What happens, connection-wise, is as follows:
Extrapolating from this, you can also see that tasks which don’t use any network-borne operations will never actually initiate any connections (though they will still be run once for each host in their host list, if any.)
Fabric’s connection cache never closes connections itself – it leaves this up to whatever is using it. The fab tool does this bookkeeping for you: it iterates over all open connections and closes them just before it exits (regardless of whether the tasks failed or not.)
Library users will need to ensure they explicitly close all open connections before their program exits. This can be accomplished by calling disconnect_all at the end of your script.
Note
disconnect_all may be moved to a more public location in the future; we’re still working on making the library aspects of Fabric more solidified and organized.
Fabric maintains an in-memory, two-tier password cache to help remember your login and sudo passwords in certain situations; this helps avoid tedious re-entry when multiple systems share the same password [1], or if a remote system’s sudo configuration doesn’t do its own caching.
The first layer is a simple default or fallback password cache, env.password. This env var stores a single password which (if non-empty) will be tried in the event that the host-specific cache (see below) has no entry for the current host string.
env.passwords (plural!) serves as a per-user/per-host cache, storing the most recently entered password for every unique user/host/port combination. Due to this cache, connections to multiple different users and/or hosts in the same session will only require a single password entry for each. (Previous versions of Fabric used only the single, default password cache and thus required password re-entry every time the previously entered password became invalid.)
Depending on your configuration and the number of hosts your session will connect to, you may find setting either or both of these env vars to be useful. However, Fabric will automatically fill them in as necessary without any additional configuration.
Specifically, each time a password prompt is presented to the user, the value entered is used to update both the single default password cache, and the cache value for the current value of env.host_string.
| [1] | We highly recommend the use of SSH key-based access instead of relying on homogeneous password setups, as it’s significantly more secure. |
The most common method for utilizing Fabric is via its command-line tool, fab, which should have been placed on your shell’s executable path when Fabric was installed. fab tries hard to be a good Unix citizen, using a standard style of command-line switches, help output, and so forth.
In its most simple form, fab may be called with no options at all, and with one or more arguments, which should be task names, e.g.:
$ fab task1 task2
As detailed in Overview and Tutorial and Execution model, this will run task1 followed by task2, assuming that Fabric was able to find a fabfile nearby containing Python functions with those names.
However, it’s possible to expand this simple usage into something more flexible, by using the provided options and/or passing arguments to individual tasks.
New in version 0.9.2.
Fabric leverages a lesser-known command line convention and may be called in the following manner:
$ fab [options] -- [shell command]
where everything after the -- is turned into a temporary run call, and is not parsed for fab options. If you’ve defined a host list at the module level or on the command line, this usage will act like a one-line anonymous task.
For example, let’s say you just wanted to get the kernel info for a bunch of systems; you could do this:
$ fab -H system1,system2,system3 -- uname -a
which would be literally equivalent to the following fabfile:
from fabric.api import run
def anonymous():
run("uname -a")
as if it were executed thusly:
$ fab -H system1,system2,system3 anonymous
Most of the time you will want to just write out the task in your fabfile (anything you use once, you’re likely to use again) but this feature provides a handy, fast way to quickly dash off an SSH-borne command while leveraging your fabfile’s connection settings.
A quick overview of all possible command line options can be found via fab --help. If you’re looking for details on a specific option, we go into detail below.
Note
fab uses Python’s optparse library, meaning that it honors typical Linux or GNU style short and long options, as well as freely mixing options and arguments. E.g. fab task1 -H hostname task2 -i path/to/keyfile is just as valid as the more straightforward fab -H hostname -i path/to/keyfile task1 task2.
Sets env.no_agent to True, forcing our SSH layer not to talk to the SSH agent when trying to unlock private key files.
New in version 0.9.1.
Sets env.abort_on_prompts to True, forcing Fabric to abort whenever it would prompt for input.
New in version 1.1.
Sets env.rcfile to the given file path, which Fabric will try to load on startup and use to update environment variables.
Prints the entire docstring for the given task, if there is one. Does not currently print out the task’s function signature, so descriptive docstrings are a good idea. (They’re always a good idea, of course – just moreso here.)
Sets env.disable_known_hosts to True, preventing Fabric from loading the user’s SSH known_hosts file.
The fabfile name pattern to search for (defaults to fabfile.py), or alternately an explicit file path to load as the fabfile (e.g. /path/to/my/fabfile.py.)
See also
Allows control over the output format of --list. short is equivalent to --shortlist, normal is the same as simply omitting this option entirely (i.e. the default), and nested prints out a nested namespace tree.
New in version 1.1.
See also
Displays a standard help message, with all possible options and a brief overview of what they do, then exits.
A comma-separated list of output levels to hide by default.
Sets env.exclude_hosts to the given comma-delimited list of host strings to then keep out of the final host list.
New in version 1.1.
When set to a file path, will load the given file as an SSH identity file (usually a private key.) This option may be repeated multiple times. Sets (or appends to) env.key_filename.
Sets env.no_keys to True, forcing the SSH layer to not look for SSH private key files in one’s home directory.
New in version 0.9.1.
Sets env.keepalive to the given (integer) value, specifying an SSH keepalive interval.
New in version 1.1.
Forces output to be buffered line-by-line instead of byte-by-byte. Often useful or required for parallel execution.
New in version 1.3.
Imports a fabfile as normal, but then prints a list of all discovered tasks and exits. Will also print the first line of each task’s docstring, if it has one, next to it (truncating if necessary.)
Changed in version 0.9.1: Added docstring to output.
See also
Sets env.password to the given string; it will then be used as the default password when making SSH connections or calling the sudo program.
Sets env.parallel to True, causing tasks to run in parallel.
New in version 1.3.
See also
Sets env.always_use_pty to False, causing all run/sudo calls to behave as if one had specified pty=False.
New in version 1.0.
Sets env.reject_unknown_hosts to True, causing Fabric to abort when connecting to hosts not found in the user’s SSH known_hosts file.
Sets env.shell to the given string, overriding the default shell wrapper used to execute remote commands.
Similar to --list, but without any embellishment, just task names separated by newlines with no indentation or docstrings.
New in version 0.9.2.
See also
A comma-separated list of output levels to be added to those that are shown by default.
Sets env.user to the given string; it will then be used as the default username when making SSH connections.
Displays Fabric’s version number, then exits.
Sets env.warn_only to True, causing Fabric to continue execution even when commands encounter error conditions.
Sets env.pool_size, which specifies how many processes to run concurrently during parallel execution.
New in version 1.3.
See also
The options given in Command-line options apply to the invocation of fab as a whole; even if the order is mixed around, options still apply to all given tasks equally. Additionally, since tasks are just Python functions, it’s often desirable to pass in arguments to them at runtime.
Answering both these needs is the concept of “per-task arguments”, which is a special syntax you can tack onto the end of any task name:
Additionally, since this process involves string parsing, all values will end up as Python strings, so plan accordingly. (We hope to improve upon this in future versions of Fabric, provided an intuitive syntax can be found.)
For example, a “create a new user” task might be defined like so (omitting most of the actual logic for brevity):
def new_user(username, admin='no', comment="No comment provided"):
log_action("New User (%s): %s" % (username, comment))
pass
You can specify just the username:
$ fab new_user:myusername
Or treat it as an explicit keyword argument:
$ fab new_user:username=myusername
If both args are given, you can again give them as positional args:
$ fab new_user:myusername,yes
Or mix and match, just like in Python:
$ fab new_user:myusername,admin=yes
The log_action call above is useful for illustrating escaped commas, like so:
$ fab new_user:myusername,admin=no,comment='Gary\, new developer (starts Monday)'
Note
Quoting the backslash-escaped comma is required, as not doing so will cause shell syntax errors. Quotes are also needed whenever an argument involves other shell-related characters such as spaces.
All of the above are translated into the expected Python function calls. For example, the last call above would become:
>>> new_user('myusername', admin='yes', comment='Gary, new developer (starts Monday)')
As mentioned in the section on task execution, there are a handful of per-task keyword arguments (host, hosts, role and roles) which do not actually map to the task functions themselves, but are used for setting per-task host and/or role lists.
These special kwargs are removed from the args/kwargs sent to the task function itself; this is so that you don’t run into TypeErrors if your task doesn’t define the kwargs in question. (It also means that if you do define arguments with these names, you won’t be able to specify them in this manner – a regrettable but necessary sacrifice.)
Note
If both the plural and singular forms of these kwargs are given, the value of the plural will win out and the singular will be discarded.
When using the plural form of these arguments, one must use semicolons (;) since commas are already being used to separate arguments from one another. Furthermore, since your shell is likely to consider semicolons a special character, you’ll want to quote the host list string to prevent shell interpretation, e.g.:
$ fab new_user:myusername,hosts="host1;host2"
Again, since the hosts kwarg is removed from the argument list sent to the new_user task function, the actual Python invocation would be new_user('myusername'), and the function would be executed on a host list of ['host1', 'host2'].
