def validate_host_viable():
# this is something ccm does when starting, but preemptively check to avoid
# spinning up the cluster if it's not going to work
try:
common.check_socket_available(('::1', 9042))
except:
raise unittest.SkipTest('failed binding ipv6 loopback ::1 on 9042')
def load(self, options):
for itf in self.network_interfaces.values():
if itf:
common.check_socket_available(itf)
cdir = self.get_cassandra_dir()
loader_bin = common.join_bin(cdir, "bin", "sstableloader")
env = common.make_cassandra_env(cdir, self.get_path())
if not "-d" in options:
l = [node.network_interfaces["storage"][0] for node in self.cluster.nodes.values() if node.is_live()]
options = ["-d", ",".join(l)] + options
# print "Executing with", options
os.execve(loader_bin, [common.platform_binary("sstableloader")] + options, env)
def start(self, no_wait=False, verbose=False, wait_for_binary_proto=False,
wait_other_notice=True, jvm_args=None, profile_options=None,
quiet_start=False, allow_root=False):
if jvm_args is None:
jvm_args = []
common.assert_jdk_valid_for_cassandra_version(self.cassandra_version())
# check whether all loopback aliases are available before starting any nodes
for node in list(self.nodes.values()):
if not node.is_running():
for itf in node.network_interfaces.values():
if itf is not None:
if not common.check_socket_available(itf, return_on_error=True):
addr, port = itf
common.error("Inet address {}:{} is not available; a cluster may already be running or you may need to add the loopback alias".format(addr, port))
sys.exit(1)
started = []
for node in list(self.nodes.values()):
if not node.is_running():
mark = 0
if os.path.exists(node.logfilename()):
mark = node.mark_log()
p = node.start(update_pid=False, jvm_args=jvm_args, profile_options=profile_options, verbose=verbose, quiet_start=quiet_start, allow_root=allow_root)
# Prior to JDK8, starting every node at once could lead to a
# nanotime collision where the RNG that generates a node's tokens
# gives identical tokens to several nodes. Thus, we stagger
# the node starts
if common.get_jdk_version() < '1.8':
time.sleep(1)
started.append((node, p, mark))
if no_wait:
time.sleep(2) # waiting 2 seconds to check for early errors and for the pid to be set
else:
for node, p, mark in started:
try:
start_message = "Listening for thrift clients..." if self.cassandra_version() < "2.2" else "Starting listening for CQL clients"
node.watch_log_for(start_message, timeout=60, process=p, verbose=verbose, from_mark=mark)
except RuntimeError:
return None
self.__update_pids(started)
for node, p, _ in started:
if not node.is_running():
raise NodeError("Error starting {0}.".format(node.name), p)
if not no_wait:
if wait_other_notice:
for (node, _, mark), (other_node, _, _) in itertools.permutations(started, 2):
node.watch_log_for_alive(other_node, from_mark=mark)
if wait_for_binary_proto:
for node, p, mark in started:
node.wait_for_binary_interface(process=p, verbose=verbose, from_mark=mark)
return started
def start(self,
join_ring=True,
no_wait=False,
verbose=False,
update_pid=True,
wait_other_notice=False,
replace_token=None,
replace_address=None,
jvm_args=[],
wait_for_binary_proto=False,
profile_options=None,
use_jna=False):
"""
Start the node. Options includes:
- join_ring: if false, start the node with -Dcassandra.join_ring=False
- no_wait: by default, this method returns when the node is started and listening to clients.
If no_wait=True, the method returns sooner.
- wait_other_notice: if True, this method returns only when all other live node of the cluster
have marked this node UP.
- replace_token: start the node with the -Dcassandra.replace_token option.
- replace_address: start the node with the -Dcassandra.replace_address option.
