zuul
/
zuul
Code Issues Proposed changes

Ansible launcher: copy lib/plugins at startup 

Because we run Zuul continuously deployed from source and installed
via pip, whenever a Zuul commit lands, pip re-installs Zuul.  When
that happens, there may be several minutes where part of the Zuul
source code is no longer on disk.  Because the ansible launcher
points directly into the installed Zuul source for the zuul library
and plugins, jobs launched during that window will fail.  To correct
this, make a copy of the necessary ansible files on startup and point
the jobs to that location.

Change-Id: Ib0b6f07967ba314d34fad4835c4e2e84584e446c
This commit is contained in:
James E. Blair 2016-06-21 08:09:16 -07:00
parent 749ca63eb4
commit b547ae9542
1 changed files with 26 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
zuul/launcher/ansiblelaunchserver.py
View File

@ -163,6 +163,24 @@ class LaunchServer(object):
state_dir = '/var/lib/zuul'
path = os.path.join(state_dir, 'launcher.socket')
self.command_socket = commandsocket.CommandSocket(path)
ansible_dir = os.path.join(state_dir, 'ansible')
plugins_dir = os.path.join(ansible_dir, 'plugins')
self.callback_dir = os.path.join(plugins_dir, 'callback_plugins')
if not os.path.exists(self.callback_dir):
os.makedirs(self.callback_dir)
self.library_dir = os.path.join(ansible_dir, 'library')
if not os.path.exists(self.library_dir):
os.makedirs(self.library_dir)
callback_path = os.path.dirname(os.path.abspath(
zuul.ansible.plugins.callback_plugins.__file__))
for fn in os.listdir(callback_path):
shutil.copy(os.path.join(callback_path, fn), self.callback_dir)
library_path = os.path.dirname(os.path.abspath(
zuul.ansible.library.__file__))
for fn in os.listdir(library_path):
shutil.copy(os.path.join(library_path, fn), self.library_dir)
for section in config.sections():
m = self.site_section_re.match(section)
@ -435,7 +453,8 @@ class LaunchServer(object):
self.sites, args['name'], args['host'],
args['description'], args['labels'],
self.hostname, self.zmq_send_queue,
self.termination_queue, self.keep_jobdir)
self.termination_queue, self.keep_jobdir,
self.callback_dir, self.library_dir)
self.node_workers[worker.name] = worker
worker.thread = threading.Thread(target=worker.run)
@ -489,7 +508,8 @@ class LaunchServer(object):
class NodeWorker(object):
def __init__(self, config, jobs, builds, sites, name, host,
description, labels, manager_name, zmq_send_queue,
termination_queue, keep_jobdir):
termination_queue, keep_jobdir, callback_dir,
library_dir):
self.log = logging.getLogger("zuul.NodeWorker.%s" % (name,))
self.log.debug("Creating node worker %s" % (name,))
self.config = config
@ -534,6 +554,8 @@ class NodeWorker(object):
self.username = config.get('launcher', 'username')
else:
self.username = 'zuul'
self.callback_dir = callback_dir
self.library_dir = library_dir
def isAlive(self):
# Meant to be called from the manager
@ -1175,16 +1197,8 @@ class NodeWorker(object):
config.write('log_path = %s\n' % os.path.join(
jobdir.logs, 'ansible.txt'))
config.write('gathering = explicit\n')
callback_path = zuul.ansible.plugins.callback_plugins.__file__
callback_path = os.path.abspath(callback_path)
callback_path = os.path.dirname(callback_path)
config.write('callback_plugins = %s\n' % callback_path)
library_path = zuul.ansible.library.__file__
library_path = os.path.abspath(library_path)
library_path = os.path.dirname(library_path)
config.write('library = %s\n' % library_path)
config.write('callback_plugins = %s\n' % self.callback_dir)
config.write('library = %s\n' % self.library_dir)
return timeout