Fabric currently honors a simple user settings file, or fabricrc (think bashrc but for fab) which should contain one or more key-value pairs, one per line. These lines will be subject to string.split('='), and thus can currently only be used to specify string settings. Any such key-value pairs will be used to update env when fab runs, and is loaded prior to the loading of any fabfile.
By default, Fabric looks for ~/.fabricrc, and this may be overridden by specifying the -c flag to fab.
For example, if your typical SSH login username differs from your workstation username, and you don’t want to modify env.user in a project’s fabfile (possibly because you expect others to use it as well) you could write a fabricrc file like so:
user = ssh_user_name
Then, when running fab, your fabfile would load up with env.user set to 'ssh_user_name'. Other users of that fabfile could do the same, allowing the fabfile itself to be cleanly agnostic regarding the default username.
This document contains miscellaneous sections about fabfiles, both how to best write them, and how to use them once written.
Fabric is capable of loading Python modules (e.g. fabfile.py) or packages (e.g. a fabfile/ directory containing an __init__.py). By default, it looks for something named either fabfile or fabfile.py.
The fabfile discovery algorithm searches in the invoking user’s current working directory or any parent directories. Thus, it is oriented around “project” use, where one keeps e.g. a fabfile.py at the root of a source code tree. Such a fabfile will then be discovered no matter where in the tree the user invokes fab.
The specific name to be searched for may be overridden on the command-line with the -f option, or by adding a fabricrc line which sets the value of fabfile. For example, if you wanted to name your fabfile fab_tasks.py, you could create such a file and then call fab -f fab_tasks.py <task name>, or add fabfile = fab_tasks.py to ~/.fabricrc.
If the given fabfile name contains path elements other than a filename (e.g. ../fabfile.py or /dir1/dir2/custom_fabfile) it will be treated as a file path and directly checked for existence without any sort of searching. When in this mode, tilde-expansion will be applied, so one may refer to e.g. ~/personal_fabfile.py.
Note
Fabric does a normal import (actually an __import__) of your fabfile in order to access its contents – it does not do any eval-ing or similar. In order for this to work, Fabric temporarily adds the found fabfile’s containing folder to the Python load path (and removes it immediately afterwards.)
Changed in version 0.9.2: The ability to load package fabfiles.
Because Fabric is just Python, you can import its components any way you want. However, for the purposes of encapsulation and convenience (and to make life easier for Fabric’s packaging script) Fabric’s public API is maintained in the fabric.api module.
All of Fabric’s Operations, Context Managers, Decorators and Utils are included in this module as a single, flat namespace. This enables a very simple and consistent interface to Fabric within your fabfiles:
from fabric.api import *
# call run(), sudo(), etc etc
This is not technically best practices (for a number of reasons) and if you’re only using a couple of Fab API calls, it is probably a good idea to explicitly from fabric.api import env, run or similar. However, in most nontrivial fabfiles, you’ll be using all or most of the API, and the star import:
from fabric.api import *
will be a lot easier to write and read than:
from fabric.api import abort, cd, env, get, hide, hosts, local, prompt, \
put, require, roles, run, runs_once, settings, show, sudo, warn
so in this case we feel pragmatism overrides best practices.
For important information on what exactly Fabric will consider as a task when it loads your fabfile, as well as notes on how best to import other code, please see Defining tasks in the Execution model documentation.
Fabric’s primary operations, run and sudo, are capable of sending local input to the remote end, in a manner nearly identical to the ssh program. For example, programs which display password prompts (e.g. a database dump utility, or changing a user’s password) will behave just as if you were interacting with them directly.
However, as with ssh itself, Fabric’s implementation of this feature is subject to a handful of limitations which are not always intuitive. This document discusses such issues in detail.
Note
Readers unfamiliar with the basics of Unix stdout and stderr pipes, and/or terminal devices, may wish to visit the Wikipedia pages for Unix pipelines and Pseudo terminals respectively.
The first issue to be aware of is that of the stdout and stderr streams, and why they are separated or combined as needed.
Fabric 0.9.x and earlier, and Python itself, buffer output on a line-by-line basis: text is not printed to the user until a newline character is found. This works fine in most situations but becomes problematic when one needs to deal with partial-line output such as prompts.
Note
Line-buffered output can make programs appear to halt or freeze for no reason, as prompts print out text without a newline, waiting for the user to enter their input and press Return.
Newer Fabric versions buffer both input and output on a character-by-character basis in order to make interaction with prompts possible. This has the convenient side effect of enabling interaction with complex programs utilizing the “curses” libraries or which otherwise redraw the screen (think top).
Unfortunately, printing to stderr and stdout simultaneously (as many programs do) means that when the two streams are printed independently one byte at a time, they can become garbled or meshed together. While this can sometimes be mitigated by line-buffering one of the streams and not the other, it’s still a serious issue.
To solve this problem, Fabric uses a setting in our SSH layer which merges the two streams at a low level and causes output to appear more naturally. This setting is represented in Fabric as the combine_stderr env var and keyword argument, and is True by default.
Due to this default setting, output will appear correctly, but at the cost of an empty .stderr attribute on the return values of run/sudo, as all output will appear to be stdout.
Conversely, users requiring a distinct stderr stream at the Python level and who aren’t bothered by garbled user-facing output (or who are hiding stdout and stderr from the command in question) may opt to set this to False as needed.
The other main issue to consider when presenting interactive prompts to users is that of echoing the user’s own input.
Typical terminal applications or bona fide text terminals (e.g. when using a Unix system without a running GUI) present programs with a terminal device called a tty or pty (for pseudo-terminal). These automatically echo all text typed into them back out to the user (via stdout), as interaction without seeing what you had just typed would be difficult. Terminal devices are also able to conditionally turn off echoing, allowing secure password prompts.
However, it’s possible for programs to be run without a tty or pty present at all (consider cron jobs, for example) and in this situation, any stdin data being fed to the program won’t be echoed. This is desirable for programs being run without any humans around, and it’s also Fabric’s old default mode of operation.
Unfortunately, in the context of executing commands via Fabric, when no pty is present to echo a user’s stdin, Fabric must echo it for them. This is sufficient for many applications, but it presents problems for password prompts, which become insecure.
In the interests of security and meeting the principle of least surprise (insofar as users are typically expecting things to behave as they would when run in a terminal emulator), Fabric 1.0 and greater force a pty by default. With a pty enabled, Fabric simply allows the remote end to handle echoing or hiding of stdin and does not echo anything itself.
Note
In addition to allowing normal echo behavior, a pty also means programs that behave differently when attached to a terminal device will then do so. For example, programs that colorize output on terminals but not when run in the background will print colored output. Be wary of this if you inspect the return value of run or sudo!
For situations requiring the pty behavior turned off, the --no-pty command-line argument and always_use_pty env var may be used.
As a final note, keep in mind that use of pseudo-terminals effectively implies combining stdout and stderr – in much the same way as the combine_stderr setting does. This is because a terminal device naturally sends both stdout and stderr to the same place – the user’s display – thus making it impossible to differentiate between them.
However, at the Fabric level, the two groups of settings are distinct from one another and may be combined in various ways. The default is for both to be set to True; the other combinations are as follows:
Fabric’s primary use case is via fabfiles and the fab tool, and this is reflected in much of the documentation. However, Fabric’s internals are written in such a manner as to be easily used without fab or fabfiles at all – this document will show you how.
There’s really only a couple of considerations one must keep in mind, when compared to writing a fabfile and using fab to run it: how connections are really made, and how disconnections occur.
We’ve documented how Fabric really connects to its hosts before, but it’s currently somewhat buried in the middle of the overall execution docs. Specifically, you’ll want to skip over to the Connections section and read it real quick. (You should really give that entire document a once-over, but it’s not absolutely required.)
As that section mentions, the key is simply that run, sudo and the other operations only look in one place when connecting: env.host_string. All of the other mechanisms for setting hosts are interpreted by the fab tool when it runs, and don’t matter when running as a library.
That said, most use cases where you want to marry a given task X and a given list of hosts Y can, as of Fabric 1.3, be handled with the execute function via execute(X, hosts=Y). Please see execute‘s documentation for details – manual host string manipulation should be rarely necessary.
The other main thing that fab does for you is to disconnect from all hosts at the end of a session; otherwise, Python will sit around forever waiting for those network resources to be released.
Fabric 0.9.4 and newer have a function you can use to do this easily: disconnect_all. Simply make sure your code calls this when it terminates (typically in the finally clause of an outer try: finally statement – lest errors in your code prevent disconnections from happening!) and things ought to work pretty well.
If you’re on Fabric 0.9.3 or older, you can simply do this (disconnect_all just adds a bit of nice output to this logic):
from fabric.state import connections
for key in connections.keys():
connections[key].close()
del connections[key]
This document is an early draft, and may not cover absolutely every difference between fab use and library use. However, the above should highlight the largest stumbling blocks. When in doubt, note that in the Fabric source code, fabric/main.py contains the bulk of the extra work done by fab, and may serve as a useful reference.