"""
if self.is_running():
raise NodeError("%s is already running" % self.name)
for itf in list(self.network_interfaces.values()):
if itf is not None and replace_address is None:
common.check_socket_available(itf)
if wait_other_notice:
marks = [ (node, node.mark_log()) for node in list(self.cluster.nodes.values()) if node.is_running() ]
cdir = self.get_install_dir()
launch_bin = common.join_bin(cdir, 'bin', 'dse')
# Copy back the dse scripts since profiling may have modified it the previous time
shutil.copy(launch_bin, self.get_bin_dir())
launch_bin = common.join_bin(self.get_path(), 'bin', 'dse')
# If Windows, change entries in .bat file to split conf from binaries
if common.is_win():
self.__clean_bat()
if profile_options is not None:
config = common.get_config()
if not 'yourkit_agent' in config:
raise NodeError("Cannot enable profile. You need to set 'yourkit_agent' to the path of your agent in a {0}/config".format(common.get_default_path_display_name()))
cmd = '-agentpath:%s' % config['yourkit_agent']
if 'options' in profile_options:
cmd = cmd + '=' + profile_options['options']
print_(cmd)
# Yes, it's fragile as shit
pattern=r'cassandra_parms="-Dlog4j.configuration=log4j-server.properties -Dlog4j.defaultInitOverride=true'
common.replace_in_file(launch_bin, pattern, ' ' + pattern + ' ' + cmd + '"')
os.chmod(launch_bin, os.stat(launch_bin).st_mode | stat.S_IEXEC)
env = common.make_dse_env(self.get_install_dir(), self.get_path())
if common.is_win():
self._clean_win_jmx();
pidfile = os.path.join(self.get_path(), 'cassandra.pid')
args = [launch_bin, 'cassandra']
if self.workload is not None:
if 'hadoop' in self.workload:
args.append('-t')
if 'solr' in self.workload:
args.append('-s')
if 'spark' in self.workload:
args.append('-k')
if 'cfs' in self.workload:
args.append('-c')
args += [ '-p', pidfile, '-Dcassandra.join_ring=%s' % str(join_ring) ]
if replace_token is not None:
args.append('-Dcassandra.replace_token=%s' % str(replace_token))
if replace_address is not None:
args.append('-Dcassandra.replace_address=%s' % str(replace_address))
if use_jna is False:
args.append('-Dcassandra.boot_without_jna=true')
args = args + jvm_args
process = None
if common.is_win():
# clean up any old dirty_pid files from prior runs
if (os.path.isfile(self.get_path() + "/dirty_pid.tmp")):
os.remove(self.get_path() + "/dirty_pid.tmp")
process = subprocess.Popen(args, cwd=self.get_bin_dir(), env=env, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
else:
process = subprocess.Popen(args, env=env, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
# Our modified batch file writes a dirty output with more than just the pid - clean it to get in parity
# with *nix operation here.
if common.is_win():
self.__clean_win_pid()
self._update_pid(process)
elif update_pid:
if no_wait:
time.sleep(2) # waiting 2 seconds nevertheless to check for early errors and for the pid to be set
else:
for line in process.stdout:
if verbose:
print_(line.rstrip('\n'))
self._update_pid(process)
if not self.is_running():
raise NodeError("Error starting node %s" % self.name, process)
if wait_other_notice:
for node, mark in marks:
node.watch_log_for_alive(self, from_mark=mark)
if wait_for_binary_proto:
self.watch_log_for("Starting listening for CQL clients")
# we're probably fine at that point but just wait some tiny bit more because
# the msg is logged just before starting the binary protocol server
time.sleep(0.2)
if self.cluster.hasOpscenter():
self._start_agent()
return process
def start(self,
join_ring=True,
no_wait=False,
verbose=False,
update_pid=True,
wait_other_notice=False,
replace_token=None,
replace_address=None,
jvm_args=None,
wait_for_binary_proto=False,
profile_options=None,
use_jna=False,
quiet_start=False):
"""
Start the node. Options includes:
- join_ring: if false, start the node with -Dcassandra.join_ring=False
- no_wait: by default, this method returns when the node is started
and listening to clients.
If no_wait=True, the method returns sooner.
- wait_other_notice: if True, this method returns only when all other
live node of the cluster
have marked this node UP.
