Merge pull request #1356 from minrk/proxy-race

rework spawn futures to fix races
This commit is contained in:
Carol Willing
2017-08-26 11:07:55 -04:00
committed by GitHub
9 changed files with 166 additions and 102 deletions

View File

@@ -104,22 +104,17 @@ class APIHandler(BaseHandler):
'pending': None,
'last_activity': user.last_activity.isoformat(),
}
if user.spawners['']._spawn_pending:
model['pending'] = 'spawn'
elif user.spawners['']._stop_pending:
model['pending'] = 'stop'
model['pending'] = user.spawners[''].pending or None
if self.allow_named_servers:
servers = model['servers'] = {}
for name, spawner in user.spawners.items():
if spawner.ready:
servers[name] = s = {'name': name}
if spawner._spawn_pending:
s['pending'] = 'spawn'
elif spawner._stop_pending:
s['pending'] = 'stop'
if spawner.pending:
s['pending'] = spawner.pending
if spawner.server:
s['url'] = user.url + name
s['url'] = user.url + name + '/'
return model
def group_model(self, group):

View File

@@ -178,19 +178,34 @@ class UserAPIHandler(APIHandler):
class UserServerAPIHandler(APIHandler):
"""Start and stop single-user servers"""
@gen.coroutine
@admin_or_self
def post(self, name, server_name=''):
user = self.find_user(name)
if server_name:
if not self.allow_named_servers:
raise web.HTTPError(400, "Named servers are not enabled.")
if server_name and not self.allow_named_servers:
raise web.HTTPError(400, "Named servers are not enabled.")
if self.allow_named_servers and not server_name:
server_name = user.default_server_name()
spawner = user.spawners[server_name]
pending = spawner.pending
if pending == 'spawn':
self.set_header('Content-Type', 'text/plain')
self.set_status(202)
return
elif pending:
raise web.HTTPError(400, "%s is pending %s" % (spawner._log_name, pending))
if spawner.ready:
# include notify, so that a server that died is noticed immediately
state = yield spawner.poll_and_notify()
# set _spawn_pending flag to prevent races while we wait
spawner._spawn_pending = True
try:
state = yield spawner.poll_and_notify()
finally:
spawner._spawn_pending = False
if state is None:
raise web.HTTPError(400, "%s's server %s is already running" % (name, server_name))
raise web.HTTPError(400, "%s is already running" % spawner._log_name)
options = self.get_json_body()
yield self.spawn_single_user(user, server_name, options=options)
@@ -209,17 +224,18 @@ class UserServerAPIHandler(APIHandler):
raise web.HTTPError(404, "%s has no server named '%s'" % (name, server_name))
spawner = user.spawners[server_name]
if spawner._stop_pending:
if spawner.pending == 'stop':
self.log.debug("%s already stopping", spawner._log_name)
self.set_header('Content-Type', 'text/plain')
self.set_status(202)
return
if not spawner.ready:
raise web.HTTPError(400, "%s's server %s is not running" % (name, server_name))
raise web.HTTPError(400, "%s is not running" % spawner._log_name)
# include notify, so that a server that died is noticed immediately
status = yield spawner.poll_and_notify()
if status is not None:
raise web.HTTPError(400, "%s's server %s is not running" % (name, server_name))
raise web.HTTPError(400, "%s is not running" % spawner._log_name)
yield self.stop_single_user(user, server_name)
status = 202 if spawner._stop_pending else 204
self.set_header('Content-Type', 'text/plain')

