Fexpect: Dealing with prompts in fabric with pexpect
Tuesday 06 March 2012
Update: Fexpect has been improved and is now on pypi and github (new blogpost)
Fabric is a popular python library for executing commands on a remote computer over ssh. It is commonly used for deployment or provisioning tasks. Often however, these tasks ask for user feedback through a prompt (“Continue? Y/N”). Fabric has no built-in way to code how to deal with such a prompt, which means that execution of the script will pause until the user provides the answer to the prompt.
There are several other tools available to ‘expect’ such a prompt and answer with a scripted response. Among those are the unix tools ‘yes’ and ‘expect’ [1]. Others have suggested to use these for situations that require prompt interaction in fabric scripts. One common use case is for an OS-upgrade, which usually includes several prompts.
A more elegant solution would be an extension of fabric. I recently suggested such a plugin on the fabric mailing list [2], and will describe my solution in more details here. It involves using the python version of ‘expect’, pexpect [3].
In short, it entails compiling a short fexpect script, sending it along with the fexpect library to the target machine using put(), and executing it.
Fabric scripts have to formulate expectations, i.e. prompts, and the corresponding answer (‘y’):
from fexpect import expect
expectation = expect("Continue \[yN\]",'y')
Notice that we escape the brackets, since the expectation is formulated as a regular expression to match the prompt.
We can combine expectations in a list, since often a command prompts multiple questions:
prompts = []
prompts += expect(‘What is your name?’,'Jasper')
prompts += expect('Where do you live?','Frankfurt')
And then we run the command with a context manager for the expectations:
from fexpect import expecting, run #or import ‘as erun’ not to confuse with fabric.api.run with expecting(prompts): run(‘command’)
Here are some extracts from the code to see what’s going on:
The context manager just stores the expectations in the fabric state:
class ExpectationContext(object): def __init__(self,expectations): self.expectations = expectations def __enter__(self): env.expectations = self.expectations def __exit__(self, type, value, tb): env.expectations = []
A custom run function (sudo also available):
def run(cmd): #sudo wrapper wrappedCmd = wrapExpectations(cmd,env) return fabric.api.run(wrappedCmd)
Which calls the actual ‘magic’:
def wrapExpectations(cmd,env):
script = createScript(cmd,env)
remoteScript = '/tmp/fexpect_'+shortuuid.uuid()
fabric.api.put(resource('pexpect.py'),'/tmp/')
fabric.api.put(StringIO(script),remoteScript)
wrappedCmd = 'python '+remoteScript
return wrappedCmd
Which in turn calls a simple function which creates the pexpect script.
Get the complete module and some tests at github: fexpect
Links:
- [1] unix expect
- [2] original post to fab-user@nongnu.org
- [3] pexpect
Comments? Write me an email or discuss on facebook, google+ or twitter.