- replace_token: start the node with the -Dcassandra.replace_token
option.
- replace_address: start the node with the
-Dcassandra.replace_address option.
"""
if jvm_args is None:
jvm_args = []
scylla_cassandra_mapping = {
'-Dcassandra.replace_address_first_boot':
'--replace-address-first-boot'
}
# Replace args in the form
# ['-Dcassandra.foo=bar'] to ['-Dcassandra.foo', 'bar']
translated_args = []
new_jvm_args = []
for jvm_arg in jvm_args:
if '=' in jvm_arg:
split_option = jvm_arg.split("=")
e_msg = ("Option %s not in the form '-Dcassandra.foo=bar'. "
"Please check your test" % jvm_arg)
assert len(split_option) == 2, e_msg
option, value = split_option
# If we have information on how to translate the jvm option,
# translate it
if option in scylla_cassandra_mapping:
translated_args += [
scylla_cassandra_mapping[option], value
]
# Otherwise, just pass it as is
else:
new_jvm_args.append(jvm_arg)
else:
new_jvm_args.append(jvm_arg)
jvm_args = new_jvm_args
if self.is_running():
raise NodeError("%s is already running" % self.name)
for itf in list(self.network_interfaces.values()):
if itf is not None and replace_address is None:
try:
common.check_socket_available(itf)
except Exception as msg:
print("{}. Looking for offending processes...".format(msg))
for proc in psutil.process_iter():
if any(self.cluster.ipprefix in cmd
for cmd in proc.cmdline()):
print("name={} pid={} cmdline={}".format(
proc.name(), proc.pid, proc.cmdline()))
raise msg
marks = []
if wait_other_notice:
marks = [(node, node.mark_log())
for node in list(self.cluster.nodes.values())
if node.is_running()]
self.mark = self.mark_log()
launch_bin = common.join_bin(self.get_path(), 'bin', 'scylla')
options_file = os.path.join(self.get_path(), 'conf', 'scylla.yaml')
os.chmod(launch_bin, os.stat(launch_bin).st_mode | stat.S_IEXEC)
# TODO: we do not support forcing specific settings
# TODO: workaround for api-address as we do not load it
# from config file scylla#59
conf_file = os.path.join(self.get_conf_dir(), common.SCYLLA_CONF)
with open(conf_file, 'r') as f:
data = yaml.safe_load(f)
jvm_args = jvm_args + ['--api-address', data['api_address']]
jvm_args = jvm_args + [
'--collectd-hostname',
'%s.%s' % (socket.gethostname(), self.name)
]
# Let's add jvm_args and the translated args
args = [
launch_bin, '--options-file', options_file, '--log-to-stdout', '1'
] + jvm_args + translated_args
# Lets search for default overrides in SCYLLA_EXT_OPTS
scylla_ext_opts = os.getenv('SCYLLA_EXT_OPTS', "").split()
opts_i = 0
orig_args = list(args)
while opts_i < len(scylla_ext_opts):
if scylla_ext_opts[opts_i].startswith("--scylla-manager="):
opts_i += 1
elif scylla_ext_opts[opts_i].startswith('-'):
add = False
if scylla_ext_opts[opts_i] not in orig_args:
add = True
args.append(scylla_ext_opts[opts_i])
opts_i += 1
while opts_i < len(scylla_ext_opts) and not scylla_ext_opts[
opts_i].startswith('-'):
if add:
args.append(scylla_ext_opts[opts_i])
opts_i += 1
if '--developer-mode' not in args:
args += ['--developer-mode', 'true']
if '--smp' not in args:
# If --smp is not passed from cmdline, use default (--smp 1)
args += ['--smp', str(self._smp)]
elif self._smp_set_during_test:
# If node.set_smp() is called during the test, ignore the --smp
# passed from the cmdline.
args[args.index('--smp') + 1] = str(self._smp)
else:
