def matching_image_for_host(distro=None, release=None, *args, **kwargs):
if distro is None or release is None:
# TODO: Use redhat-release, &c
rel_string = Run(data=True)(["bash", "-c", """
set -o errexit -o nounset -o pipefail
( source /etc/os-release && tr A-Z a-z <<<"$ID\t$VERSION_ID" )
"""])
probed_distro, probed_release = rel_string.strip().split()
distro, release = (distro or probed_distro, release or probed_release)
return image_token("%s:%s" % (distro, release), *args, **kwargs)
def remove(self, *args, **kwargs):
errors = 0
lk = deimos.flock.LK(self.lock, LOCK_EX|LOCK_NB)
try:
lk.lock()
except deimos.flock.Err:
msg = "Lock unavailable -- is cleanup already running?"
if self.optimistic:
log.info(msg)
return 0
else:
log.error(msg)
raise e
try:
for d in self.dirs(*args, **kwargs):
state = deimos.state.state(d)
if state is None:
log.warning("Not able to load state from: %s", d)
continue
try:
cmd = ["rm", "-rf", d + "/"]
cmd += [state._mesos()]
if state.cid() is not None:
cmd += [state._docker()]
Run()(cmd)
except subprocess.CalledProcessError:
errors += 1
finally:
lk.unlock()
if errors != 0:
log.error("There were failures on %d directories", errors)
return 4
def refresh_docker_image_info(image):
try:
text = Run(data=True)(docker("inspect", image))
parsed = json.loads(text)[0]
images[image] = parsed
return parsed
except subprocess.CalledProcessError as e:
return None
def probe(ident, quiet=False):
fields = "{{.ID}} {{.State.Pid}} {{.State.ExitCode}}"
level = logging.DEBUG if quiet else logging.WARNING
argv = docker("inspect", "--format=" + fields, ident)
run = Run(data=True, error_level=level)
text = run(argv).strip()
cid, pid, exit = text.split()
return Status(cid=cid, pid=pid, exit=(exit if pid == 0 else None))
def stop_docker_and_resume(self, signum):
if self.state is not None and self.state.cid() is not None:
cid = self.state.cid()
log.info("Trying to stop Docker container: %s", cid)
try:
Run()(deimos.docker.stop(cid))
except subprocess.CalledProcessError:
pass
return deimos.sig.Resume()
def containers(self, *args):
log.info(" ".join(args))
data = Run(data=True)(deimos.docker.docker("ps", "--no-trunc", "-q"))
mesos_ids = []
for line in data.splitlines():
cid = line.strip()
state = deimos.state.State(self.state_root, docker_id=cid)
if not state.exists():
continue
try:
state.lock("wait", LOCK_SH | LOCK_NB)
except deimos.flock.Err: # LOCK_EX held, so launch() is running
mesos_ids += [state.mesos_container_id()]
containers = Containers()
for mesos_id in mesos_ids:
container = containers.containers.add()
container.value = mesos_id
recordio.writeProto(containers)
return 0
def containers(self, *args):
log.info(" ".join(args))
data = Run(data=True)(deimos.docker.docker("ps", "--no-trunc", "-q"))
mesos_ids = []
for line in data.splitlines():
cid = line.strip()
state = deimos.state.State(self.state_root, docker_id=cid)
if not state.exists():
continue
try:
state.lock("wait", LOCK_SH | LOCK_NB)
except deimos.flock.Err: # LOCK_EX held, so launch() is running
mesos_ids += [state.mesos_container_id()]
containers = Containers()
for mesos_id in mesos_ids:
container = containers.containers.add()
container.value = mesos_id
recordio.writeProto(containers)
return 0
def destroy(self, destroy_pb, *args):
log.info(" ".join(args))
container_id = destroy_pb.container_id.value
state = deimos.state.State(self.state_root, mesos_id=container_id)
state.await_launch()
lk_d = state.lock("destroy", LOCK_EX)
if state.exit() is None:
Run()(deimos.docker.stop(state.cid()))
else:
log.info("Container is stopped")
return 0
def destroy(self, container_id, *args):
log.info(" ".join([container_id] + list(args)))
state = deimos.state.State(self.state_root, mesos_id=container_id)
state.await_launch()
lk_d = state.lock("destroy", LOCK_EX)
if state.exit() is not None:
Run()(deimos.docker.stop(state.cid()))
else:
log.info("Container is stopped")
if not sys.stdout.closed:
