def quorum_unblock_cmd(lib, argv, modifiers):
"""
Options:
* --force - no error when removing non existing property and no warning
about this action
"""
modifiers.ensure_only_supported("--force")
if argv:
raise CmdLineInputError()
output, retval = utils.run(
["corosync-cmapctl", "-g", "runtime.votequorum.wait_for_all_status"]
)
if retval != 0:
utils.err("unable to check quorum status")
if output.split("=")[-1].strip() != "1":
utils.err("cluster is not waiting for nodes to establish quorum")
all_nodes, report_list = get_existing_nodes_names(
utils.get_corosync_conf_facade()
)
if report_list:
utils.process_library_reports(report_list)
unjoined_nodes = set(all_nodes) - set(utils.getCorosyncActiveNodes())
if not unjoined_nodes:
utils.err("no unjoined nodes found")
if not modifiers.get("--force"):
answer = utils.get_terminal_input(
(
"WARNING: If node(s) {nodes} are not powered off or they do"
+ " have access to shared resources, data corruption and/or"
+ " cluster failure may occur. Are you sure you want to"
+ " continue? [y/N] "
).format(nodes=", ".join(unjoined_nodes))
)
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
for node in unjoined_nodes:
# pass --force so no warning will be displayed
stonith.stonith_confirm(
lib, [node], parse_args.InputModifiers({"--force": ""})
)
output, retval = utils.run(
["corosync-cmapctl", "-s", "quorum.cancel_wait_for_all", "u8", "1"]
)
if retval != 0:
utils.err("unable to cancel waiting for nodes")
print("Quorum unblocked")
startup_fencing = utils.get_set_properties().get("startup-fencing", "")
utils.set_cib_property(
"startup-fencing",
"false" if startup_fencing.lower() != "false" else "true"
)
utils.set_cib_property("startup-fencing", startup_fencing)
print("Waiting for nodes canceled")
def quorum_unblock_cmd(lib, argv, modifiers):
"""
Options:
* --force - no error when removing non existing property and no warning
about this action
"""
modifiers.ensure_only_supported("--force")
if argv:
raise CmdLineInputError()
output, retval = utils.run(
["corosync-cmapctl", "-g", "runtime.votequorum.wait_for_all_status"]
)
if retval != 0:
utils.err("unable to check quorum status")
if output.split("=")[-1].strip() != "1":
utils.err("cluster is not waiting for nodes to establish quorum")
all_nodes, report_list = get_existing_nodes_names(
utils.get_corosync_conf_facade()
)
if report_list:
process_library_reports(report_list)
unjoined_nodes = set(all_nodes) - set(utils.getCorosyncActiveNodes())
if not unjoined_nodes:
utils.err("no unjoined nodes found")
if not modifiers.get("--force"):
answer = utils.get_terminal_input(
(
"WARNING: If node(s) {nodes} are not powered off or they do"
+ " have access to shared resources, data corruption and/or"
+ " cluster failure may occur. Are you sure you want to"
+ " continue? [y/N] "
).format(nodes=", ".join(unjoined_nodes))
)
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
for node in unjoined_nodes:
# pass --force so no warning will be displayed
stonith.stonith_confirm(
lib, [node], parse_args.InputModifiers({"--force": ""})
)
output, retval = utils.run(
["corosync-cmapctl", "-s", "quorum.cancel_wait_for_all", "u8", "1"]
)
if retval != 0:
utils.err("unable to cancel waiting for nodes")
print("Quorum unblocked")
startup_fencing = utils.get_set_properties().get("startup-fencing", "")
utils.set_cib_property(
"startup-fencing",
"false" if startup_fencing.lower() != "false" else "true"
)
utils.set_cib_property("startup-fencing", startup_fencing)
print("Waiting for nodes canceled")
def is_service_running(service):
"""
Used in module pcs.config
Commandline options: no options
"""
if utils.is_systemctl():