The fab tool is very verbose by default and prints out almost everything it can, including the remote end’s stderr and stdout streams, the command strings being executed, and so forth. While this is necessary in many cases in order to know just what’s going on, any nontrivial Fabric task will quickly become difficult to follow as it runs.
To aid in organizing task output, Fabric output is grouped into a number of non-overlapping levels or groups, each of which may be turned on or off independently. This provides flexible control over what is displayed to the user.
Note
All levels, save for debug, are on by default.
The standard, atomic output levels/groups are as follows:
Changed in version 0.9.2: Added “Executing task” lines to the running output level.
Changed in version 0.9.2: Added the user output level.
There is a final atomic output level, debug, which behaves slightly differently from the rest:
debug: Turn on debugging (which is off by default.) Currently, this is largely used to view the “full” commands being run; take for example this run call:
run('ls "/home/username/Folder Name With Spaces/"')
Normally, the running line will show exactly what is passed into run, like so:
[hostname] run: ls "/home/username/Folder Name With Spaces/"
With debug on, and assuming you’ve left shell set to True, you will see the literal, full string as passed to the remote server:
[hostname] run: /bin/bash -l -c "ls \"/home/username/Folder Name With Spaces\""
Enabling debug output will also display full Python tracebacks during aborts.
Note
Where modifying other pieces of output (such as in the above example where it modifies the ‘running’ line to show the shell and any escape characters), this setting takes precedence over the others; so if running is False but debug is True, you will still be shown the ‘running’ line in its debugging form.
Changed in version 1.0: Debug output now includes full Python tracebacks during aborts.
In addition to the atomic/standalone levels above, Fabric also provides a couple of convenience aliases which map to multiple other levels. These may be referenced anywhere the other levels are referenced, and will effectively toggle all of the levels they are mapped to.
You may toggle any of Fabric’s output levels in a number of ways; for examples, please see the API docs linked in each bullet point:
Direct modification of fabric.state.output: fabric.state.output is a dictionary subclass (similar to env) whose keys are the output level names, and whose values are either True (show that particular type of output) or False (hide it.)
fabric.state.output is the lowest-level implementation of output levels and is what Fabric’s internals reference when deciding whether or not to print their output.
Context managers: hide and show are twin context managers that take one or more output level names as strings, and either hide or show them within the wrapped block. As with Fabric’s other context managers, the prior values are restored when the block exits.
Command-line arguments: You may use the --hide and/or --show arguments to fab options and arguments, which behave exactly like the context managers of the same names (but are, naturally, globally applied) and take comma-separated strings as input.
New in version 1.3.
By default, Fabric executes all specified tasks serially (see Execution strategy for details.) This document describes Fabric’s options for running tasks on multiple hosts in parallel, via per-task decorators and/or global command-line switches.
Because Fabric 1.x is not fully threadsafe (and because in general use, task functions do not typically interact with one another) this functionality is implemented via the Python multiprocessing module. It creates one new process for each host and task combination, optionally using a (configurable) sliding window to prevent too many processes from running at the same time.
For example, imagine a scenario where you want to update Web application code on a number of Web servers, and then reload the servers once the code has been distributed everywhere (to allow for easier rollback if code updates fail.) One could implement this with the following fabfile:
from fabric.api import *
def update():
with cd("/srv/django/myapp"):
run("git pull")
def reload():
sudo("service apache2 reload")
and execute it on a set of 3 servers, in serial, like so:
$ fab -H web1,web2,web3 update reload
Normally, without any parallel execution options activated, Fabric would run in order:
With parallel execution activated (via -P – see below for details), this turns into:
Hopefully the benefits of this are obvious – if update took 5 seconds to run and reload took 2 seconds, serial execution takes (5+2)*3 = 21 seconds to run, while parallel execution takes only a third of the time, (5+2) = 7 seconds on average.
Since the minimum “unit” that parallel execution affects is a task, the functionality may be enabled or disabled on a task-by-task basis using the parallel and serial decorators. For example, this fabfile:
from fabric.api import *
@parallel
def runs_in_parallel():
pass
def runs_serially():
pass
when run in this manner:
$ fab -H host1,host2,host3 runs_in_parallel runs_serially
will result in the following execution sequence:
One may also force all tasks to run in parallel by using the command-line flag -P or the env variable env.parallel. However, any task specifically wrapped with serial will ignore this setting and continue to run serially.
For example, the following fabfile will result in the same execution sequence as the one above:
from fabric.api import *
def runs_in_parallel():
pass
@serial
def runs_serially():
pass
when invoked like so:
$ fab -H host1,host2,host3 -P runs_in_parallel runs_serially
As before, runs_in_parallel will run in parallel, and runs_serially in sequence.
With large host lists, a user’s local machine can get overwhelmed by running too many concurrent Fabric processes. Because of this, you may opt to use a moving bubble approach that limits Fabric to a specific number of concurrently active processes.
By default, no bubble is used and all hosts are run in one concurrent pool. You can override this on a per-task level by specifying the pool_size keyword argument to parallel, or globally via -z.
For example, to run on 5 hosts at a time:
from fabric.api import *
@parallel(pool_size=5)
def heavy_task():
# lots of heavy local lifting or lots of IO here
Or skip the pool_size kwarg and instead:
$ fab -P -z 5 heavy_task
Fabric’s default mode of printing to the terminal is byte-by-byte, in order to support Interaction with remote programs. This often gives poor results when running in parallel mode, as the multiple processes may write to your terminal’s standard out stream simultaneously.
To help offset this problem, Fabric’s option for linewise output is automatically enabled whenever parallelism is active. This will cause you to lose most of the benefits outlined in the above link Fabric’s remote interactivity features, but as those do not map well to parallel invocations, it’s typically a fair trade.
There’s no way to avoid the multiple processes mixing up on a line-by-line basis, but you will at least be able to tell them apart by the host-string line prefix.
Note
Future versions will add improved logging support to make troubleshooting parallel runs easier.
Fabric currently makes use of a pure-Python SSH re-implementation for managing connections, meaning that there are occasionally spots where it is limited by that library’s capabilities. Below are areas of note where Fabric will exhibit behavior that isn’t consistent with, or as flexible as, the behavior of the ssh command-line program.
SSH’s host key tracking mechanism keeps tabs on all the hosts you attempt to connect to, and maintains a ~/.ssh/known_hosts file with mappings between identifiers (IP address, sometimes with a hostname as well) and SSH keys. (For details on how this works, please see the OpenSSH documentation.)
The ssh library is capable of loading up your known_hosts file, and will then compare any host it connects to, with that mapping. Settings are available to determine what happens when an unknown host (a host whose username or IP is not found in known_hosts) is seen:
Whether to reject or add hosts, as above, is controlled in Fabric via the env.reject_unknown_hosts option, which is False by default for convenience’s sake. We feel this is a valid tradeoff between convenience and security; anyone who feels otherwise can easily modify their fabfiles at module level to set env.reject_unknown_hosts = True.
The point of SSH’s key/fingerprint tracking is so that man-in-the-middle attacks can be detected: if an attacker redirects your SSH traffic to a computer under his control, and pretends to be your original destination server, the host keys will not match. Thus, the default behavior of SSH (and its Python implementation) is to immediately abort the connection when a host previously recorded in known_hosts suddenly starts sending us a different host key.
In some edge cases such as some EC2 deployments, you may want to ignore this potential problem. Our SSH layer, at the time of writing, doesn’t give us control over this exact behavior, but we can sidestep it by simply skipping the loading of known_hosts – if the host list being compared to is empty, then there’s no problem. Set env.disable_known_hosts to True when you want this behavior; it is False by default, in order to preserve default SSH behavior.
Warning
Enabling env.disable_known_hosts will leave you wide open to man-in-the-middle attacks! Please use with caution.
As of Fabric 1.1, there are two distinct methods you may use in order to define which objects in your fabfile show up as tasks:
Note
These two methods are mutually exclusive: if Fabric finds any new-style task objects in your fabfile or in modules it imports, it will assume you’ve committed to this method of task declaration and won’t consider any non-Task callables. If no new-style tasks are found, it reverts to the classic behavior.
The rest of this document explores these two methods in detail.
Note
To see exactly what tasks in your fabfile may be executed via fab, use fab --list.
Fabric 1.1 introduced the Task class to facilitate new features and enable some programming best practices, specifically:
With the introduction of Task, there are two ways to set up new tasks:
Use of new-style tasks also allows you to set up namespaces.
The quickest way to make use of new-style task features is to wrap basic task functions with @task:
from fabric.api import task, run
@task
def mytask():
run("a command")
When this decorator is used, it signals to Fabric that only functions wrapped in the decorator are to be loaded up as valid tasks. (When not present, classic-style task behavior kicks in.)
@task may also be called with arguments to customize its behavior. Any arguments not documented below are passed into the constructor of the task_class being used, with the function itself as the first argument (see Using custom subclasses with @task for details.)
Here’s a quick example of using the alias keyword argument to facilitate use of both a longer human-readable task name, and a shorter name which is quicker to type:
from fabric.api import task
@task(alias='dwm')
def deploy_with_migrations():
pass
Calling --list on this fabfile would show both the original deploy_with_migrations and its alias dwm:
$ fab --list
Available commands:
deploy_with_migrations
dwm
When more than one alias for the same function is needed, simply swap in the aliases kwarg, which takes an iterable of strings instead of a single string.