# If we're called as part of the signal handler set up by launch,
# STDOUT is probably closed already. Writing the Protobuf would
# only result in a bevy of error messages.
proto_out(protos.ExternalStatus, message="destroy: ok")
return 0
def stop_docker_and_resume(self, signum):
if self.state is not None and self.state.cid() is not None:
cid = self.state.cid()
log.info("Trying to stop Docker container: %s", cid)
container_id = self.state.docker_id
log.debug("Unwiring the container %s from MidoNet", container_id)
try:
log.info("state = %s", dir(self.state))
midonet.unwire_container_from_midonet(container_id)
log.debug("Successfully unwired the container %s from MidoNet "\
"bridge", container_id)
except Exception as ex:
log.error(traceback.format_exc())
try:
Run()(deimos.docker.stop(cid))
except subprocess.CalledProcessError:
pass
return deimos.sig.Resume()
def destroy(self, destroy_pb, *args):
log.info(" ".join(args))
container_id = destroy_pb.container_id.value
state = deimos.state.State(self.state_root, mesos_id=container_id)
state.await_launch()
lk_d = state.lock("destroy", LOCK_EX)
if state.exit() is None:
container_id = state.docker_id
log.debug("Unwiring the container %s from MidoNet", container_id)
try:
log.info("state_root = %s", dir(state))
midonet.unwire_container_from_midonet(container_id)
log.debug("Successfully unwired the container %s from MidoNet " \
"bridge", container_id)
except Exception as ex:
log.error(traceback.format_exc())
Run()(deimos.docker.stop(state.cid()))
else:
log.info("Container is stopped")
return 0
def launch(self, launch_pb, *args):
log.info(" ".join(args))
fork = False if "--no-fork" in args else True
deimos.sig.install(self.log_signal)
run_options = []
launchy = deimos.mesos.Launch(launch_pb)
state = deimos.state.State(self.state_root,
mesos_id=launchy.container_id)
state.push()
lk_l = state.lock("launch", LOCK_EX)
state.executor_id = launchy.executor_id
state.push()
state.ids()
mesos_directory() # Redundant?
if launchy.directory:
os.chdir(launchy.directory)
# TODO: if launchy.user:
# os.seteuid(launchy.user)
url, options = launchy.container
options, trailing_argv = split_on(options, "//")
url, options = self.container_settings.override(url, options)
true_argv = launchy.argv if trailing_argv is None else trailing_argv
image = self.determine_image(url, launchy)
log.info("image = %s", image)
run_options += ["--sig-proxy"]
run_options += ["--rm"] # This is how we ensure container cleanup
run_options += ["--cidfile", state.resolve("cid")]
place_uris(launchy, self.shared_dir, self.optimistic_unpack)
run_options += ["-w", self.workdir]
# Docker requires an absolute path to a source filesystem, separated
# from the bind path in the container with a colon, but the absolute
# path to the Mesos sandbox might have colons in it (TaskIDs with
# timestamps can cause this situation). So we create a soft link to it
# and mount that.
shared_full = os.path.abspath(self.shared_dir)
sandbox_symlink = state.sandbox_symlink(shared_full)
run_options += ["-v", "%s:%s" % (sandbox_symlink, self.workdir)]
cpus, mems = launchy.cpu_and_mem
env = launchy.env
run_options += options
env_dict = dict(env)
if env_dict.get("MIDONET_BRIDGE_ID", None):
run_options += ["--net=none"]
# We need to wrap the call to Docker in a call to the Mesos executor
# if no executor is passed as part of the task. We need to pass the
# MESOS_* environment variables in to the container if we're going to
# start an executor.
observer_argv = None
if launchy.needs_observer:
# NB: The "@@docker@@" variant is a work around for Mesos's option
# parser. There is a fix in the pipeline.
observer_argv = [
mesos_executor(), "--override",
deimos.path.me(), "observe", state.mesos_id
]