View File

@@ -20,7 +20,7 @@ from .. import __version__
from .. import orm
from ..objects import Server
from ..spawner import LocalProcessSpawner
from ..utils import url_path_join, exponential_backoff
from ..utils import default_server_name, url_path_join, exponential_backoff
# pattern for the authentication token header
auth_header_pat = re.compile(r'^(?:token|bearer)\s+([^\s]+)$', flags=re.IGNORECASE)
@@ -376,8 +376,16 @@ class BaseHandler(RequestHandler):
@gen.coroutine
def spawn_single_user(self, user, server_name='', options=None):
if server_name in user.spawners and user.spawners[server_name].pending == 'spawn':
raise RuntimeError("Spawn already pending for: %s" % user.name)
user_server_name = user.name
if self.allow_named_servers and not server_name:
server_name = default_server_name(user)
if server_name:
user_server_name = '%s:%s' % (user.name, server_name)
if server_name in user.spawners and user.spawners[server_name].pending:
pending = user.spawners[server_name].pending
raise RuntimeError("%s pending %s" % (user_server_name, pending))
# count active servers and pending spawns
# we could do careful bookkeeping to avoid
@@ -391,32 +399,26 @@ class BaseHandler(RequestHandler):
active_server_limit = self.active_server_limit
if concurrent_spawn_limit and spawn_pending_count >= concurrent_spawn_limit:
self.log.info(
'%s pending spawns, throttling',
spawn_pending_count,
)
raise web.HTTPError(
429,
"User startup rate limit exceeded. Try again in a few minutes.")
self.log.info(
'%s pending spawns, throttling',
spawn_pending_count,
)
raise web.HTTPError(
429,
"User startup rate limit exceeded. Try again in a few minutes.",
)
if active_server_limit and active_count >= active_server_limit:
self.log.info(
'%s servers active, no space available',
active_count,
)
raise web.HTTPError(
429,
"Active user limit exceeded. Try again in a few minutes.")
self.log.info(
'%s servers active, no space available',
active_count,
)
raise web.HTTPError(429, "Active user limit exceeded. Try again in a few minutes.")
tic = IOLoop.current().time()
user_server_name = user.name
if server_name:
user_server_name = '%s:%s' % (user.name, server_name)
else:
user_server_name = user.name
self.log.debug("Initiating spawn for %s", user_server_name)
f = user.spawn(server_name, options)
spawn_future = user.spawn(server_name, options)
self.log.debug("%i%s concurrent spawns",
spawn_pending_count,
@@ -426,22 +428,28 @@ class BaseHandler(RequestHandler):
'/%i' % active_server_limit if active_server_limit else '')
spawner = user.spawners[server_name]
# set spawn_pending now, so there's no gap where _spawn_pending is False
# while we are waiting for _proxy_pending to be set
spawner._spawn_pending = True
@gen.coroutine
def finish_user_spawn(f=None):
def finish_user_spawn():
"""Finish the user spawn by registering listeners and notifying the proxy.
If the spawner is slow to start, this is passed as an async callback,
otherwise it is called immediately.
"""
if f and f.exception() is not None:
# failed, don't add to the proxy
return
# wait for spawn Future
try:
yield spawn_future
except Exception:
spawner._spawn_pending = False
raise
toc = IOLoop.current().time()
self.log.info("User %s took %.3f seconds to start", user_server_name, toc-tic)
self.statsd.timing('spawner.success', (toc - tic) * 1000)
spawner._proxy_pending = True
try:
spawner._proxy_pending = True
yield self.proxy.add_user(user, server_name)
except Exception:
self.log.exception("Failed to add %s to proxy!", user_server_name)
@@ -451,37 +459,41 @@ class BaseHandler(RequestHandler):
spawner.add_poll_callback(self.user_stopped, user, server_name)
finally:
spawner._proxy_pending = False
spawner._spawn_pending = False
try:
yield gen.with_timeout(timedelta(seconds=self.slow_spawn_timeout), f)
yield gen.with_timeout(timedelta(seconds=self.slow_spawn_timeout), finish_user_spawn())
except gen.TimeoutError:
# waiting_for_response indicates server process has started,
# but is yet to become responsive.
if not spawner._waiting_for_response:
if spawner._spawn_pending and not spawner._waiting_for_response:
# still in Spawner.start, which is taking a long time
# we shouldn't poll while spawn is incomplete.
self.log.warning("User %s is slow to start (timeout=%s)",