In a similar manner to aliases, it’s sometimes useful to designate a given task within a module as the “default” task, which may be called by referencing just the module name. This can save typing and/or allow for neater organization when there’s a single “main” task and a number of related tasks or subroutines.
For example, a deploy submodule might contain tasks for provisioning new servers, pushing code, migrating databases, and so forth – but it’d be very convenient to highlight a task as the default “just deploy” action. Such a deploy.py module might look like this:
from fabric.api import task
@task
def migrate():
pass
@task
def push():
pass
@task
def provision():
pass
@task
def full_deploy():
if not provisioned:
provision()
push()
migrate()
With the following task list (assuming a simple top level fabfile.py that just imports deploy):
$ fab --list
Available commands:
deploy.full_deploy
deploy.migrate
deploy.provision
deploy.push
Calling deploy.full_deploy on every deploy could get kind of old, or somebody new to the team might not be sure if that’s really the right task to run.
Using the default kwarg to @task, we can tag e.g. full_deploy as the default task:
@task(default=True)
def full_deploy():
pass
Doing so updates the task list like so:
$ fab --list
Available commands:
deploy
deploy.full_deploy
deploy.migrate
deploy.provision
deploy.push
Note that full_deploy still exists as its own explicit task – but now deploy shows up as a sort of top level alias for full_deploy.
If multiple tasks within a module have default=True set, the last one to be loaded (typically the one lowest down in the file) will take precedence.
Using @task(default=True) in the top level fabfile will cause the denoted task to execute when a user invokes fab without any task names (similar to e.g. make.) When using this shortcut, it is not possible to specify arguments to the task itself – use a regular invocation of the task if this is necessary.
If you’re used to classic-style tasks, an easy way to think about Task subclasses is that their run method is directly equivalent to a classic task; its arguments are the task arguments (other than self) and its body is what gets executed.
For example, this new-style task:
class MyTask(Task):
name = "deploy"
def run(self, environment, domain="whatever.com"):
run("git clone foo")
sudo("service apache2 restart")
instance = MyTask()
is exactly equivalent to this function-based task:
@task
def deploy(environment, domain="whatever.com"):
run("git clone foo")
sudo("service apache2 restart")
Note how we had to instantiate an instance of our class; that’s simply normal Python object-oriented programming at work. While it’s a small bit of boilerplate right now – for example, Fabric doesn’t care about the name you give the instantiation, only the instance’s name attribute – it’s well worth the benefit of having the power of classes available.
We plan to extend the API in the future to make this experience a bit smoother.
It’s possible to marry custom Task subclasses with @task. This may be useful in cases where your core execution logic doesn’t do anything class/object-specific, but you want to take advantage of class metaprogramming or similar techniques.
Specifically, any Task subclass which is designed to take in a callable as its first constructor argument (as the built-in WrappedCallableTask does) may be specified as the task_class argument to @task.
Fabric will automatically instantiate a copy of the given class, passing in the wrapped function as the first argument. All other args/kwargs given to the decorator (besides the “special” arguments documented in Arguments) are added afterwards.
Here’s a brief and somewhat contrived example to make this obvious:
from fabric.api import task
from fabric.tasks import Task
class CustomTask(Task):
def __init__(self, func, myarg, *args, **kwargs):
super(CustomTask, self).__init__(*args, **kwargs)
self.func = func
self.myarg = myarg
def run(self, *args, **kwargs):
return self.func(*args, **kwargs)
@task(task_class=CustomTask, myarg='value', alias='at')
def actual_task():
pass
When this fabfile is loaded, a copy of CustomTask is instantiated, effectively calling:
task_obj = CustomTask(actual_task, myarg='value')
Note how the alias kwarg is stripped out by the decorator itself and never reaches the class instantiation; this is identical in function to how command-line task arguments work.
With classic tasks, fabfiles were limited to a single, flat set of task names with no real way to organize them. In Fabric 1.1 and newer, if you declare tasks the new way (via @task or your own Task subclass instances) you may take advantage of namespacing:
Let’s build up a fabfile package from simple to complex and see how this works.
We start with a single __init__.py containing a few tasks (the Fabric API import omitted for brevity):
@task
def deploy():
...
@task
def compress():
...
The output of fab --list would look something like this:
deploy
compress
There’s just one namespace here: the “root” or global namespace. Looks simple now, but in a real-world fabfile with dozens of tasks, it can get difficult to manage.
As mentioned above, Fabric will examine any imported module objects for tasks, regardless of where that module exists on your Python import path. For now we just want to include our own, “nearby” tasks, so we’ll make a new submodule in our package for dealing with, say, load balancers – lb.py:
@task
def add_backend():
...
And we’ll add this to the top of __init__.py:
import lb
Now fab --list shows us:
deploy
compress
lb.add_backend
Again, with only one task in its own submodule, it looks kind of silly, but the benefits should be pretty obvious.
Namespacing isn’t limited to just one level. Let’s say we had a larger setup and wanted a namespace for database related tasks, with additional differentiation inside that. We make a sub-package named db/ and inside it, a migrations.py module:
@task
def list():
...
@task
def run():
...
We need to make sure that this module is visible to anybody importing db, so we add it to the sub-package’s __init__.py:
import migrations
As a final step, we import the sub-package into our root-level __init__.py, so now its first few lines look like this:
import lb
import db
After all that, our file tree looks like this:
.
├── __init__.py
├── db
│ ├── __init__.py
│ └── migrations.py
└── lb.py
and fab --list shows:
deploy
compress
lb.add_backend
db.migrations.list
db.migrations.run
We could also have specified (or imported) tasks directly into db/__init__.py, and they would show up as db.<whatever> as you might expect.
You may limit what Fabric “sees” when it examines imported modules, by using the Python convention of a module level __all__ variable (a list of variable names.) If we didn’t want the db.migrations.run task to show up by default for some reason, we could add this to the top of db/migrations.py:
__all__ = ['list']
Note the lack of 'run' there. You could, if needed, import run directly into some other part of the hierarchy, but otherwise it’ll remain hidden.
We’ve been keeping our fabfile package neatly organized and importing it in a straightforward manner, but the filesystem layout doesn’t actually matter here. All Fabric’s loader cares about is the names the modules are given when they’re imported.
For example, if we changed the top of our root __init__.py to look like this:
import db as database
Our task list would change thusly:
deploy
compress
lb.add_backend
database.migrations.list
database.migrations.run
This applies to any other import – you could import third party modules into your own task hierarchy, or grab a deeply nested module and make it appear near the top level.
As a final note, we’ve been using the default Fabric --list output during this section – it makes it more obvious what the actual task names are. However, you can get a more nested or tree-like view by passing nested to the --list-format option:
$ fab --list-format=nested --list
Available commands (remember to call as module.[...].task):
deploy
compress
lb:
add_backend
database:
migrations:
list
run
While it slightly obfuscates the “real” task names, this view provides a handy way of noting the organization of tasks in large namespaces.
When no new-style Task-based tasks are found, Fabric will consider any callable object found in your fabfile, except the following:
Python’s import statement effectively includes the imported objects in your module’s namespace. Since Fabric’s fabfiles are just Python modules, this means that imports are also considered as possible classic-style tasks, alongside anything defined in the fabfile itself.
Note
This only applies to imported callable objects – not modules. Imported modules only come into play if they contain new-style tasks, at which point this section no longer applies.
Because of this, we strongly recommend that you use the import module form of importing, followed by module.callable(), which will result in a cleaner fabfile API than doing from module import callable.
For example, here’s a sample fabfile which uses urllib.urlopen to get some data out of a webservice:
from urllib import urlopen
from fabric.api import run
def webservice_read():
objects = urlopen('http://my/web/service/?foo=bar').read().split()
print(objects)
This looks simple enough, and will run without error. However, look what happens if we run fab --list on this fabfile:
$ fab --list
Available commands:
webservice_read List some directories.
urlopen urlopen(url [, data]) -> open file-like object
Our fabfile of only one task is showing two “tasks”, which is bad enough, and an unsuspecting user might accidentally try to call fab urlopen, which probably won’t work very well. Imagine any real-world fabfile, which is likely to be much more complex, and hopefully you can see how this could get messy fast.
For reference, here’s the recommended way to do it:
import urllib
from fabric.api import run
def webservice_read():
objects = urllib.urlopen('http://my/web/service/?foo=bar').read().split()
print(objects)
It’s a simple change, but it’ll make anyone using your fabfile a bit happier.
Some frequently encountered questions, coupled with answers/solutions/excuses, may be found on the Frequently Asked Questions (FAQ) page.
Fabric maintains two sets of API documentation, autogenerated from the source code’s docstrings (which are typically very thorough.)
The core API is loosely defined as those functions, classes and methods which form the basic building blocks of Fabric (such as run and sudo) upon which everything else (the below “contrib” section, and user fabfiles) builds.
