def on_create_request(self, rsrc_id, rsrc_data):
with lc.LogContext(_LOGGER, rsrc_id,
adapter_cls=lc.ContainerAdapter) as log:
log.info('Creating presence: %s', rsrc_data)
app_name = appcfg.app_name(rsrc_id)
# Register running.
path = z.path.running(app_name)
_LOGGER.info('Register running: %s, %s', path, self.hostname)
if not self._safe_create(rsrc_id, path, self.hostname):
_LOGGER.info('Waiting to expire: %s', path)
return None
self.presence[app_name][path] = rsrc_id
# Register endpoints.
for endpoint in rsrc_data.get('endpoints', []):
internal_port = endpoint['port']
ep_name = endpoint.get('name', str(internal_port))
ep_port = endpoint['real_port']
ep_proto = endpoint.get('proto', 'tcp')
hostport = self.hostname + ':' + str(ep_port)
path = z.path.endpoint(app_name, ep_proto, ep_name)
_LOGGER.info('Register endpoint: %s, %s', path, hostport)
if not self._safe_create(rsrc_id, path, hostport):
_LOGGER.info('Waiting to expire: %s', path)
return None
self.presence[app_name][path] = rsrc_id
# Register identity.
identity_group = rsrc_data.get('identity_group')
if identity_group:
identity = rsrc_data.get('identity', _INVALID_IDENTITY)
identity_data = {'host': self.hostname, 'app': app_name}
path = z.path.identity_group(identity_group, str(identity))
_LOGGER.info('Register identity: %s, %s', path, identity_data)
if not self._safe_create(rsrc_id, path, identity_data):
_LOGGER.info('Waiting to expire: %s', path)
return None
self.presence[app_name][path] = rsrc_id
return {}
def load_app_safe(container, container_dir, app_json=STATE_JSON):
"""Load app manifest as object.
If app manifest is corrupted or invalid, return object with key attributes.
"""
try:
return load_app(container_dir, app_json=app_json)
except ValueError as err:
_LOGGER.error('Manifest file is corrupted or invalid: %s', err)
appname = appcfg.app_name(container)
return utils.to_obj({
'name': appname,
'app': appcfg.appname_basename(appname),
'task': appcfg.appname_task_id(appname),
'uniqueid': appcfg.app_unique_id(container),
})
def on_delete_request(self, rsrc_id):
with lc.LogContext(_LOGGER, rsrc_id,
adapter_cls=lc.ContainerAdapter) as log:
log.info('Deleting presence')
app_name = appcfg.app_name(rsrc_id)
to_delete = [
path for path in self.presence[app_name]
if self.presence[app_name][path] == rsrc_id
]
for path in to_delete:
self._safe_delete(path)
del self.presence[app_name][path]
if not self.presence[app_name]:
del self.presence[app_name]
return True
def finish(tm_env, container_dir):
"""Frees allocated resources and mark then as available.
"""
container = os.path.basename(container_dir)
with lc.LogContext(_LOGGER, container, lc.ContainerAdapter):
_LOGGER.info('finishing %r', container_dir)
data_dir = os.path.join(container_dir, 'data')
appname = appcfg.app_name(container)
app = runtime.load_app_safe(container, data_dir)
if app:
_cleanup(tm_env, data_dir, app)
apphook.cleanup(tm_env, app, container_dir)
# All resources are cleaned up. If the app terminated inside the
# container, remove the node from Zookeeper, which will notify the
# scheduler that it is safe to reuse the host for other load.
exitinfo, aborted, oom, terminated = _collect_finish_info(data_dir)
if aborted is not None:
_post_aborted_event(tm_env, appname, aborted)
elif oom:
_post_oom_event(tm_env, appname)
elif terminated:
# Terminated (or evicted).
# Don't post event, this is handled by the scheduler.
_LOGGER.info('Terminated: %s', appname)
else:
# Container finished because service exited.
# It is important that this is checked last.
if exitinfo is not None:
_post_exit_event(tm_env, appname, exitinfo)
def _synchronize(self):
"""Synchronize apps to running/cleanup.
We need to re-validate three things on startup:
- All configured apps should have an associated cache entry.
Otherwise, create a link to cleanup.
- All configured apps with a cache entry and with a cleanup file
should be linked to cleanup. Otherwise, link to running.
- Additional cache entries should be configured to run.
On restart we need to validate another three things:
- All configured apps that have a running link should be checked
if in the cache. If not then terminate the app.
- All configured apps that have a cleanup link should be left
alone as this is handled.
- Additional cache entries should be configured to run.
On startup run.sh will clear running and cleanup which simplifies
the logic for us as we can check running/cleanup first. Then check
startup conditions and finally non-configured apps that are in cache.
NOTE: a link cannot exist in running and cleanup at the same time.
"""
# Disable R0912(too-many-branches)
# pylint: disable=R0912
configured = {
os.path.basename(filename)
for filename in glob.glob(os.path.join(self.tm_env.apps_dir, '*'))
}
cached = {
os.path.basename(filename): appcfg.eventfile_unique_name(filename)
for filename in glob.glob(os.path.join(self.tm_env.cache_dir, '*'))
}
for container in configured:
appname = appcfg.app_name(container)
if os.path.exists(os.path.join(self.tm_env.running_dir, appname)):
# App already running.. check if in cache.
# No need to check if needs cleanup as that is handled
if appname not in cached or cached[appname] != container:
self._terminate(appname)
else:
_LOGGER.info('Ignoring %s as it is running', appname)
cached.pop(appname, None)
elif os.path.exists(os.path.join(self.tm_env.cleanup_dir,
appname)):
# Already in the process of being cleaned up
_LOGGER.info('Ignoring %s as it is in cleanup', appname)
cached.pop(appname, None)
else:
needs_cleanup = True
if appname in cached and cached[appname] == container:
data_dir = os.path.join(self.tm_env.apps_dir, container,
'data')
for cleanup_file in ['exitinfo', 'aborted', 'oom']:
path = os.path.join(data_dir, cleanup_file)
if os.path.exists(path):
_LOGGER.debug('Found cleanup file %r', path)
break
else:
if self._configure(appname):
needs_cleanup = False
_LOGGER.debug('Added existing app %r', appname)
cached.pop(appname, None)
if needs_cleanup:
fs.symlink_safe(
os.path.join(self.tm_env.cleanup_dir, appname),
os.path.join(self.tm_env.apps_dir, container))
_LOGGER.debug('Removed %r', appname)
for appname in six.iterkeys(cached):
if self._configure(appname):
_LOGGER.debug('Added new app %r', appname)
self._refresh_supervisor()