path: root/scripts/killvm

#!/usr/bin/python -tt
# skvidal
# take builder/instance name
# look for it on buildvmhost boxes
# if it is there/up
# confirm to kill it
# destroy it
# lvremove the disk
# undefine it
# 

import os
import sys
import socket
from socket import gaierror
import ansible
import ansible.runner
import ansible.playbook
import time
from ansible import callbacks
from pprint import pprint
import optparse

def get_ans_results(results, hostname):
    if hostname in results['dark']:
        return results['dark'][hostname]
    if hostname in results['contacted']:
        return results['contacted'][hostname]
    
    return {}

def confirm():
    ans = raw_input()
    if ans.lower() == 'yes':
        return True
    return False



def find_instance(conn, instance):
    vmdict  = get_vm_to_host_map(conn)

    if instance in vmdict:
        return vmdict[instance]
    return None
    
def get_vm_to_host_map(conn):

    conn.module_name='virt'
    conn.module_args='command=list_vms'

    res = conn.run()

    vm_to_host = {}

    for (host,data) in sorted(res['contacted'].items()):
        for vm in data['list_vms']:
            vm_to_host[vm] = host

    return vm_to_host
    
def check_for_ans_error(results, hostname, err_codes=[], success_codes=[0], 
                             return_on_error=['stdout', 'stderr']):
    # returns True or False + dict
    # dict includes 'msg'
    # may include 'rc', 'stderr', 'stdout' and any other
    # requested result codes
    err_results = {}
    
    if 'dark' in results and hostname in results['dark']:
        err_results['msg'] = "Error: Could not contact/connect to %s." % hostname
        return (True, err_results)
    
    error = False

    if err_codes or success_codes:
        if hostname in results['contacted']:
            if 'rc' in results['contacted'][hostname]:
                rc = int(results['contacted'][hostname]['rc'])
                err_results['rc'] = rc
                # check for err codes first
                if rc in err_codes:
                    error = True
                    err_results['msg'] = 'rc %s matched err_codes' % rc
                elif rc not in success_codes:
                    error = True
                    err_results['msg'] = 'rc %s not in success_codes' % rc
            elif 'failed' in results['contacted'][hostname] and results['contacted'][hostname]['failed']:
                error = True
                err_results['msg'] = 'results included failed as true'
                 
        if error:
            for item in return_on_error:
                if item in results['contacted'][hostname]:
                    err_results[item] = results['contacted'][hostname][item]
                    
    return error, err_results

def vm_is_defined(conn, vm, vmhost):
    # get list of vms
    conn.module_name = 'virt'
    conn.module_args = 'command=list_vms'
    results = get_ans_results(conn.run(), vmhost)
    # if vm is in in there
    if vm in results.get('list_vms', []):
        return True
    return False
    

def vm_is_alive(conn, vm, vmhost):
    if not vm_is_defined(conn, vm, vmhost):
        return False
    conn.module_name = 'virt'
    conn.module_args = 'command=status guest=%s' % vm
    results = get_ans_results(conn.run(), vmhost)
    if results.get('status', None)  == 'running':
        return True

    return False

def wait_for_host(hn, timeout=300):
    # watch for that host ssh to come up
    conn = ansible.runner.Runner(host_list=hn +',', pattern=hn, remote_user='root')
    is_up = False
    start = time.time()
    while not is_up:
        if time.time() - start >= timeout:
            raise Exception, "Hit Timeout waiting for %s to boot" % hn
        conn.module_name='ping'
        res = get_ans_results(conn.run(), hn)
        if res.get('ping'):
            is_up=True
        else:
            time.sleep(2)
        



def parse_args(args):
    parser = optparse.OptionParser('\nkillvm [options] vm')
    parser.add_option('-i', '--inventory', default='/srv/web/infra/ansible/inventory', 
                   help="path to ansible inventory file")
    parser.add_option('-p', '--pattern', default='buildvmhost:bvirthost:virthost:colo-virt', 
                   help="ansible host pattern to use to look up vmhosts")                   
    parser.add_option('-y', '--yes', default=False, dest='yes', action="store_true",
                    help='Do not confirm any of the destructive actions - just do them')
    parser.add_option('--vg', default='/dev/vg_host01', dest='vg',
                    help='path to volumegroup to use on vmhost for vm disk: %default')                    

    opts, args = parser.parse_args(args)

    if not os.path.exists(opts.inventory):
        print "Could not find ansible inventory at: %s" % opts.inventory
        sys.exit(1)
        
    if len(args) != 1:
        parser.print_usage()
        sys.exit(1)


    return opts, args



def main():

    opts, args = parse_args(sys.argv[1:])
    # args
    vm = args[0]

    try:
        ip = socket.gethostbyname(vm)
    except gaierror,e:
        print 'Could not find ip for %s' % vm
        return 1

    if vm.find('.') == -1:
        print '%s was not a fqdn, cmon!' % vm
        return 1
           
    s_vm = vm.split('.')[0]

    print 'Checking for %s' % vm

    conn = ansible.runner.Runner(host_list=opts.inventory, pattern=opts.pattern, timeout=20, forks=30, remote_user='root')
    vmhost = find_instance(conn, instance=vm)
    if not vmhost:
        print 'Could not find vm %s on any virthost in %s' % (vm, opts.pattern)
        sys.exit(1)

    print 'Found on %s' % vmhost    

    vmhost_conn = ansible.runner.Runner(host_list=vmhost+',', pattern=vmhost, remote_user='root')
    if vm_is_defined(vmhost_conn, vm, vmhost):
        if vm_is_alive(vmhost_conn, vm, vmhost):
            if not opts.yes:
                print "%s is running. Okay to Destroy? ('yes' to confirm): " % vm,
                if not confirm():
                    print 'Exiting on user input'
                    return 1
                
            # destroy it
            vmhost_conn.module_args = "command=destroy guest=%s" % vm
            err, err_res = check_for_ans_error(vmhost_conn.run(), vmhost)
            if err:
                print 'Error destroying %s on %s' % (vm, vmhost)
                print err_res
                return 1
                
        # undefine it
        if not opts.yes:
            print "%s is defined. Okay to Undefine? ('yes' to confirm): " % vm,
            if not confirm():
                print 'Exiting on user input'
                return 1
        
        vmhost_conn.module_args = "command=undefine guest=%s" % vm
        err, err_res = check_for_ans_error(vmhost_conn.run(), vmhost)
        if err:
            print 'Error undefining %s on %s' % (vm, vmhost)
            print err_res
            return 1

    # check for the lv being allocated already
    lv_check = '/sbin/lvs %s/%s --noheadings' % (opts.vg, s_vm)
    vmhost_conn.module_name='command'
    vmhost_conn.module_args=lv_check
    results = get_ans_results(vmhost_conn.run(), vmhost)
    if 'rc' not in results:
        print 'Could not talk to vmhost about disks'
        return 1

    if results['rc'] == 0:
        print 'Removing old disk: %s/%s' % (opts.vg, s_vm)
        # lvremove its disk
        lvrm='/sbin/lvremove -f %s/%s' % (opts.vg, s_vm)
        vmhost_conn.module_name='command'   
        vmhost_conn.module_args=lvrm
        results = get_ans_results(vmhost_conn.run(), vmhost)
        if results.get('rc', None) != 0:
            print "Could not remove lv for old vm %s" % vm
            print results
            return 1


     

if __name__ == "__main__":
    sys.exit(main())