New in version 0.9.2.
Functions for wrapping strings in ANSI color codes.
Each function within this module returns the input string text, wrapped with ANSI color codes for the appropriate color.
For example, to print some text as green on supporting terminals:
from fabric.colors import green
print(green("This text is green!"))
Because these functions simply return modified strings, you can nest them:
from fabric.colors import red, green
print(red("This sentence is red, except for " + green("these words, which are green") + "."))
If bold is set to True, the ANSI flag for bolding will be flipped on for that particular invocation, which usually shows up as a bold or brighter version of the original color on most terminals.
Context managers for use with the with statement.
Note
When using Python 2.5, you will need to start your fabfile with from __future__ import with_statement in order to make use of the with statement (which is a regular, non __future__ feature of Python 2.6+.)
Note
If you are using multiple directly nested with statements, it can be convenient to use multiple context expressions in one single with statement. Instead of writing:
with cd('/path/to/app'):
with prefix('workon myvenv'):
run('./manage.py syncdb')
run('./manage.py loaddata myfixture')
you can write:
with cd('/path/to/app'), prefix('workon myvenv'):
run('./manage.py syncdb')
run('./manage.py loaddata myfixture')
Note that you need Python 2.6+ for this to work. On Python 2.5, you can do the following:
from contextlib import nested
with nested(cd('/path/to/app'), prefix('workon myvenv')):
...
Finally, note that settings implements nested itself – see its API doc for details.
Context manager that keeps directory state when calling remote operations.
Any calls to run, sudo, get, or put within the wrapped block will implicitly have a string similar to "cd <path> && " prefixed in order to give the sense that there is actually statefulness involved.
Because use of cd affects all such invocations, any code making use of those operations, such as much of the contrib section, will also be affected by use of cd.
Like the actual ‘cd’ shell builtin, cd may be called with relative paths (keep in mind that your default starting directory is your remote user’s $HOME) and may be nested as well.
Below is a “normal” attempt at using the shell ‘cd’, which doesn’t work due to how shell-less SSH connections are implemented – state is not kept between invocations of run or sudo:
run('cd /var/www')
run('ls')
The above snippet will list the contents of the remote user’s $HOME instead of /var/www. With cd, however, it will work as expected:
with cd('/var/www'):
run('ls') # Turns into "cd /var/www && ls"
Finally, a demonstration (see inline comments) of nesting:
with cd('/var/www'):
run('ls') # cd /var/www && ls
with cd('website1'):
run('ls') # cd /var/www/website1 && ls
Note
This context manager is currently implemented by appending to (and, as always, restoring afterwards) the current value of an environment variable, env.cwd. However, this implementation may change in the future, so we do not recommend manually altering env.cwd – only the behavior of cd will have any guarantee of backwards compatibility.
Note
Space characters will be escaped automatically to make dealing with such directory names easier.
Changed in version 1.0: Applies to get and put in addition to the command-running operations.
See also
Context manager for setting the given output groups to False.
groups must be one or more strings naming the output groups defined in output. The given groups will be set to False for the duration of the enclosed block, and restored to their previous value afterwards.
For example, to hide the “[hostname] run:” status lines, as well as preventing printout of stdout and stderr, one might use hide as follows:
def my_task():
with hide('running', 'stdout', 'stderr'):
run('ls /var/www')
Context manager for updating local current working directory.
This context manager is identical to cd, except that it changes a different env var (lcwd, instead of cwd) and thus only affects the invocation of local and the local arguments to get/put.
Relative path arguments are relative to the local user’s current working directory, which will vary depending on where Fabric (or Fabric-using code) was invoked. You can check what this is with os.getcwd. It may be useful to pin things relative to the location of the fabfile in use, which may be found in env.real_fabfile
New in version 1.0.
Append the given path to the PATH used to execute any wrapped commands.
Any calls to run or sudo within the wrapped block will implicitly have a string similar to "PATH=$PATH:<path> " prepended before the given command.
You may customize the behavior of path by specifying the optional behavior keyword argument, as follows:
Note
This context manager is currently implemented by modifying (and, as always, restoring afterwards) the current value of environment variables, env.path and env.path_behavior. However, this implementation may change in the future, so we do not recommend manually altering them directly.
New in version 1.0.
Prefix all wrapped run/sudo commands with given command plus &&.
This is nearly identical to cd, except that nested invocations append to a list of command strings instead of modifying a single string.
Most of the time, you’ll want to be using this alongside a shell script which alters shell state, such as ones which export or alter shell environment variables.
For example, one of the most common uses of this tool is with the workon command from virtualenvwrapper:
with prefix('workon myvenv'):
run('./manage.py syncdb')
In the above snippet, the actual shell command run would be this:
$ workon myvenv && ./manage.py syncdb
This context manager is compatible with cd, so if your virtualenv doesn’t cd in its postactivate script, you could do the following:
with cd('/path/to/app'):
with prefix('workon myvenv'):
run('./manage.py syncdb')
run('./manage.py loaddata myfixture')
Which would result in executions like so:
$ cd /path/to/app && workon myvenv && ./manage.py syncdb
$ cd /path/to/app && workon myvenv && ./manage.py loaddata myfixture
Finally, as alluded to near the beginning, prefix may be nested if desired, e.g.:
with prefix('workon myenv'):
run('ls')
with prefix('source /some/script'):
run('touch a_file')
The result:
$ workon myenv && ls
$ workon myenv && source /some/script && touch a_file
Contrived, but hopefully illustrative.
Nest context managers and/or override env variables.
settings serves two purposes:
These behaviors may be specified at the same time if desired. An example will hopefully illustrate why this is considered useful:
def my_task():
with settings(
hide('warnings', 'running', 'stdout', 'stderr'),
warn_only=True
):
if run('ls /etc/lsb-release'):
return 'Ubuntu'
elif run('ls /etc/redhat-release'):
return 'RedHat'
The above task executes a run statement, but will warn instead of aborting if the ls fails, and all output – including the warning itself – is prevented from printing to the user. The end result, in this scenario, is a completely silent task that allows the caller to figure out what type of system the remote host is, without incurring the handful of output that would normally occur.
Thus, settings may be used to set any combination of environment variables in tandem with hiding (or showing) specific levels of output, or in tandem with any other piece of Fabric functionality implemented as a context manager.
Context manager for setting the given output groups to True.
groups must be one or more strings naming the output groups defined in output. The given groups will be set to True for the duration of the enclosed block, and restored to their previous value afterwards.
For example, to turn on debug output (which is typically off by default):
def my_task():
with show('debug'):
run('ls /var/www')
As almost all output groups are displayed by default, show is most useful for turning on the normally-hidden debug group, or when you know or suspect that code calling your own code is trying to hide output with hide.
Convenience decorators for use in fabfiles.
Decorator defining which host or hosts to execute the wrapped function on.
For example, the following will ensure that, barring an override on the command line, my_func will be run on host1, host2 and host3, and with specific users on host1 and host3:
@hosts('user1@host1', 'host2', 'user2@host3')
def my_func():
pass
hosts may be invoked with either an argument list (@hosts('host1'), @hosts('host1', 'host2')) or a single, iterable argument (@hosts(['host1', 'host2'])).
Note that this decorator actually just sets the function’s .hosts attribute, which is then read prior to executing the function.
Changed in version 0.9.2: Allow a single, iterable argument (@hosts(iterable)) to be used instead of requiring @hosts(*iterable).
Decorator defining a list of role names, used to look up host lists.
A role is simply defined as a key in env whose value is a list of one or more host connection strings. For example, the following will ensure that, barring an override on the command line, my_func will be executed against the hosts listed in the webserver and dbserver roles:
env.roledefs.update({
'webserver': ['www1', 'www2'],
'dbserver': ['db1']
})
@roles('webserver', 'dbserver')
def my_func():
pass
As with hosts, roles may be invoked with either an argument list or a single, iterable argument. Similarly, this decorator uses the same mechanism as hosts and simply sets <function>.roles.
Changed in version 0.9.2: Allow a single, iterable argument to be used (same as hosts).
Decorator preventing wrapped function from running more than once.
By keeping internal state, this decorator allows you to mark a function such that it will only run once per Python interpreter session, which in typical use means “once per invocation of the fab program”.
Any function wrapped with this decorator will silently fail to execute the 2nd, 3rd, ..., Nth time it is called, and will return the value of the original run.
Warning
This decorator is not compatible with Fabric’s parallel execution mode; when used alongside parallel or -P, or when decorating subtasks of parallel tasks, each parallel copy of the decorated task will itself run one time, resulting in multiple runs.
Forces the wrapped function to always run sequentially, never in parallel.
This decorator takes precedence over the global value of env.parallel. However, if a task is decorated with both serial and parallel, parallel wins.
New in version 1.3.
Forces the wrapped function to run in parallel, instead of sequentially.
This decorator takes precedence over the global value of env.parallel. It also takes precedence over serial if a task is decorated with both.
New in version 1.3.
Decorator declaring the wrapped function to be a new-style task.