dummy_output, retval = utils.run(["systemctl", "status", service])
else:
dummy_output, retval = utils.run(["service", service, "status"])
return retval == 0
def quorum_unblock_cmd(argv):
if len(argv) > 0:
usage.quorum(["unblock"])
sys.exit(1)
if utils.is_rhel6():
utils.err("operation is not supported on CMAN clusters")
output, retval = utils.run(
["corosync-cmapctl", "-g", "runtime.votequorum.wait_for_all_status"]
)
if retval != 0:
utils.err("unable to check quorum status")
if output.split("=")[-1].strip() != "1":
utils.err("cluster is not waiting for nodes to establish quorum")
unjoined_nodes = (
set(utils.getNodesFromCorosyncConf())
-
set(utils.getCorosyncActiveNodes())
)
if not unjoined_nodes:
utils.err("no unjoined nodes found")
if "--force" not in utils.pcs_options:
answer = utils.get_terminal_input(
(
"WARNING: If node(s) {nodes} are not powered off or they do"
+ " have access to shared resources, data corruption and/or"
+ " cluster failure may occur. Are you sure you want to"
+ " continue? [y/N] "
).format(nodes=", ".join(unjoined_nodes))
)
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
for node in unjoined_nodes:
stonith.stonith_confirm([node], skip_question=True)
output, retval = utils.run(
["corosync-cmapctl", "-s", "quorum.cancel_wait_for_all", "u8", "1"]
)
if retval != 0:
utils.err("unable to cancel waiting for nodes")
print("Quorum unblocked")
startup_fencing = utils.get_set_properties().get("startup-fencing", "")
utils.set_cib_property(
"startup-fencing",
"false" if startup_fencing.lower() != "false" else "true"
)
utils.set_cib_property("startup-fencing", startup_fencing)
print("Waiting for nodes canceled")
def quorum_unblock_cmd(argv):
if len(argv) > 0:
usage.quorum(["unblock"])
sys.exit(1)
if utils.is_rhel6():
utils.err("operation is not supported on CMAN clusters")
output, retval = utils.run(
["corosync-cmapctl", "-g", "runtime.votequorum.wait_for_all_status"]
)
if retval != 0:
utils.err("unable to check quorum status")
if output.split("=")[-1].strip() != "1":
utils.err("cluster is not waiting for nodes to establish quorum")
unjoined_nodes = (
set(utils.getNodesFromCorosyncConf())
-
set(utils.getCorosyncActiveNodes())
)
if not unjoined_nodes:
utils.err("no unjoined nodes found")
if "--force" not in utils.pcs_options:
answer = utils.get_terminal_input(
(
"WARNING: If node(s) {nodes} are not powered off or they do"
+ " have access to shared resources, data corruption and/or"
+ " cluster failure may occur. Are you sure you want to"
+ " continue? [y/N] "
).format(nodes=", ".join(unjoined_nodes))
)
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
for node in unjoined_nodes:
stonith.stonith_confirm([node], skip_question=True)
output, retval = utils.run(
["corosync-cmapctl", "-s", "quorum.cancel_wait_for_all", "u8", "1"]
)
if retval != 0:
utils.err("unable to cancel waiting for nodes")
print("Quorum unblocked")
startup_fencing = utils.get_set_properties().get("startup-fencing", "")
utils.set_cib_property(
"startup-fencing",
"false" if startup_fencing.lower() != "false" else "true"
)
utils.set_cib_property("startup-fencing", startup_fencing)
print("Waiting for nodes canceled")
def quorum_unblock_cmd(lib, argv, modifiers):
"""
Options:
* --force - no error when removing non existing property and no warning
about this action
"""
modifiers.ensure_only_supported("--force")
if argv:
raise CmdLineInputError()
output, retval = utils.run(
["corosync-cmapctl", "-g", "runtime.votequorum.wait_for_all_status"])
if (retval == 1 and "Error CS_ERR_NOT_EXIST" in output) or (
retval == 0 and output.rsplit("=", maxsplit=1)[-1].strip() != "1"):
utils.err("cluster is not waiting for nodes to establish quorum")