# Update self._smp based on command line parameter.
# It may be used below, along with self._mem_mb_per_cpu, for calculating --memory
self._smp = int(args[args.index('--smp') + 1])
if '--memory' not in args:
# If --memory is not passed from cmdline, use default (512M per cpu)
args += [
'--memory', '{}M'.format(self._mem_mb_per_cpu * self._smp)
]
elif self._mem_set_during_test:
# If node.set_mem_mb_per_cpu() is called during the test, ignore the --memory
# passed from the cmdline.
args[args.index('--memory') + 1] = '{}M'.format(
self._mem_mb_per_cpu * self._smp)
if '--default-log-level' not in args:
args += ['--default-log-level', self.__global_log_level]
# TODO add support for classes_log_level
if '--collectd' not in args:
args += ['--collectd', '0']
if '--cpuset' not in args:
args += ['--overprovisioned']
if '--prometheus-address' not in args:
args += ['--prometheus-address', data['api_address']]
if replace_address:
args += ['--replace-address', replace_address]
args += ['--unsafe-bypass-fsync', '1']
scylla_process = self._start_scylla(args, marks, update_pid,
wait_other_notice,
wait_for_binary_proto)
self._start_jmx(data)
if not self._wait_java_up(data):
e_msg = (
"Error starting node %s: unable to connect to scylla-jmx" %
self.name)
raise NodeError(e_msg, scylla_process)
self.is_running()
return scylla_process
def start(self,
join_ring=True,
no_wait=False,
verbose=False,
update_pid=True,
wait_other_notice=False,
replace_token=None,
replace_address=None,
jvm_args=[],
wait_for_binary_proto=False,
profile_options=None,
use_jna=False):
"""
Start the node. Options includes:
- join_ring: if false, start the node with -Dcassandra.join_ring=False
- no_wait: by default, this method returns when the node is started and listening to clients.
If no_wait=True, the method returns sooner.
- wait_other_notice: if True, this method returns only when all other live node of the cluster
have marked this node UP.
- replace_token: start the node with the -Dcassandra.replace_token option.
- replace_address: start the node with the -Dcassandra.replace_address option.
"""
if self.is_running():
raise NodeError("%s is already running" % self.name)
for itf in list(self.network_interfaces.values()):
if itf is not None and replace_address is None:
common.check_socket_available(itf)
if wait_other_notice:
marks = [ (node, node.mark_log()) for node in list(self.cluster.nodes.values()) if node.is_running() ]
cdir = self.get_cassandra_dir()
cass_bin = common.join_bin(cdir, 'bin', 'cassandra')
# Copy back the cassandra scripts since profiling may have modified it the previous time
shutil.copy(cass_bin, self.get_bin_dir())
cass_bin = common.join_bin(self.get_path(), 'bin', 'cassandra')
# If Windows, change entries in .bat file to split conf from binaries
if common.is_win():
self.__clean_bat()
if profile_options is not None:
config = common.get_config()
if not 'yourkit_agent' in config:
raise NodeError("Cannot enable profile. You need to set 'yourkit_agent' to the path of your agent in a ~/.ccm/config")
cmd = '-agentpath:%s' % config['yourkit_agent']
if 'options' in profile_options:
cmd = cmd + '=' + profile_options['options']
print_(cmd)
# Yes, it's fragile as shit
pattern=r'cassandra_parms="-Dlog4j.configuration=log4j-server.properties -Dlog4j.defaultInitOverride=true'
common.replace_in_file(cass_bin, pattern, ' ' + pattern + ' ' + cmd + '"')
os.chmod(cass_bin, os.stat(cass_bin).st_mode | stat.S_IEXEC)
env = common.make_cassandra_env(cdir, self.get_path())
if common.is_win():
self._clean_win_jmx();
pidfile = os.path.join(self.get_path(), 'cassandra.pid')
args = [ cass_bin, '-p', pidfile, '-Dcassandra.join_ring=%s' % str(join_ring) ]
if replace_token is not None:
args.append('-Dcassandra.replace_token=%s' % str(replace_token))
if replace_address is not None:
args.append('-Dcassandra.replace_address=%s' % str(replace_address))
if use_jna is False:
args.append('-Dcassandra.boot_without_jna=true')
args = args + jvm_args
process = None
if common.is_win():
# clean up any old dirty_pid files from prior runs
if (os.path.isfile(self.get_path() + "/dirty_pid.tmp")):
os.remove(self.get_path() + "/dirty_pid.tmp")
process = subprocess.Popen(args, cwd=self.get_bin_dir(), env=env, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
else:
process = subprocess.Popen(args, env=env, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
# Our modified batch file writes a dirty output with more than just the pid - clean it to get in parity
# with *nix operation here.
if common.is_win():
self.__clean_win_pid()
self._update_pid(process)
elif update_pid:
if no_wait:
time.sleep(2) # waiting 2 seconds nevertheless to check for early errors and for the pid to be set
else:
for line in process.stdout:
if verbose:
print_(line.rstrip('\n'))
self._update_pid(process)
if not self.is_running():
raise NodeError("Error starting node %s" % self.name, process)
if wait_other_notice:
for node, mark in marks:
node.watch_log_for_alive(self, from_mark=mark)
if wait_for_binary_proto:
self.watch_log_for("Starting listening for CQL clients")
# we're probably fine at that point but just wait some tiny bit more because
# the msg is logged just before starting the binary protocol server
time.sleep(0.2)
return process
def start(self, no_wait=False, verbose=False, wait_for_binary_proto=False,
wait_other_notice=True, jvm_args=None, profile_options=None,
quiet_start=False, allow_root=False):
if jvm_args is None:
jvm_args = []
common.assert_jdk_valid_for_cassandra_version(self.cassandra_version())
# check whether all loopback aliases are available before starting any nodes
for node in list(self.nodes.values()):
if not node.is_running():
for itf in node.network_interfaces.values():
if itf is not None:
if not common.check_socket_available(itf, return_on_error=True):
addr, port = itf
common.error("Inet address {}:{} is not available; a cluster may already be running or you may need to add the loopback alias".format(addr, port))
sys.exit(1)
started = []
for node in list(self.nodes.values()):
if not node.is_running():
mark = 0
if os.path.exists(node.logfilename()):
mark = node.mark_log()
p = node.start(update_pid=False, jvm_args=jvm_args, profile_options=profile_options, verbose=verbose, quiet_start=quiet_start, allow_root=allow_root)
# Prior to JDK8, starting every node at once could lead to a
# nanotime collision where the RNG that generates a node's tokens
# gives identical tokens to several nodes. Thus, we stagger
# the node starts
if common.get_jdk_version() < '1.8':
time.sleep(1)
started.append((node, p, mark))
if no_wait:
time.sleep(2) # waiting 2 seconds to check for early errors and for the pid to be set
else:
for node, p, mark in started:
try:
start_message = "Listening for thrift clients..." if self.cassandra_version() < "2.2" else "Starting listening for CQL clients"
node.watch_log_for(start_message, timeout=60, process=p, verbose=verbose, from_mark=mark)
except RuntimeError:
return None
self.__update_pids(started)
for node, p, _ in started:
if not node.is_running():
raise NodeError("Error starting {0}.".format(node.name), p)
if not no_wait:
if wait_other_notice:
for (node, _, mark), (other_node, _, _) in itertools.permutations(started, 2):
node.watch_log_for_alive(other_node, from_mark=mark)
if wait_for_binary_proto:
for node, p, mark in started:
node.wait_for_binary_interface(process=p, verbose=verbose, from_mark=mark)
return started
def start(self,
join_ring=True,
no_wait=False,
verbose=False,
update_pid=True,
wait_other_notice=False,
replace_token=None,
replace_address=None,
jvm_args=None,
wait_for_binary_proto=False,
profile_options=None,
use_jna=False,
quiet_start=False,
allow_root=False):
"""
Start the node. Options includes:
- join_ring: if false, start the node with -Dcassandra.join_ring=False
- no_wait: by default, this method returns when the node is started and listening to clients.