May be invoked as a simple, argument-less decorator (i.e. @task) or with arguments customizing its behavior (e.g. @task(alias='myalias')).
Please see the new-style task documentation for details on how to use this decorator.
Changed in version 1.2: Added the alias, aliases, task_class and default keyword arguments. See Arguments for details.
Decorator equivalent of fabric.context_managers.settings.
Allows you to wrap an entire function as if it was called inside a block with the settings context manager. This may be useful if you know you want a given setting applied to an entire function body, or wish to retrofit old code without indenting everything.
For example, to turn aborts into warnings for an entire task function:
@with_settings(warn_only=True)
def foo():
...
See also
New in version 1.1.
Classes and subroutines dealing with network connections and related topics.
Disconnect from all currently connected servers.
Used at the end of fab‘s main loop, and also intended for use by library users.
Functions to be used in fabfiles and other non-core code, such as run()/sudo().
Download one or more files from a remote host.
get returns an iterable containing the absolute paths to all local files downloaded, which will be empty if local_path was a StringIO object (see below for more on using StringIO). This object will also exhibit a .failed attribute containing any remote file paths which failed to download, and a .succeeded attribute equivalent to not .failed.
remote_path is the remote file or directory path to download, which may contain shell glob syntax, e.g. "/var/log/apache2/*.log", and will have tildes replaced by the remote home directory. Relative paths will be considered relative to the remote user’s home directory, or the current remote working directory as manipulated by cd. If the remote path points to a directory, that directory will be downloaded recursively.
local_path is the local file path where the downloaded file or files will be stored. If relative, it will honor the local current working directory as manipulated by lcd. It may be interpolated, using standard Python dict-based interpolation, with the following variables:
Note
When remote_path is an absolute directory path, only the inner directories will be recreated locally and passed into the above variables. So for example, get('/var/log', '%(path)s') would start writing out files like apache2/access.log, postgresql/8.4/postgresql.log, etc, in the local working directory. It would not write out e.g. var/log/apache2/access.log.
Additionally, when downloading a single file, %(dirname)s and %(path)s do not make as much sense and will be empty and equivalent to %(basename)s, respectively. Thus a call like get('/var/log/apache2/access.log', '%(path)s') will save a local file named access.log, not var/log/apache2/access.log.
This behavior is intended to be consistent with the command-line scp program.
If left blank, local_path defaults to "%(host)s/%(path)s" in order to be safe for multi-host invocations.
Warning
If your local_path argument does not contain %(host)s and your get call runs against multiple hosts, your local files will be overwritten on each successive run!
If local_path does not make use of the above variables (i.e. if it is a simple, explicit file path) it will act similar to scp or cp, overwriting pre-existing files if necessary, downloading into a directory if given (e.g. get('/path/to/remote_file.txt', 'local_directory') will create local_directory/remote_file.txt) and so forth.
local_path may alternately be a file-like object, such as the result of open('path', 'w') or a StringIO instance.
Note
Attempting to get a directory into a file-like object is not valid and will result in an error.
Note
This function will use seek and tell to overwrite the entire contents of the file-like object, in order to be consistent with the behavior of put (which also considers the entire file). However, unlike put, the file pointer will not be restored to its previous location, as that doesn’t make as much sense here and/or may not even be possible.
Note
Due to how our SSH layer works, a temporary file will still be written to your hard disk even if you specify a file-like object such as a StringIO for the local_path argument. Cleanup is performed, however – we just note this for users expecting straight-to-memory transfers. (We hope to patch our SSH layer in the future to enable true straight-to-memory downloads.)
Changed in version 1.0: Now honors the remote working directory as manipulated by cd, and the local working directory as manipulated by lcd.
Changed in version 1.0: Now allows file-like objects in the local_path argument.
Changed in version 1.0: local_path may now contain interpolated path- and host-related variables.
Changed in version 1.0: Directories may be specified in the remote_path argument and will trigger recursive downloads.
Changed in version 1.0: Return value is now an iterable of downloaded local file paths, which also exhibits the .failed and .succeeded attributes.
Invoke a fully interactive shell on the remote end.
If command is given, it will be sent down the pipe before handing control over to the invoking user.
This function is most useful for when you need to interact with a heavily shell-based command or series of commands, such as when debugging or when fully interactive recovery is required upon remote program failure.
It should be considered an easy way to work an interactive shell session into the middle of a Fabric script and is not a drop-in replacement for run, which is also capable of interacting with the remote end (albeit only while its given command is executing) and has much stronger programmatic abilities such as error handling and stdout/stderr capture.
Specifically, open_shell provides a better interactive experience than run, but use of a full remote shell prevents Fabric from determining whether programs run within the shell have failed, and pollutes the stdout/stderr stream with shell output such as login banners, prompts and echoed stdin.
Thus, this function does not have a return value and will not trigger Fabric’s failure handling if any remote programs result in errors.
New in version 1.0.
Upload one or more files to a remote host.
put returns an iterable containing the absolute file paths of all remote files uploaded. This iterable also exhibits a .failed attribute containing any local file paths which failed to upload (and may thus be used as a boolean test.) You may also check .succeeded which is equivalent to not .failed.
local_path may be a relative or absolute local file or directory path, and may contain shell-style wildcards, as understood by the Python glob module. Tilde expansion (as implemented by os.path.expanduser) is also performed.
local_path may alternately be a file-like object, such as the result of open('path') or a StringIO instance.
Note
In this case, put will attempt to read the entire contents of the file-like object by rewinding it using seek (and will use tell afterwards to preserve the previous file position).
Note
Use of a file-like object in put‘s local_path argument will cause a temporary file to be utilized due to limitations in our SSH layer’s API.
remote_path may also be a relative or absolute location, but applied to the remote host. Relative paths are relative to the remote user’s home directory, but tilde expansion (e.g. ~/.ssh/) will also be performed if necessary.
An empty string, in either path argument, will be replaced by the appropriate end’s current working directory.
While the SFTP protocol (which put uses) has no direct ability to upload files to locations not owned by the connecting user, you may specify use_sudo=True to work around this. When set, this setting causes put to upload the local files to a temporary location on the remote end, and then use sudo to move them to remote_path.
In some use cases, it is desirable to force a newly uploaded file to match the mode of its local counterpart (such as when uploading executable scripts). To do this, specify mirror_local_mode=True.
Alternately, you may use the mode kwarg to specify an exact mode, in the same vein as os.chmod or the Unix chmod command.
put will honor cd, so relative values in remote_path will be prepended by the current remote working directory, if applicable. Thus, for example, the below snippet would attempt to upload to /tmp/files/test.txt instead of ~/files/test.txt:
with cd('/tmp'):
put('/path/to/local/test.txt', 'files')
Use of lcd will affect local_path in the same manner.
Examples:
put('bin/project.zip', '/tmp/project.zip')
put('*.py', 'cgi-bin/')
put('index.html', 'index.html', mode=0755)
Changed in version 1.0: Now honors the remote working directory as manipulated by cd, and the local working directory as manipulated by lcd.
Changed in version 1.0: Now allows file-like objects in the local_path argument.
Changed in version 1.0: Directories may be specified in the local_path argument and will trigger recursive uploads.
Changed in version 1.0: Return value is now an iterable of uploaded remote file paths which also exhibits the .failed and .succeeded attributes.
Reboot the remote system, disconnect, and wait for wait seconds.
After calling this operation, further execution of run or sudo will result in a normal reconnection to the server, including any password prompts.
New in version 0.9.2.
Run a shell command on a remote host.
If shell is True (the default), run will execute the given command string via a shell interpreter, the value of which may be controlled by setting env.shell (defaulting to something similar to /bin/bash -l -c "<command>".) Any double-quote (") or dollar-sign ($) characters in command will be automatically escaped when shell is True.
run will return the result of the remote program’s stdout as a single (likely multiline) string. This string will exhibit failed and succeeded boolean attributes specifying whether the command failed or succeeded, and will also include the return code as the return_code attribute.
Any text entered in your local terminal will be forwarded to the remote program as it runs, thus allowing you to interact with password or other prompts naturally. For more on how this works, see Interaction with remote programs.
You may pass pty=False to forego creation of a pseudo-terminal on the remote end in case the presence of one causes problems for the command in question. However, this will force Fabric itself to echo any and all input you type while the command is running, including sensitive passwords. (With pty=True, the remote pseudo-terminal will echo for you, and will intelligently handle password-style prompts.) See Pseudo-terminals for details.
Similarly, if you need to programmatically examine the stderr stream of the remote program (exhibited as the stderr attribute on this function’s return value), you may set combine_stderr=False. Doing so has a high chance of causing garbled output to appear on your terminal (though the resulting strings returned by run will be properly separated). For more info, please read Combining stdout and stderr.
Examples:
run("ls /var/www/")
run("ls /home/myuser", shell=False)
output = run('ls /var/www/site1')
New in version 1.0: The succeeded and stderr return value attributes, the combine_stderr kwarg, and interactive behavior.
Changed in version 1.0: The default value of pty is now True.
Changed in version 1.0.2: The default value of combine_stderr is now None instead of True. However, the default behavior is unchanged, as the global setting is still True.