if retval != 0:
utils.err("unable to check quorum status")
all_nodes, report_list = get_existing_nodes_names(
utils.get_corosync_conf_facade())
if report_list:
process_library_reports(report_list)
unjoined_nodes = set(all_nodes) - set(utils.getCorosyncActiveNodes())
if not unjoined_nodes:
utils.err("no unjoined nodes found")
if not utils.get_continue_confirmation_or_force(
f"If node(s) {format_list(unjoined_nodes)} are not powered off or they "
"do have access to shared resources, data corruption and/or cluster "
"failure may occur",
modifiers.get("--force"),
):
return
for node in unjoined_nodes:
# pass --force so no warning will be displayed
stonith.stonith_confirm(lib, [node],
parse_args.InputModifiers({"--force": ""}))
output, retval = utils.run(
["corosync-cmapctl", "-s", "quorum.cancel_wait_for_all", "u8", "1"])
if retval != 0:
utils.err("unable to cancel waiting for nodes")
print_to_stderr("Quorum unblocked")
startup_fencing = utils.get_set_properties().get("startup-fencing", "")
utils.set_cib_property(
"startup-fencing",
"false" if startup_fencing.lower() != "false" else "true",
)
utils.set_cib_property("startup-fencing", startup_fencing)
print_to_stderr("Waiting for nodes canceled")
def pcs(testfile, args = ""):
"""Run pcs with -f on specified file
Return tuple with:
shell stdoutdata
shell returncode
"""
if args == "":
args = testfile
testfile = temp_cib
arg_split = args.split()
arg_split_temp = []
in_quote = False
for arg in arg_split:
if in_quote:
arg_split_temp[-1] = arg_split_temp[-1] + " " + arg.replace("'","")
if arg.find("'") != -1:
in_quote = False
else:
arg_split_temp.append(arg.replace("'",""))
if arg.find("'") != -1 and not (arg[0] == "'" and arg[-1] == "'"):
in_quote = True
conf_opts = []
if "--corosync_conf" not in args:
corosync_conf = rc("corosync.conf")
conf_opts.append("--corosync_conf="+corosync_conf)
if "--cluster_conf" not in args:
cluster_conf = rc("cluster.conf")
conf_opts.append("--cluster_conf="+cluster_conf)
return utils.run([pcs_location, "-f", testfile] + conf_opts + arg_split_temp)
def cluster_status(lib, argv, modifiers):
"""
Options:
* -f - CIB file
* --request-timeout - HTTP timeout for checking status of pcsd, no effect
if -f is specified
"""
modifiers.ensure_only_supported("-f", "--request-timeout")
if argv:
raise CmdLineInputError()
(output, retval) = utils.run(["crm_mon", "-1", "-r"])
if retval != 0:
utils.err("cluster is not currently running on this node")
first_empty_line = False
print("Cluster Status:")
for line in output.splitlines():
if line == "":
if first_empty_line:
break
first_empty_line = True
continue
print("", line)
if not modifiers.is_specified("-f") and utils.hasCorosyncConf():
print()
print_pcsd_daemon_status(lib, modifiers)
def full_status():
if "--hide-inactive" in utils.pcs_options and "--full" in utils.pcs_options:
utils.err("you cannot specify both --hide-inactive and --full")
monitor_command = ["crm_mon", "--one-shot"]
if "--hide-inactive" not in utils.pcs_options:
monitor_command.append('--inactive')
if "--full" in utils.pcs_options:
monitor_command.extend(
["--show-detail", "--show-node-attributes", "--failcounts"]
)
output, retval = utils.run(monitor_command)
if (retval != 0):
utils.err("cluster is not currently running on this node")
if not utils.usefile or "--corosync_conf" in utils.pcs_options:
cluster_name = utils.getClusterName()
print("Cluster name: %s" % cluster_name)
if utils.stonithCheck():
print("WARNING: no stonith devices and stonith-enabled is not false")
if not utils.is_rhel6() and utils.corosyncPacemakerNodeCheck():
print("WARNING: corosync and pacemaker node names do not match (IPs used in setup?)")
