def test_scheduler_state(self):
expected = Scheduler(FrameworkInfo(
user='******',
name='test_fw_name',
checkpoint=True))
expected.tasks = dict(taks1='cluster1', task2='cluster2')
self._state_provider.dump_scheduler_state(expected)
actual = self._state_provider.load_scheduler_state()
assert expected.framework_info == actual.framework_info
assert expected.tasks == actual.tasks
def test_scheduler_runs():
"""
Verifies that the scheduler successfully launches 3 "no-op" MySQL tasks.
NOTE: Due to the limitation of zake the scheduler's ZK operations are not propagated to
executors in separate processes but they are unit-tested separately.
"""
import mesos.native
# Make sure fake_mysos_executor.pex is available to be fetched by Mesos slave.
assert os.path.isfile('dist/fake_mysos_executor.pex')
storage = FakeStorage(SequentialThreadingHandler())
zk_client = FakeClient(storage=storage)
zk_client.start()
zk_url = "zk://fake_host/home/mysos/clusters"
cluster_name = "test_cluster"
num_nodes = 3
state_provider = LocalStateProvider(safe_mkdtemp())
framework_info = FrameworkInfo(user=getpass.getuser(),
name="mysos",
checkpoint=False)
state = Scheduler(framework_info)
scheduler = MysosScheduler(state, state_provider, getpass.getuser(),
os.path.abspath("dist/fake_mysos_executor.pex"),
"./fake_mysos_executor.pex", zk_client, zk_url,
Amount(40, Time.SECONDS), "/fakepath",
gen_encryption_key())
scheduler_driver = mesos.native.MesosSchedulerDriver(
scheduler, framework_info, "local")
scheduler_driver.start()
# Wait until the scheduler is connected and becomes available.
assert scheduler.connected.wait(30)
scheduler.create_cluster(cluster_name, "mysql_user", num_nodes)
# A slave is promoted to be the master.
deadline(
lambda: wait_for_master(
get_cluster_path(posixpath.join(zk_url, 'discover'), cluster_name),
zk_client), Amount(40, Time.SECONDS))
assert scheduler_driver.stop() == DRIVER_STOPPED
def main(args, options):
log.info("Options in use: %s", options)
if not options.api_port:
app.error('Must specify --port')
if not options.mesos_master:
app.error('Must specify --mesos_master')
if not options.framework_user:
app.error('Must specify --framework_user')
if not options.executor_uri:
app.error('Must specify --executor_uri')
if not options.executor_cmd:
app.error('Must specify --executor_cmd')
if not options.zk_url:
app.error('Must specify --zk_url')
if not options.admin_keypath:
app.error('Must specify --admin_keypath')
try:
election_timeout = parse_time(options.election_timeout)
framework_failover_timeout = parse_time(
options.framework_failover_timeout)
except InvalidTime as e:
app.error(e.message)
try:
_, zk_servers, zk_root = zookeeper.parse(options.zk_url)
except Exception as e:
app.error("Invalid --zk_url: %s" % e.message)
web_assets_dir = os.path.join(options.work_dir, "web")
pkgutil.unpack_assets(web_assets_dir, MYSOS_MODULE, ASSET_RELPATH)
log.info("Extracted web assets into %s" % options.work_dir)
fw_principal = None
fw_secret = None
if options.framework_authentication_file:
try:
with open(options.framework_authentication_file, "r") as f:
cred = yaml.load(f)
fw_principal = cred["principal"]
fw_secret = cred["secret"]
log.info(
"Loaded credential (principal=%s) for framework authentication"
% fw_principal)
except IOError as e:
app.error(
"Unable to read the framework authentication key file: %s"
% e)
except (KeyError, yaml.YAMLError) as e:
app.error(
"Invalid framework authentication key file format %s" % e)
log.info("Starting Mysos scheduler")
kazoo = KazooClient(zk_servers)
kazoo.start()
if options.state_storage == 'zk':
log.info("Using ZooKeeper (path: %s) for state storage" % zk_root)
state_provider = ZooKeeperStateProvider(kazoo, zk_root)
else:
log.info("Using local disk for state storage")
state_provider = LocalStateProvider(options.work_dir)
try:
state = state_provider.load_scheduler_state()
except StateProvider.Error as e:
app.error(e.message)
if state:
log.info("Successfully restored scheduler state")
framework_info = state.framework_info
if framework_info.HasField('id'):
log.info("Recovered scheduler's FrameworkID is %s" %
framework_info.id.value)
else:
log.info("No scheduler state to restore")
framework_info = FrameworkInfo(
user=options.framework_user,
name=FRAMEWORK_NAME,
checkpoint=True,
failover_timeout=framework_failover_timeout.as_(Time.SECONDS),
role=options.framework_role)
if fw_principal:
framework_info.principal = fw_principal
state = Scheduler(framework_info)
state_provider.dump_scheduler_state(state)
scheduler = MysosScheduler(state,
state_provider,
options.framework_user,
options.executor_uri,
options.executor_cmd,
kazoo,
options.zk_url,
election_timeout,
options.admin_keypath,
installer_args=options.installer_args,
backup_store_args=options.backup_store_args,
executor_environ=options.executor_environ,
framework_role=options.framework_role)
if fw_principal and fw_secret:
cred = Credential(principal=fw_principal, secret=fw_secret)
scheduler_driver = mesos.native.MesosSchedulerDriver(
scheduler, framework_info, options.mesos_master, cred)
else:
scheduler_driver = mesos.native.MesosSchedulerDriver(
scheduler, framework_info, options.mesos_master)
scheduler_driver.start()
server = HttpServer()
server.mount_routes(MysosServer(scheduler, web_assets_dir))
et = ExceptionalThread(target=server.run,
args=('0.0.0.0', options.api_port, 'cherrypy'))
et.daemon = True
et.start()
try:
# Wait for the scheduler to stop.
# The use of 'stopped' event instead of scheduler_driver.join() is necessary to stop the
# process with SIGINT.
while not scheduler.stopped.wait(timeout=0.5):
pass
except KeyboardInterrupt:
log.info('Interrupted, exiting.')
else:
log.info('Scheduler exited.')
app.shutdown(
1
) # Mysos scheduler is supposed to be long-running thus the use of exit status 1.
from mesos.native import MesosSchedulerDriver
from python_scheduler import PythonScheduler
from json import dumps, loads
from threading import Thread
from uuid import uuid1
import logging
import mesos
from mesos.interface.mesos_pb2 import FrameworkInfo
import os
import subprocess
import sys
import time
if __name__ == "__main__":
print sys.argv[1:]
framework = FrameworkInfo()
framework.name = "DistributedShell"
framework.user = "" # Let mesos fill current user
framework.role = "*"
framework.checkpoint = False
framework.failover_timeout = 0.0
#create instance of schedule and connect to mesos
scheduler = PythonScheduler()
#submit shell commands
task = " ".join(sys.argv[1:])
print "task: %s", task
scheduler.submitTasks(task)
mesosURL = "1.1.1.1:5050" # IP / hostname of mesos master
driver = MesosSchedulerDriver(scheduler, framework, mesosURL)
#run the driver
def main(args, options):
log.info("Options in use: %s", options)
if not options.api_port:
app.error('Must specify --port')
if not options.mesos_master:
app.error('Must specify --mesos_master')
if not options.framework_user:
app.error('Must specify --framework_user')
if not options.executor_uri:
app.error('Must specify --executor_uri')
if not options.executor_cmd:
app.error('Must specify --executor_cmd')
if not options.zk_url:
app.error('Must specify --zk_url')
if not options.admin_keypath:
app.error('Must specify --admin_keypath')
try:
election_timeout = parse_time(options.election_timeout)
framework_failover_timeout = parse_time(options.framework_failover_timeout)
except InvalidTime as e:
app.error(e.message)
try:
_, zk_servers, zk_root = zookeeper.parse(options.zk_url)
except Exception as e:
app.error("Invalid --zk_url: %s" % e.message)
web_assets_dir = os.path.join(options.work_dir, "web")
pkgutil.unpack_assets(web_assets_dir, MYSOS_MODULE, ASSET_RELPATH)
log.info("Extracted web assets into %s" % options.work_dir)
fw_principal = None
fw_secret = None
if options.framework_authentication_file:
try:
with open(options.framework_authentication_file, "r") as f:
cred = yaml.load(f)
fw_principal = cred["principal"]
fw_secret = cred["secret"]
log.info("Loaded credential (principal=%s) for framework authentication" % fw_principal)
except IOError as e:
app.error("Unable to read the framework authentication key file: %s" % e)
except (KeyError, yaml.YAMLError) as e:
app.error("Invalid framework authentication key file format %s" % e)
log.info("Starting Mysos scheduler")
kazoo = KazooClient(zk_servers)
kazoo.start()
if options.state_storage == 'zk':
log.info("Using ZooKeeper (path: %s) for state storage" % zk_root)
state_provider = ZooKeeperStateProvider(kazoo, zk_root)
else:
log.info("Using local disk for state storage")
state_provider = LocalStateProvider(options.work_dir)
try:
state = state_provider.load_scheduler_state()
except StateProvider.Error as e:
app.error(e.message)
if state:
log.info("Successfully restored scheduler state")
framework_info = state.framework_info
if framework_info.HasField('id'):
log.info("Recovered scheduler's FrameworkID is %s" % framework_info.id.value)
else:
log.info("No scheduler state to restore")
framework_info = FrameworkInfo(
user=options.framework_user,
name=FRAMEWORK_NAME,
checkpoint=True,
failover_timeout=framework_failover_timeout.as_(Time.SECONDS),
role=options.framework_role)
if fw_principal:
framework_info.principal = fw_principal
state = Scheduler(framework_info)
state_provider.dump_scheduler_state(state)
scheduler = MysosScheduler(
state,
state_provider,
options.framework_user,
options.executor_uri,
options.executor_cmd,
kazoo,
options.zk_url,
election_timeout,
options.admin_keypath,
installer_args=options.installer_args,
backup_store_args=options.backup_store_args,
executor_environ=options.executor_environ,
framework_role=options.framework_role)
if fw_principal and fw_secret:
cred = Credential(principal=fw_principal, secret=fw_secret)
scheduler_driver = mesos.native.MesosSchedulerDriver(
scheduler,
framework_info,
options.mesos_master,
cred)
else:
scheduler_driver = mesos.native.MesosSchedulerDriver(
scheduler,
framework_info,
options.mesos_master)
scheduler_driver.start()
server = HttpServer()
server.mount_routes(MysosServer(scheduler, web_assets_dir))
et = ExceptionalThread(
target=server.run, args=('0.0.0.0', options.api_port, 'cherrypy'))
et.daemon = True
et.start()
try:
# Wait for the scheduler to stop.
# The use of 'stopped' event instead of scheduler_driver.join() is necessary to stop the
# process with SIGINT.
while not scheduler.stopped.wait(timeout=0.5):
pass
except KeyboardInterrupt:
log.info('Interrupted, exiting.')
else:
log.info('Scheduler exited.')
app.shutdown(1) # Mysos scheduler is supposed to be long-running thus the use of exit status 1.