Run a shell command on a remote host, with superuser privileges.
sudo is identical in every way to run, except that it will always wrap the given command in a call to the sudo program to provide superuser privileges.
sudo accepts an additional user argument, which is passed to sudo and allows you to run as some user other than root. On most systems, the sudo program can take a string username or an integer userid (uid); user may likewise be a string or an int.
Examples:
sudo("~/install_script.py")
sudo("mkdir /var/www/new_docroot", user="www-data")
sudo("ls /home/jdoe", user=1001)
result = sudo("ls /tmp/")
Changed in version 1.0: See the changed and added notes for run.
Run a command on the local system.
local is simply a convenience wrapper around the use of the builtin Python subprocess module with shell=True activated. If you need to do anything special, consider using the subprocess module directly.
local is not currently capable of simultaneously printing and capturing output, as run/sudo do. The capture kwarg allows you to switch between printing and capturing as necessary, and defaults to False.
When capture=False, the local subprocess’ stdout and stderr streams are hooked up directly to your terminal, though you may use the global output controls output.stdout and output.stderr to hide one or both if desired. In this mode, local returns None.
When capture=True, this function will return the contents of the command’s stdout as a string-like object; as with run and sudo, this return value exhibits the return_code, stderr, failed and succeeded attributes. See run for details.
local will honor the lcd context manager, allowing you to control its current working directory independently of the remote end (which honors cd).
Changed in version 1.0: Added the succeeded and stderr attributes.
Changed in version 1.0: Now honors the lcd context manager.
Changed in version 1.0: Changed the default value of capture from True to False.
Prompt user with text and return the input (like raw_input).
A single space character will be appended for convenience, but nothing else. Thus, you may want to end your prompt text with a question mark or a colon, e.g. prompt("What hostname?").
If key is given, the user’s input will be stored as env.<key> in addition to being returned by prompt. If the key already existed in env, its value will be overwritten and a warning printed to the user.
If default is given, it is displayed in square brackets and used if the user enters nothing (i.e. presses Enter without entering any text). default defaults to the empty string. If non-empty, a space will be appended, so that a call such as prompt("What hostname?", default="foo") would result in a prompt of What hostname? [foo] (with a trailing space after the [foo].)
The optional keyword argument validate may be a callable or a string:
Either way, prompt will re-prompt until validation passes (or the user hits Ctrl-C).
Note
prompt honors env.abort_on_prompts and will call abort instead of prompting if that flag is set to True. If you want to block on user input regardless, try wrapping with settings.
Examples:
# Simplest form:
environment = prompt('Please specify target environment: ')
# With default, and storing as env.dish:
prompt('Specify favorite dish: ', 'dish', default='spam & eggs')
# With validation, i.e. requiring integer input:
prompt('Please specify process nice level: ', key='nice', validate=int)
# With validation against a regular expression:
release = prompt('Please supply a release name',
validate=r'^\w+-\d+(\.\d+)?$')
# Prompt regardless of the global abort-on-prompts setting:
with settings(abort_on_prompts=False):
prompt('I seriously need an answer on this! ')
Check for given keys in the shared environment dict and abort if not found.
Positional arguments should be strings signifying what env vars should be checked for. If any of the given arguments do not exist, Fabric will abort execution and print the names of the missing keys.
The optional keyword argument used_for may be a string, which will be printed in the error output to inform users why this requirement is in place. used_for is printed as part of a string similar to:
"Th(is|ese) variable(s) (are|is) used for %s"
so format it appropriately.
The optional keyword argument provided_by may be a list of functions or function names or a single function or function name which the user should be able to execute in order to set the key or keys; it will be included in the error output if requirements are not met.
Note: it is assumed that the keyword arguments apply to all given keys as a group. If you feel the need to specify more than one used_for, for example, you should break your logic into multiple calls to require().
Changed in version 1.1: Allow iterable provided_by values instead of just single values.
Abstract base class for objects wishing to be picked up as Fabric tasks.
Instances of subclasses will be treated as valid tasks when present in fabfiles loaded by the fab tool.
For details on how to implement and use Task subclasses, please see the usage documentation on new-style tasks.
New in version 1.1.
Return the host list the given task should be using.
See How host lists are constructed for detailed documentation on how host lists are set.
Execute task (callable or name), honoring host/role decorators, etc.
task may be an actual callable object, or it may be a registered task name, which is used to look up a callable just as if the name had been given on the command line (including namespaced tasks, e.g. "deploy.migrate".
The task will then be executed once per host in its host list, which is (again) assembled in the same manner as CLI-specified tasks: drawing from -H, env.hosts, the hosts or roles decorators, and so forth.
host, hosts, role, roles and exclude_hosts kwargs will be stripped out of the final call, and used to set the task’s host list, as if they had been specified on the command line like e.g. fab taskname:host=hostname.
Any other arguments or keyword arguments will be passed verbatim into task when it is called, so execute(mytask, 'arg1', kwarg1='value') will (once per host) invoke mytask('arg1', kwarg1='value').
See also
The execute usage docs, for an expanded explanation and some examples.
New in version 1.3.
Internal subroutines for e.g. aborting execution with an error message, or performing indenting on multiline output.
Abort execution, print msg to stderr and exit with error status (1.)
This function currently makes use of sys.exit, which raises SystemExit. Therefore, it’s possible to detect and recover from inner calls to abort by using except SystemExit or similar.
Print text immediately, without any prefix or line ending.
This function is simply an alias of puts with different default argument values, such that the text is printed without any embellishment and immediately flushed.
It is useful for any situation where you wish to print text which might otherwise get buffered by Python’s output buffering (such as within a processor intensive for loop). Since such use cases typically also require a lack of line endings (such as printing a series of dots to signify progress) it also omits the traditional newline by default.
Note
Since fastprint calls puts, it is likewise subject to the user output level.
New in version 0.9.2.
See also
Return text indented by the given number of spaces.
If text is not a string, it is assumed to be a list of lines and will be joined by \n prior to indenting.
When strip is True, a minimum amount of whitespace is removed from the left-hand side of the given string (so that relative indents are preserved, but otherwise things are left-stripped). This allows you to effectively “normalize” any previous indentation for some inputs.
An alias for print whose output is managed by Fabric’s output controls.
In other words, this function simply prints to sys.stdout, but will hide its output if the user output level is set to False.
If show_prefix=False, puts will omit the leading [hostname] which it tacks on by default. (It will also omit this prefix if env.host_string is empty.)
Newlines may be disabled by setting end to the empty string (''). (This intentionally mirrors Python 3’s print syntax.)
You may force output flushing (e.g. to bypass output buffering) by setting flush=True.
New in version 0.9.2.
See also
Print warning message, but do not abort execution.
This function honors Fabric’s output controls and will print the given msg to stderr, provided that the warnings output level (which is active by default) is turned on.
Fabric’s contrib package contains commonly useful tools (often merged in from user fabfiles) for tasks such as user I/O, modifying remote files, and so forth. While the core API is likely to remain small and relatively unchanged over time, this contrib section will grow and evolve (while trying to remain backwards-compatible) as more use-cases are solved and added.
Console/terminal user interface functionality.
Ask user a yes/no question and return their response as True or False.
question should be a simple, grammatically complete question such as “Do you wish to continue?”, and will have a string similar to ” [Y/n] ” appended automatically. This function will not append a question mark for you.
By default, when the user presses Enter without typing anything, “yes” is assumed. This can be changed by specifying default=False.
New in version 0.9.2.
These functions streamline the process of initializing Django’s settings module environment variable. Once this is done, your fabfile may import from your Django project, or Django itself, without requiring the use of manage.py plugins or having to set the environment variable yourself every time you use your fabfile.
Currently, these functions only allow Fabric to interact with local-to-your-fabfile Django installations. This is not as limiting as it sounds; for example, you can use Fabric as a remote “build” tool as well as using it locally. Imagine the following fabfile:
from fabric.api import run, local, hosts, cd
from fabric.contrib import django
django.project('myproject')
from myproject.myapp.models import MyModel
def print_instances():
for instance in MyModel.objects.all():
print(instance)
@hosts('production-server')
def print_production_instances():
with cd('/path/to/myproject'):
run('fab print_instances')
With Fabric installed on both ends, you could execute print_production_instances locally, which would trigger print_instances on the production server – which would then be interacting with your production Django database.
As another example, if your local and remote settings are similar, you can use it to obtain e.g. your database settings, and then use those when executing a remote (non-Fabric) command. This would allow you some degree of freedom even if Fabric is only installed locally:
from fabric.api import run
from fabric.contrib import django
django.settings_module('myproject.settings')
from django.conf import settings
def dump_production_database():
run('mysqldump -u %s -p=%s %s > /tmp/prod-db.sql' % (
settings.DATABASE_USER,
settings.DATABASE_PASSWORD,
settings.DATABASE_NAME
))
The above snippet will work if run from a local, development environment, again provided your local settings.py mirrors your remote one in terms of database connection info.
Sets DJANGO_SETTINGS_MODULE to '<name>.settings'.