print(output)
if not utils.usefile:
if "--full" in utils.pcs_options:
print_pcsd_daemon_status()
print()
utils.serviceStatus(" ")
def fixture_attrs(self, nodes, attrs=None):
attrs = dict() if attrs is None else attrs
xml_lines = ["<nodes>"]
for node_id, node_name in enumerate(nodes, 1):
xml_lines.extend(
[
'<node id="{0}" uname="{1}">'.format(node_id, node_name),
'<instance_attributes id="nodes-{0}">'.format(node_id),
]
)
nv = '<nvpair id="nodes-{id}-{name}" name="{name}" value="{val}"/>'
for name, value in attrs.get(node_name, dict()).items():
xml_lines.append(nv.format(id=node_id, name=name, val=value))
xml_lines.extend(["</instance_attributes>", "</node>"])
xml_lines.append("</nodes>")
utils_usefile_original = utils.usefile
utils_filename_original = utils.filename
utils.usefile = True
utils.filename = self.temp_cib.name
output, retval = utils.run(
["cibadmin", "--modify", "--xml-text", "\n".join(xml_lines)]
)
utils.usefile = utils_usefile_original
utils.filename = utils_filename_original
assert output == ""
assert retval == 0
def stonith_confirm(lib, argv, modifiers):
"""
Options:
* --force - do not warn user
"""
del lib
modifiers.ensure_only_supported("--force")
if len(argv) != 1:
utils.err("must specify one (and only one) node to confirm fenced")
node = argv.pop(0)
if not modifiers.get("--force"):
answer = utils.get_terminal_input(
(
"WARNING: If node {node} is not powered off or it does"
+ " have access to shared resources, data corruption and/or"
+ " cluster failure may occur. Are you sure you want to"
+ " continue? [y/N] "
).format(node=node)
)
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
args = ["stonith_admin", "-C", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to confirm fencing of node '%s'\n" % node + output)
else:
print("Node: %s confirmed fenced" % node)
def stonith_confirm(lib, argv, modifiers):
"""
Options:
* --force - do not warn user
"""
modifiers.ensure_only_supported("--force")
if len(argv) != 1:
utils.err("must specify one (and only one) node to confirm fenced")
node = argv.pop(0)
if not modifiers.get("--force"):
answer = utils.get_terminal_input(
("WARNING: If node {node} is not powered off or it does" +
" have access to shared resources, data corruption and/or" +
" cluster failure may occur. Are you sure you want to" +
" continue? [y/N] ").format(node=node))
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
args = ["stonith_admin", "-C", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to confirm fencing of node '%s'\n" % node + output)
else:
print("Node: %s confirmed fenced" % node)
def pcs(testfile, args):
"""
Run pcs with -f on specified file
Return tuple with:
shell stdoutdata
shell returncode
"""
arg_split = args.split()
arg_split_temp = []
in_quote = False
for arg in arg_split:
if in_quote:
arg_split_temp[-1] = arg_split_temp[-1] + " " + arg.replace(
"'", "")
if arg.find("'") != -1:
in_quote = False
else:
arg_split_temp.append(arg.replace("'", ""))
if arg.find("'") != -1 and not (arg[0] == "'" and arg[-1] == "'"):
in_quote = True
conf_opts = []
if "--corosync_conf" not in args:
corosync_conf = rc("corosync.conf")
conf_opts.append("--corosync_conf=" + corosync_conf)
if "--cluster_conf" not in args:
cluster_conf = rc("cluster.conf")
conf_opts.append("--cluster_conf=" + cluster_conf)
return utils.run([__pcs_location, "-f", testfile] + conf_opts +
arg_split_temp)
def fixture_attrs(self, nodes, attrs=None):
attrs = dict() if attrs is None else attrs
xml_lines = ['<nodes>']
for node_id, node_name in enumerate(nodes, 1):
xml_lines.extend([
'<node id="{0}" uname="{1}">'.format(node_id, node_name),
'<instance_attributes id="nodes-{0}">'.format(node_id),
])
nv = '<nvpair id="nodes-{id}-{name}" name="{name}" value="{val}"/>'
for name, value in attrs.get(node_name, dict()).items():
xml_lines.append(nv.format(id=node_id, name=name, val=value))
xml_lines.extend([
'</instance_attributes>',
'</node>'
])
xml_lines.append('</nodes>')
utils_usefile_original = utils.usefile
utils_filename_original = utils.filename
utils.usefile = True
utils.filename = temp_cib
output, retval = utils.run([
"cibadmin", "--modify", '--xml-text', "\n".join(xml_lines)
])
utils.usefile = utils_usefile_original
utils.filename = utils_filename_original
assert output == ""
assert retval == 0
def stop_cluster(argv):
if len(argv) > 0:
stop_cluster_nodes(argv)
return
if "--force" not in utils.pcs_options:
output, dummy_retval = utils.run(["corosync-quorumtool", "-p", "-s"])
# retval is 0 on success if node is not in partition with quorum
# retval is 1 on error OR on success if node has quorum
try:
if (corosync_live.QuorumStatus.from_string(
output).stopping_local_node_cause_quorum_loss()):
utils.err("Stopping the node will cause a loss of the quorum" +
", use --force to override")
except corosync_live.QuorumStatusException:
if not utils.is_node_offline_by_quorumtool_output(output):
utils.err(
"Unable to determine whether stopping the node will cause "
+ "a loss of the quorum, use --force to override")
# else the node seems to be stopped already, proceed to be sure
stop_all = ("--pacemaker" not in utils.pcs_options
and "--corosync" not in utils.pcs_options)
if stop_all or "--pacemaker" in utils.pcs_options:
stop_cluster_pacemaker()
if stop_all or "--corosync" in utils.pcs_options:
stop_cluster_corosync()
def pcs(cib_file, args, corosync_conf_opt=None, mock_settings=None):
"""
Run pcs with -f on specified file
Return tuple with:
shell stdoutdata
shell returncode
"""
if mock_settings is None:
mock_settings = {}
env_mock_settings_prefix = "PCS.SETTINGS."