state.lock("observe", LOCK_EX | LOCK_NB) # Explanation of Locks
# When the observer is running, we would like its call to
# observe() to finish before all the wait(); and we'd like the
# observer to have a chance to report TASK_FINISHED before the
# calls to wait() report their results (which would result in a
# TASK_FAILED).
#
# For this reason, we take the "observe" lock in launch(), before
# we call the observer and before releasing the "launch" or "wait"
# locks.
#
# Calls to observe() actually skip locking "observe"; but wait()
# calls must take this lock. The "observe" lock is held by
# launch() until the observer executor completes, at which point
# we can be reasonably sure its status was propagated to the Mesos
# slave.
else:
env += mesos_env() + [("MESOS_DIRECTORY", self.workdir)]
self.place_dockercfg()
runner_argv = deimos.docker.run(run_options,
image,
true_argv,
env=env,
ports=launchy.ports,
cpus=cpus,
mems=mems)
log_mesos_env(logging.DEBUG)
observer = None
with open("stdout", "w") as o: # This awkward multi 'with' is a
with open("stderr", "w") as e: # concession to 2.6 compatibility
with open(os.devnull) as devnull:
log.info(deimos.cmd.present(runner_argv))
self.runner = subprocess.Popen(runner_argv,
stdin=devnull,
stdout=o,
stderr=e)
state.pid(self.runner.pid)
state.await_cid()
log.debug("Wiring the container to MidoNet")
try:
bridge_id = env_dict.get(
"MIDONET_BRIDGE_ID",
"78488c47-d1de-4d16-a27a-4e6419dc4f88")
container_id = state.docker_id
ip_addr = env_dict.get("MIDONET_IP_ADDRESS",
"192.168.100.42")
default_gw = env_dict.get("MIDONET_DEFAULT_GATEWAY",
None)
midonet.wire_container_to_midonet(
container_id, bridge_id, ip_addr, default_gw)
log.debug("Successfully wired the container %s to MidoNet " \
"bridge %s", container_id, bridge_id)
except Exception as ex:
log.error(traceback.format_exc())
state.push()
lk_w = state.lock("wait", LOCK_EX)
lk_l.unlock()
if fork:
pid = os.fork()
if pid is not 0:
state.ids()
log.info("Forking watcher into child...")
return
state.ids()
if observer_argv is not None:
log.info(deimos.cmd.present(observer_argv))
call = deimos.cmd.in_sh(observer_argv, allstderr=False)
# TODO: Collect these leaking file handles.
obs_out = open(state.resolve("observer.out"), "w+")
obs_err = open(state.resolve("observer.err"), "w+")
# If the Mesos executor sees LIBPROCESS_PORT=0 (which
# is passed by the slave) there are problems when it
# attempts to bind. ("Address already in use").
# Purging both LIBPROCESS_* net variables, to be safe.
for v in ["LIBPROCESS_PORT", "LIBPROCESS_IP"]:
if v in os.environ:
del os.environ[v]
observer = subprocess.Popen(call,
stdin=devnull,
stdout=obs_out,
stderr=obs_err,
close_fds=True)
data = Run(data=True)(deimos.docker.wait(state.cid()))
state.exit(data)
lk_w.unlock()
for p, arr in [(self.runner, runner_argv), (observer, observer_argv)]:
if p is None:
continue
thread = threading.Thread(target=p.wait)
thread.start()
thread.join(10)
if thread.is_alive():
log.warning(deimos.cmd.present(arr, "SIGTERM after 10s"))
p.terminate()
thread.join(1)
if thread.is_alive():
log.warning(deimos.cmd.present(arr, "SIGKILL after 1s"))
p.kill()
msg = deimos.cmd.present(arr, p.wait())
if p.wait() == 0:
log.info(msg)
else:
log.warning(msg)
return state.exit()
def place_dockercfg(self):
dockercfg = self.index_settings.dockercfg
if dockercfg is not None:
log.info("Copying to .dockercfg: %s" % dockercfg)
Run()(["cp", dockercfg, ".dockercfg"])
def pull(image):
Run(data=True)(docker("pull", image))
refresh_docker_image_info(image)
def launch(self, container_id, *args):
log.info(" ".join([container_id] + list(args)))
deimos.sig.install(self.sig_proxy)
run_options = []
state = deimos.state.State(self.state_root, mesos_id=container_id)
state.push()
lk_l = state.lock("launch", LOCK_EX)
mesos_directory()
task = protos.TaskInfo()
task.ParseFromString(sys.stdin.read())
for line in proto_lines(task):
log.debug(line)
state.executor_id = executor_id(task)
state.push()
state.ids()
url, options = self.container_settings.override(*container(task))
pre, image = url.split("docker:///")
if pre != "":
raise Err("URL '%s' is not a valid docker:// URL!" % url)
if image == "":
image = self.default_image(task)
log.info("image = %s", image)
run_options += ["--sig-proxy"]
run_options += ["--rm"] # This is how we ensure container cleanup
run_options += ["--cidfile", state.resolve("cid")]
place_uris(task, self.shared_dir, self.optimistic_unpack)
run_options += ["-w", self.workdir]
# Docker requires an absolute path to a source filesystem, separated
# from the bind path in the container with a colon, but the absolute
# path to the Mesos sandbox might have colons in it (TaskIDs with
# timestamps can cause this situation). So we create a soft link to it
# and mount that.
shared_full = os.path.abspath(self.shared_dir)
sandbox_symlink = state.sandbox_symlink(shared_full)
run_options += ["-v", "%s:%s" % (sandbox_symlink, self.workdir)]
cpus, mems = cpu_and_mem(task)
env = [(_.name, _.value) for _ in task.command.environment.variables]
run_options += options
# We need to wrap the call to Docker in a call to the Mesos executor
# if no executor is passed as part of the task. We need to pass the
# MESOS_* environment variables in to the container if we're going to
# start an executor.
observer_argv = None
if needs_executor_wrapper(task):
options = ["--mesos-executor", "--observer"]
if not (len(args) > 1 and args[0] in options):
raise Err("Task %s needs --observer to be set!" % state.eid())
observer_argv = list(
args[1:]) + [deimos.path.me(), "wait", "--docker"]
else:
env += mesos_env() + [("MESOS_DIRECTORY", self.workdir)]
runner_argv = deimos.docker.run(run_options,
image,
argv(task),
env=env,
ports=ports(task),
cpus=cpus,
mems=mems)
log_mesos_env(logging.DEBUG)
observer = None
with open("stdout", "w") as o: # This awkward multi 'with' is a
with open("stderr", "w") as e: # concession to 2.6 compatibility
with open(os.devnull) as devnull:
log.info(deimos.cmd.present(runner_argv))
self.runner = subprocess.Popen(runner_argv,
stdin=devnull,
stdout=o,
stderr=e)
state.pid(self.runner.pid)
state.await_cid()
state.push()
lk_w = state.lock("wait", LOCK_EX)
lk_l.unlock()
state.ids()
proto_out(protos.ExternalStatus, message="launch: ok")
sys.stdout.close() # Mark STDOUT as closed for Python code
os.close(
1) # Use low-level call to close OS side of STDOUT
if observer_argv is not None:
observer_argv += [state.cid()]
log.info(deimos.cmd.present(observer_argv))
call = deimos.cmd.in_sh(observer_argv)
# TODO: Collect these leaking file handles.
obs_out = open(state.resolve("observer.out"), "w+")
obs_err = open(state.resolve("observer.err"), "w+")
# If the Mesos executor sees LIBPROCESS_PORT=0 (which
# is passed by the slave) there are problems when it
# attempts to bind. ("Address already in use").
# Purging both LIBPROCESS_* net variables, to be safe.
for v in ["LIBPROCESS_PORT", "LIBPROCESS_IP"]:
if v in os.environ:
del os.environ[v]
observer = subprocess.Popen(call,
stdin=devnull,
stdout=obs_out,
stderr=obs_err,
close_fds=True)
data = Run(data=True)(deimos.docker.wait(state.cid()))
state.exit(data)
lk_w.unlock()
for p, arr in [(self.runner, runner_argv), (observer, observer_argv)]:
if p is None or p.wait() == 0:
continue
log.warning(deimos.cmd.present(arr, p.wait()))
return state.exit()