This function provides a handy shortcut for the common case where one is using the Django default naming convention for their settings file and location.
Uses settings_module – see its documentation for details on why and how to use this functionality.
Set DJANGO_SETTINGS_MODULE shell environment variable to module.
Due to how Django works, imports from Django or a Django project will fail unless the shell environment variable DJANGO_SETTINGS_MODULE is correctly set (see the Django settings docs.)
This function provides a shortcut for doing so; call it near the top of your fabfile or Fabric-using code, after which point any Django imports should work correctly.
Note
This function sets a shell environment variable (via os.environ) and is unrelated to Fabric’s own internal “env” variables.
Module providing easy API for working with remote files and folders.
Append string (or list of strings) text to filename.
When a list is given, each string inside is handled independently (but in the order given.)
If text is already found in filename, the append is not run, and None is returned immediately. Otherwise, the given text is appended to the end of the given filename via e.g. echo '$text' >> $filename.
The test for whether text already exists defaults to a full line match, e.g. ^<text>$, as this seems to be the most sensible approach for the “append lines to a file” use case. You may override this and force partial searching (e.g. ^<text>) by specifying partial=True.
Because text is single-quoted, single quotes will be transparently backslash-escaped. This can be disabled with escape=False.
If use_sudo is True, will use sudo instead of run.
Changed in version 0.9.1: Added the partial keyword argument.
Changed in version 1.0: Swapped the order of the filename and text arguments to be consistent with other functions in this module.
Changed in version 1.0: Changed default value of partial kwarg to be False.
Attempt to comment out all lines in filename matching regex.
The default commenting character is # and may be overridden by the char argument.
This function uses the sed function, and will accept the same use_sudo and backup keyword arguments that sed does.
comment will prepend the comment character to the beginning of the line, so that lines end up looking like so:
this line is uncommented
#this line is commented
# this line is indented and commented
In other words, comment characters will not “follow” indentation as they sometimes do when inserted by hand. Neither will they have a trailing space unless you specify e.g. char='# '.
Note
In order to preserve the line being commented out, this function will wrap your regex argument in parentheses, so you don’t need to. It will ensure that any preceding/trailing ^ or $ characters are correctly moved outside the parentheses. For example, calling comment(filename, r'^foo$') will result in a sed call with the “before” regex of r'^(foo)$' (and the “after” regex, naturally, of r'#\1'.)
Return True if filename contains text.
By default, this function will consider a partial line match (i.e. where the given text only makes up part of the line it’s on). Specify exact=True to change this behavior so that only a line containing exactly text results in a True return value.
Double-quotes in either text or filename will be automatically backslash-escaped in order to behave correctly during the remote shell invocation.
If use_sudo is True, will use sudo instead of run.
Changed in version 1.0: Swapped the order of the filename and text arguments to be consistent with other functions in this module.
Return True if given path exists on the current remote host.
If use_sudo is True, will use sudo instead of run.
exists will, by default, hide all output (including the run line, stdout, stderr and any warning resulting from the file not existing) in order to avoid cluttering output. You may specify verbose=True to change this behavior.
Given one or more file paths, returns first one found, or None if none exist. May specify use_sudo and verbose which are passed to exists.
Run a search-and-replace on filename with given regex patterns.
Equivalent to sed -i<backup> -r -e "/<limit>/ s/<before>/<after>/<flags>g <filename>".
For convenience, before and after will automatically escape forward slashes, single quotes and parentheses for you, so you don’t need to specify e.g. http:\/\/foo\.com, instead just using http://foo\.com is fine.
If use_sudo is True, will use sudo instead of run.
sed will pass shell=False to run/sudo, in order to avoid problems with many nested levels of quotes and backslashes.
Other options may be specified with sed-compatible regex flags – for example, to make the search and replace case insensitive, specify flags="i". The g flag is always specified regardless, so you do not need to remember to include it when overriding this parameter.
New in version 1.1: The flags parameter.
Attempt to uncomment all lines in filename matching regex.
The default comment delimiter is # and may be overridden by the char argument.
This function uses the sed function, and will accept the same use_sudo and backup keyword arguments that sed does.
uncomment will remove a single whitespace character following the comment character, if it exists, but will preserve all preceding whitespace. For example, # foo would become foo (the single space is stripped) but `` # foo`` would become `` foo`` (the single space is still stripped, but the preceding 4 spaces are not.)
Render and upload a template text file to a remote host.
filename should be the path to a text file, which may contain Python string interpolation formatting and will be rendered with the given context dictionary context (if given.)
Alternately, if use_jinja is set to True and you have the Jinja2 templating library available, Jinja will be used to render the template instead. Templates will be loaded from the invoking user’s current working directory by default, or from template_dir if given.
The resulting rendered file will be uploaded to the remote file path destination. If the destination file already exists, it will be renamed with a .bak extension unless backup=False is specified.
By default, the file will be copied to destination as the logged-in user; specify use_sudo=True to use sudo instead.
The mirror_local_mode and mode kwargs are passed directly to an internal put call; please see its documentation for details on these two options.
Changed in version 1.1: Added the backup, mirror_local_mode and mode kwargs.
Useful non-core functionality, e.g. functions composing multiple operations.
Synchronize a remote directory with the current project directory via rsync.
Where upload_project() makes use of scp to copy one’s entire project every time it is invoked, rsync_project() uses the rsync command-line utility, which only transfers files newer than those on the remote end.
rsync_project() is thus a simple wrapper around rsync; for details on how rsync works, please see its manpage. rsync must be installed on both your local and remote systems in order for this operation to work correctly.
This function makes use of Fabric’s local() operation, and returns the output of that function call; thus it will return the stdout, if any, of the resultant rsync call.
rsync_project() takes the following parameters:
remote_dir: the only required parameter, this is the path to the directory on the remote server. Due to how rsync is implemented, the exact behavior depends on the value of local_dir:
- If local_dir ends with a trailing slash, the files will be dropped inside of remote_dir. E.g. rsync_project("/home/username/project", "foldername/") will drop the contents of foldername inside of /home/username/project.
- If local_dir does not end with a trailing slash (and this includes the default scenario, when local_dir is not specified), remote_dir is effectively the “parent” directory, and a new directory named after local_dir will be created inside of it. So rsync_project("/home/username", "foldername") would create a new directory /home/username/foldername (if needed) and place the files there.
local_dir: by default, rsync_project uses your current working directory as the source directory. This may be overridden by specifying local_dir, which is a string passed verbatim to rsync, and thus may be a single directory ("my_directory") or multiple directories ("dir1 dir2"). See the rsync documentation for details.
exclude: optional, may be a single string, or an iterable of strings, and is used to pass one or more --exclude options to rsync.
delete: a boolean controlling whether rsync‘s --delete option is used. If True, instructs rsync to remove remote files that no longer exist locally. Defaults to False.
extra_opts: an optional, arbitrary string which you may use to pass custom arguments or options to rsync.
Furthermore, this function transparently honors Fabric’s port and SSH key settings. Calling this function when the current host string contains a nonstandard port, or when env.key_filename is non-empty, will use the specified port and/or SSH key filename(s).
For reference, the approximate rsync command-line call that is constructed by this function is the following:
rsync [--delete] [--exclude exclude[0][, --exclude[1][, ...]]] \
-pthrvz [extra_opts] <local_dir> <host_string>:<remote_dir>
Upload the current project to a remote system via tar/gzip.
local_dir specifies the local project directory to upload, and defaults to the current working directory.
remote_dir specifies the target directory to upload into (meaning that a copy of local_dir will appear as a subdirectory of remote_dir) and defaults to the remote user’s home directory.
This function makes use of the tar and gzip programs/libraries, thus it will not work too well on Win32 systems unless one is using Cygwin or something similar. It will attempt to clean up the local and remote tarfiles when it finishes executing, even in the event of a failure.
Changed in version 1.1: Added the local_dir and remote_dir kwargs.
Please see the changelog.
Please see the roadmap.
If you’ve scoured the prose and API documentation and still can’t find an answer to your question, below are various support resources that should help. We do request that you do at least skim the documentation before posting tickets or mailing list questions, however!
The best way to get help with using Fabric is via the fab-user mailing list (currently hosted at nongnu.org.) The Fabric developers do their best to reply promptly, and the list contains an active community of other Fabric users and contributors as well.
Fabric has an official Twitter account, @pyfabric, which is used for announcements and occasional related news tidbits (e.g. “Hey, check out this neat article on Fabric!”).
To file new bugs or search existing ones, you may visit Fabric’s Github Issues page. This does require a (free, easy to set up) Github account.
We maintain a semi-official IRC channel at #fabric on Freenode (irc://irc.freenode.net) where the developers and other users may be found. As always with IRC, we can’t promise immediate responses, but some folks keep logs of the channel and will try to get back to you when they can.
There is an official Fabric MoinMoin wiki reachable at wiki.fabfile.org, although as of this writing its usage patterns are still being worked out. Like the ticket tracker, spam has forced us to put anti-spam measures up: the wiki has a simple, easy captcha in place on the edit form.