user_server_name, self.slow_spawn_timeout)
# schedule finish for when the user finishes spawning
IOLoop.current().add_future(f, finish_user_spawn)
else:
# start has finished, but the server hasn't come up
# check if the server died while we were waiting
status = yield user.spawner.poll()
if status is None:
# hit timeout, but server's running. Hope that it'll show up soon enough,
# though it's possible that it started at the wrong URL
self.log.warning("User %s is slow to become responsive (timeout=%s)",
user_server_name, self.slow_spawn_timeout)
self.log.debug("Expecting server for %s at: %s", user_server_name, spawner.server.url)
# schedule finish for when the user finishes spawning
IOLoop.current().add_future(f, finish_user_spawn)
else:
toc = IOLoop.current().time()
self.statsd.timing('spawner.failure', (toc - tic) * 1000)
raise web.HTTPError(500, "Spawner failed to start [status=%s]" % status)
else:
yield finish_user_spawn()
user_server_name, self.slow_spawn_timeout)
return
# start has finished, but the server hasn't come up
# check if the server died while we were waiting
status = yield spawner.poll()
if status is not None:
toc = IOLoop.current().time()
self.statsd.timing('spawner.failure', (toc - tic) * 1000)
raise web.HTTPError(500, "Spawner failed to start [status=%s]" % status)
if spawner._waiting_for_response:
# hit timeout waiting for response, but server's running.
# Hope that it'll show up soon enough,
# though it's possible that it started at the wrong URL
self.log.warning("User %s is slow to become responsive (timeout=%s)",
user_server_name, self.slow_spawn_timeout)
self.log.debug("Expecting server for %s at: %s",
user_server_name, spawner.server.url)
if spawner._proxy_pending:
# User.spawn finished, but it hasn't been added to the proxy
# Could be due to load or a slow proxy
self.log.warning("User %s is slow to be added to the proxy (timeout=%s)",
user_server_name, self.slow_spawn_timeout)
@gen.coroutine
def user_stopped(self, user, server_name):
@@ -501,36 +513,37 @@ class BaseHandler(RequestHandler):
if name not in user.spawners:
raise KeyError("User %s has no such spawner %r", user.name, name)
spawner = user.spawners[name]
if spawner._stop_pending:
raise RuntimeError("Stop already pending for: %s:%s" % (user.name, name))
tic = IOLoop.current().time()
yield self.proxy.delete_user(user, name)
f = user.stop()
@gen.coroutine
def finish_stop(f=None):
"""Finish the stop action by noticing that the user is stopped.
if spawner.pending:
raise RuntimeError("%s pending %s" % (user_server_name, spawner.pending))
# set user._stop_pending before doing anything async
# to avoid races
spawner._stop_pending = True
If the spawner is slow to stop, this is passed as an async callback,
otherwise it is called immediately.
@gen.coroutine
def stop():
"""Stop the server
1. remove it from the proxy
2. stop the server
3. notice that it stopped
"""
if f and f.exception() is not None:
# failed, don't do anything
return
tic = IOLoop.current().time()
try:
yield self.proxy.delete_user(user, name)
yield user.stop(name)
finally:
spawner._stop_pending = False
toc = IOLoop.current().time()
self.log.info("User %s server took %.3f seconds to stop", user.name, toc-tic)
self.log.info("User %s server took %.3f seconds to stop", user.name, toc - tic)
try:
yield gen.with_timeout(timedelta(seconds=self.slow_stop_timeout), f)
yield gen.with_timeout(timedelta(seconds=self.slow_stop_timeout), stop())
except gen.TimeoutError:
if spawner._stop_pending:
# hit timeout, but stop is still pending
self.log.warning("User %s:%s server is slow to stop", user.name, name)
# schedule finish for when the server finishes stopping
IOLoop.current().add_future(f, finish_stop)
else:
raise
else:
yield finish_stop()
#---------------------------------------------------------------
# template rendering
@@ -653,7 +666,8 @@ class UserSpawnHandler(BaseHandler):
# logged in as correct user, spawn the server
spawner = current_user.spawner
if spawner._spawn_pending or spawner._proxy_pending:
if spawner.pending:
self.log.info("%s is pending %s", spawner._log_name, spawner.pending)
# spawn has started, but not finished
self.statsd.incr('redirects.user_spawn_pending', 1)
html = self.render_template("spawn_pending.html", user=current_user)