If no_wait=True, the method returns sooner.
- wait_other_notice: if True, this method returns only when all other live node of the cluster
have marked this node UP.
- replace_token: start the node with the -Dcassandra.replace_token option.
- replace_address: start the node with the -Dcassandra.replace_address option.
"""
if jvm_args is None:
jvm_args = []
if self.is_running():
raise NodeError("%s is already running" % self.name)
for itf in list(self.network_interfaces.values()):
if itf is not None and replace_address is None:
common.check_socket_available(itf)
if wait_other_notice:
marks = [(node, node.mark_log()) for node in list(self.cluster.nodes.values()) if node.is_running()]
self.mark = self.mark_log()
cdir = self.get_install_dir()
launch_bin = common.join_bin(cdir, 'bin', 'dse')
# Copy back the dse scripts since profiling may have modified it the previous time
shutil.copy(launch_bin, self.get_bin_dir())
launch_bin = common.join_bin(self.get_path(), 'bin', 'dse')
# If Windows, change entries in .bat file to split conf from binaries
if common.is_win():
self.__clean_bat()
if profile_options is not None:
config = common.get_config()
if 'yourkit_agent' not in config:
raise NodeError("Cannot enable profile. You need to set 'yourkit_agent' to the path of your agent in a {0}/config".format(common.get_default_path_display_name()))
cmd = '-agentpath:%s' % config['yourkit_agent']
if 'options' in profile_options:
cmd = cmd + '=' + profile_options['options']
print_(cmd)
# Yes, it's fragile as shit
pattern = r'cassandra_parms="-Dlog4j.configuration=log4j-server.properties -Dlog4j.defaultInitOverride=true'
common.replace_in_file(launch_bin, pattern, ' ' + pattern + ' ' + cmd + '"')
os.chmod(launch_bin, os.stat(launch_bin).st_mode | stat.S_IEXEC)
env = self.get_env()
if common.is_win():
self._clean_win_jmx()
pidfile = os.path.join(self.get_path(), 'cassandra.pid')
args = [launch_bin, 'cassandra']
for workload in self.workloads:
if 'hadoop' in workload:
args.append('-t')
if 'solr' in workload:
args.append('-s')
if 'spark' in workload:
args.append('-k')
if 'cfs' in workload:
args.append('-c')
if 'graph' in workload:
args.append('-g')
args += ['-p', pidfile, '-Dcassandra.join_ring=%s' % str(join_ring)]
args += ['-Dcassandra.logdir=%s' % os.path.join(self.get_path(), 'logs')]
if replace_token is not None:
args.append('-Dcassandra.replace_token=%s' % str(replace_token))
if replace_address is not None:
args.append('-Dcassandra.replace_address=%s' % str(replace_address))
if use_jna is False:
args.append('-Dcassandra.boot_without_jna=true')
if allow_root:
args.append('-R')
args = args + jvm_args
self._delete_old_pid()
process = None
FNULL = open(os.devnull, 'w')
if common.is_win():
# clean up any old dirty_pid files from prior runs
if (os.path.isfile(self.get_path() + "/dirty_pid.tmp")):
os.remove(self.get_path() + "/dirty_pid.tmp")
process = subprocess.Popen(args, cwd=self.get_bin_dir(), env=env, stdout=FNULL, stderr=subprocess.PIPE)
else:
process = subprocess.Popen(args, env=env, stdout=FNULL, stderr=subprocess.PIPE)
# Our modified batch file writes a dirty output with more than just the pid - clean it to get in parity
# with *nix operation here.
if common.is_win():
self.__clean_win_pid()
self._update_pid(process)
elif update_pid:
self._update_pid(process)
if not self.is_running():
raise NodeError("Error starting node %s" % self.name, process)
if wait_other_notice:
for node, mark in marks:
node.watch_log_for_alive(self, from_mark=mark)
if wait_for_binary_proto:
self.wait_for_binary_interface(from_mark=self.mark)
if self.cluster.hasOpscenter():
self._start_agent()
return process