env = {
"{}{}".format(env_mock_settings_prefix, option): value
for option, value in mock_settings.items()
}
if test_installed:
env["PCS_TEST.TEST_INSTALLED"] = "1"
cmd = [__pcs_location] + args
if cib_file:
cmd.extend(["-f", cib_file])
if corosync_conf_opt:
cmd.extend(["--corosync_conf", corosync_conf_opt])
return utils.run(cmd, env_extend=env)
def full_status():
if "--hide-inactive" in utils.pcs_options and "--full" in utils.pcs_options:
utils.err("you cannot specify both --hide-inactive and --full")
monitor_command = ["crm_mon", "--one-shot"]
if "--hide-inactive" not in utils.pcs_options:
monitor_command.append('--inactive')
if "--full" in utils.pcs_options:
monitor_command.extend(
["--show-detail", "--show-node-attributes", "--failcounts"])
output, retval = utils.run(monitor_command)
if (retval != 0):
utils.err("cluster is not currently running on this node")
if not utils.usefile or "--corosync_conf" in utils.pcs_options:
cluster_name = utils.getClusterName()
print("Cluster name: %s" % cluster_name)
status_stonith_check()
if (not utils.usefile and not utils.is_rhel6()
and utils.corosyncPacemakerNodeCheck()):
print(
"WARNING: corosync and pacemaker node names do not match (IPs used in setup?)"
)
print(output)
if not utils.usefile:
if "--full" in utils.pcs_options and utils.hasCorosyncConf():
print_pcsd_daemon_status()
print()
utils.serviceStatus(" ")
def is_minimum_pacemaker_version(cmajor, cminor, crev):
output, dummy_retval = utils.run(["crm_mon", "--version"])
pacemaker_version = output.split("\n")[0]
r = re.compile(r"Pacemaker (\d+)\.(\d+)\.(\d+)")
m = r.match(pacemaker_version)
major = int(m.group(1))
minor = int(m.group(2))
rev = int(m.group(3))
return compare_version((major, minor, rev), (cmajor, cminor, crev)) > -1
def is_minimum_pacemaker_features(cmajor, cminor, crev):
output, dummy_retval = utils.run(["pacemakerd", "--features"])
features_version = output.split("\n")[1]
r = re.compile(r"Supporting v(\d+)\.(\d+)\.(\d+):")
m = r.search(features_version)
major = int(m.group(1))
minor = int(m.group(2))
rev = int(m.group(3))
return compare_version((major, minor, rev), (cmajor, cminor, crev)) > -1
def is_minimum_pacemaker_version(cmajor, cminor, crev):
output, dummy_retval = utils.run(["crm_mon", "--version"])
pacemaker_version = output.split("\n")[0]
r = re.compile(r"Pacemaker (\d+)\.(\d+)\.(\d+)")
m = r.match(pacemaker_version)
major = int(m.group(1))
minor = int(m.group(2))
rev = int(m.group(3))
return (
major > cmajor or (major == cmajor and minor > cminor) or (major == cmajor and minor == cminor and rev >= crev)
)
def cluster_report(argv):
if len(argv) != 1:
usage.cluster(["report"])
sys.exit(1)
outfile = argv[0]
dest_outfile = outfile + ".tar.bz2"
if os.path.exists(dest_outfile):
if "--force" not in utils.pcs_options:
utils.err(dest_outfile +
" already exists, use --force to overwrite")
else:
try:
os.remove(dest_outfile)
except OSError as e:
utils.err("Unable to remove " + dest_outfile + ": " +
e.strerror)
crm_report_opts = []
crm_report_opts.append("-f")
if "--from" in utils.pcs_options:
crm_report_opts.append(utils.pcs_options["--from"])
if "--to" in utils.pcs_options:
crm_report_opts.append("-t")
crm_report_opts.append(utils.pcs_options["--to"])
else:
yesterday = datetime.datetime.now() - datetime.timedelta(1)
crm_report_opts.append(yesterday.strftime("%Y-%m-%d %H:%M"))
crm_report_opts.append(outfile)
output, retval = utils.run([settings.crm_report] + crm_report_opts)
if (retval != 0 and
"ERROR: Cannot determine nodes; specify --nodes or --single-node"
in output):
utils.err("cluster is not configured on this node")