View File

@@ -84,10 +84,11 @@ class LoginHandler(BaseHandler):
if user:
already_running = False
if user.spawner:
if user.spawner.ready:
status = yield user.spawner.poll()
already_running = (status is None)
if not already_running and not user.spawner.options_form:
if not already_running and not user.spawner.options_form \
and not user.spawner.pending:
# logging in triggers spawn
yield self.spawn_single_user(user)
self.redirect(self.get_next_url())

View File

@@ -115,6 +115,10 @@ class SpawnHandler(BaseHandler):
self.log.warning("User is already running: %s", url)
self.redirect(url)
return
if user.spawner.pending:
raise web.HTTPError(
400, "%s is pending %s" % (user.spawner._log_name, user.spawner.pending)
)
form_options = {}
for key, byte_list in self.request.body_arguments.items():
form_options[key] = [ bs.decode('utf8') for bs in byte_list ]

View File

@@ -231,9 +231,10 @@ class Proxy(LoggingConfigurable):
user.name, spawner.proxy_spec, spawner.server.host,
)
if spawner._spawn_pending:
if spawner.pending and spawner.pending != 'spawn':
raise RuntimeError(
"User %s's spawn is pending, shouldn't be added to the proxy yet!", user.name)
"%s is pending %s, shouldn't be added to the proxy yet!" % (spawner._log_name, spawner.pending)
)
yield self.add_route(
spawner.proxy_spec,
@@ -326,7 +327,7 @@ class Proxy(LoggingConfigurable):
spec, route['target'], spawner.server,
)
futures.append(self.add_user(user, name))
elif spawner._proxy_pending:
elif spawner._spawn_pending:
good_routes.add(spawner.proxy_spec)
# check service routes

View File

@@ -49,17 +49,29 @@ class Spawner(LoggingConfigurable):
# private attributes for tracking status
_spawn_pending = False
_start_pending = False
_stop_pending = False
_proxy_pending = False
_waiting_for_response = False
@property
def _log_name(self):
"""Return username:servername or username
Used in logging for consistency with named servers.
"""
if self.name:
return '%s:%s' % (self.user.name, self.name)
else:
return self.user.name
@property
def pending(self):
"""Return the current pending event, if any
Return False if nothing is pending.
"""
if self._spawn_pending or self._proxy_pending:
if self._spawn_pending:
return 'spawn'
elif self._stop_pending:
return 'stop'

View File

@@ -38,6 +38,27 @@ def test_create_named_server(app, named_servers):
assert prefix == user.spawners[servername].server.base_url
assert prefix.endswith('/user/%s/%s/' % (username, servername))
r = yield api_request(app, 'users', username)
r.raise_for_status()
user_model = r.json()
user_model.pop('last_activity')
assert user_model == {
'name': username,
'groups': [],
'kind': 'user',
'admin': False,
'pending': None,
'server': None,
'servers': {
name: {
'name': name,
'url': url_path_join(user.url, name, '/'),
}
for name in ['1', servername]
},
}
@pytest.mark.gen_test
def test_delete_named_server(app, named_servers):
@@ -69,9 +90,9 @@ def test_delete_named_server(app, named_servers):
'servers': {
name: {
'name': name,
'url': url_path_join(user.url, name),
'url': url_path_join(user.url, name, '/'),
}
for name in ['1', servername]
for name in ['1']
},
}

View File

@@ -317,8 +317,6 @@ class User(HasTraits):
url of the server will be /user/:name/:server_name
"""
db = self.db
if self.allow_named_servers and not server_name:
server_name = default_server_name(self)
base_url = url_path_join(self.base_url, server_name) + '/'
@@ -368,7 +366,7 @@ class User(HasTraits):
if (authenticator):
yield gen.maybe_future(authenticator.pre_spawn_start(self, spawner))
spawner._spawn_pending = True
spawner._start_pending = True
# wait for spawner.start to return
try:
# run optional preparation work to bootstrap the notebook
@@ -434,6 +432,7 @@ class User(HasTraits):
user=self.name,
), exc_info=True)
# raise original exception
spawner._start_pending = False
raise e
spawner.start_polling()
@@ -475,7 +474,7 @@ class User(HasTraits):
_check_version(__version__, server_version, self.log)
finally:
spawner._waiting_for_response = False
spawner._spawn_pending = False
spawner._start_pending = False
return self
@gen.coroutine
@@ -486,6 +485,7 @@ class User(HasTraits):
"""
spawner = self.spawners[server_name]
spawner._spawn_pending = False
spawner._start_pending = False
spawner.stop_polling()
spawner._stop_pending = True
try: