Does python fabric support dynamic set env.hosts?

Question

I want to change the env.hosts dynamically because sometimes I want to deploy to one machine first, check if ok then deploy to many machines. Currently I need to set env.hosts first, how could I set the env.hosts in a method and not in global at script start?

Answer 1

Yes you can set env.hosts dynamically. One common pattern we use is:

from fabric.api import env

def staging():
    env.hosts = ['XXX.XXX.XXX.XXX', ]

def production():
    env.hosts = ['YYY.YYY.YYY.YYY', 'ZZZ.ZZZ.ZZZ.ZZZ', ]

def deploy():
   # Do something...

You would use this to chain the tasks such as fab staging deploy or fab production deploy.

Answer 2

Kind of late to the party, but I achieved this with ec2 like so (note in EC2 you do not know what the ip/hostname may be, generally speaking - so you almost have to go dynamic to really account for how the environment/systems could come up - another option would be to use dyndns, but then this would still be useful):

from fabric.api import *
import datetime
import time
import urllib2
import ConfigParser
from platform_util import *

config = ConfigParser.RawConfigParser()

@task
def load_config(configfile=None):
    '''
    ***REQUIRED***  Pass in the configuration to use - usage load_config:</path/to/config.cfg>
    '''
    if configfile != None:
        # Load up our config file
        config.read(configfile)

        # Key/secret needed for aws interaction with boto 
        # (anyone help figure out a better way to do this with sub modules, please don't say classes :-) )
        global aws_key
        global aws_sec

        aws_key = config.get("main","aws_key")
        aws_sec = config.get("main","aws_sec")

         # Stuff for fabric
        env.user                = config.get("main","fabric_ssh_user")
        env.key_filename        = config.get("main","fabric_ssh_key_filename")
        env.parallel            = config.get("main","fabric_default_parallel")

        # Load our role definitions for fabric
        for i in config.sections():
            if i != "main":
                hostlist = []
                if config.get(i,"use-regex") == 'yes':
                    for x in get_running_instances_by_regex(aws_key,aws_sec,config.get(i,"security-group"),config.get(i,"pattern")):
                        hostlist.append(x.private_ip_address)   
                    env.roledefs[i] = hostlist

                else:
                    for x in get_running_instances(aws_key,aws_sec,config.get(i,"security-group")):
                        hostlist.append(x.private_ip_address)   
                    env.roledefs[i] = hostlist

                if config.has_option(i,"base-group"):
                    if config.get(i,"base-group") == 'yes':
                        print "%s is a base group" % i
                        print env.roledefs[i]
#                       env["basegroups"][i] = True

where get_running_instances and get_running_instances_by_regex are utility functions that make use of boto (http://code.google.com/p/boto/)

ex:

import logging
import re
from boto.ec2.connection import EC2Connection
from boto.ec2.securitygroup import SecurityGroup
from boto.ec2.instance import Instance
from boto.s3.key import Key

########################################
# B-O-F get_instances
########################################
def get_instances(access_key=None, secret_key=None, security_group=None):
    '''
    Get all instances. Only within a security group if specified., doesnt' matter their state (running/stopped/etc)
    '''
    logging.debug('get_instances()')
    conn = EC2Connection(aws_access_key_id=access_key, aws_secret_access_key=secret_key)

    if security_group:
        sg = SecurityGroup(connection=conn, name=security_group)
        instances = sg.instances()
        return instances
    else:
        instances = conn.get_all_instances()
        return instances

Here is a sample of what my config looked like:

# Config file for fabric toolset
#
#  This specific configuration is for <whatever> related hosts
#  
#
[main]
aws_key = <key>
aws_sec = <secret>
fabric_ssh_user = <your_user>
fabric_ssh_key_filename = /path/to/your/.ssh/<whatever>.pem
fabric_default_parallel = 1

#
# Groupings - Fabric knows them as roledefs (check env dict)
#

# Production groupings
[app-prod]
security-group = app-prod
use-regex = no
pattern = 

[db-prod]
security-group = db-prod
use-regex = no
pattern = 

[db-prod-masters]
security-group = db-prod
use-regex = yes
pattern = mysql-[d-s]01

Answer 3

Yet another new answer to an old question. :) But I just recently found myself attempting to dynamically set hosts, and really have to disagree with the main answer. My idea of dynamic, or at least what I was attempting to do, was take an instance DNS-name that was just created by boto, and access that instance with a fab command. I couldn't do fab staging deploy, because the instance doesn't exist at fabfile-editing time.

Fortunately, fabric does support a truly dynamic host-assignment with execute. (It's possible this didn't exist when the question was first asked, of course, but now it does). Execute allows you to define both a function to be called, and the env.hosts it should use for that command. For example:

def create_EC2_box(data=fab_base_data):
    conn = boto.ec2.connect_to_region(region)
    reservations = conn.run_instances(image_id=image_id, ...)
    ...
    return instance.public_dns_name

def _ping_box():
    run('uname -a')
    run('tail /var/log/cloud-init-output.log')

def build_box():
    box_name = create_EC2_box(fab_base_data)
    new_hosts = [box_name]
    # new_hosts = ['ec2-54-152-152-123.compute-1.amazonaws.com'] # testing
    execute(_ping_box, hosts=new_hosts)

Now I can do fab build_box, and it will fire one boto call that creates an instance, and another fabric call that runs on the new instance - without having to define the instance-name at edit-time.