newoutput = ""
for line in output.split("\n"):
if line.startswith("cat:") or line.startswith(
"grep") or line.startswith("grep") or line.startswith("tail"):
continue
if "We will attempt to remove" in line:
continue
if "-p option" in line:
continue
if "However, doing" in line:
continue
if "to diagnose" in line:
continue
if "--dest" in line:
line = line.replace("--dest", "<dest>")
newoutput = newoutput + line + "\n"
if retval != 0:
utils.err(newoutput)
print(newoutput)
def testNodeProperties(self):
utils.usefile = True
utils.filename = temp_cib
o,r = utils.run(["cibadmin","-M", '--xml-text', '<nodes><node id="1" uname="rh7-1"><instance_attributes id="nodes-1"/></node><node id="2" uname="rh7-2"><instance_attributes id="nodes-2"/></node></nodes>'])
ac(o,"")
assert r == 0
o,r = pcs("property set --node=rh7-1 IP=192.168.1.1")
ac(o,"")
assert r==0
o,r = pcs("property set --node=rh7-2 IP=192.168.2.2")
ac(o,"")
assert r==0
o,r = pcs("property")
ac(o,"Cluster Properties:\nNode Attributes:\n rh7-1: IP=192.168.1.1\n rh7-2: IP=192.168.2.2\n")
assert r==0
o,r = pcs("property set --node=rh7-2 IP=")
ac(o,"")
assert r==0
o,r = pcs("property")
ac(o,"Cluster Properties:\nNode Attributes:\n rh7-1: IP=192.168.1.1\n")
assert r==0
o,r = pcs("property set --node=rh7-1 IP=192.168.1.1")
ac(o,"")
assert r==0
o,r = pcs("property set --node=rh7-2 IP=192.168.2.2")
ac(o,"")
assert r==0
o,r = pcs("property")
ac(o,"Cluster Properties:\nNode Attributes:\n rh7-1: IP=192.168.1.1\n rh7-2: IP=192.168.2.2\n")
assert r==0
o,r = pcs("property unset --node=rh7-1 IP")
ac(o,"")
assert r==0
o,r = pcs("property")
ac(o,"Cluster Properties:\nNode Attributes:\n rh7-2: IP=192.168.2.2\n")
assert r==0
o,r = pcs("property unset --node=rh7-1 IP")
ac(o,"Error: attribute: 'IP' doesn't exist for node: 'rh7-1'\n")
assert r==2
o,r = pcs("property unset --node=rh7-1 IP --force")
ac(o,"")
assert r==0
def corosync_status(dummy_lib, argv, modifiers):
"""
Options: no options
"""
modifiers.ensure_only_supported()
if argv:
raise CmdLineInputError()
(output, retval) = utils.run(["corosync-quorumtool", "-l"])
if retval != 0:
utils.err("corosync not running")
else:
print(output.rstrip())
def stonith_confirm(argv):
if len(argv) != 1:
utils.err("must specify one (and only one) node to confirm fenced")
node = argv.pop(0)
args = ["stonith_admin", "-C", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to confirm fencing of node '%s'\n" % node + output)
else:
print("Node: %s confirmed fenced" % node)
def full_status():
if "--hide-inactive" in utils.pcs_options and "--full" in utils.pcs_options:
utils.err("you cannot specify both --hide-inactive and --full")
monitor_command = ["crm_mon", "--one-shot"]
if "--hide-inactive" not in utils.pcs_options:
monitor_command.append('--inactive')
if "--full" in utils.pcs_options:
monitor_command.extend(
["--show-detail", "--show-node-attributes", "--failcounts"])
output, retval = utils.run(monitor_command)
if (retval != 0):
utils.err("cluster is not currently running on this node")
if not utils.usefile or "--corosync_conf" in utils.pcs_options:
cluster_name = utils.getClusterName()
print("Cluster name: %s" % cluster_name)
status_stonith_check()
print(output)
if "--full" in utils.pcs_options:
tickets, retval = utils.run(["crm_ticket", "-L"])
if retval != 0:
print("WARNING: Unable to get information about tickets")
print()
elif tickets:
print("Tickets:")
print("\n".join(indent(tickets.split("\n"))))
if not utils.usefile:
if "--full" in utils.pcs_options and utils.hasCorosyncConf():
print_pcsd_daemon_status()
print()
utils.serviceStatus(" ")
def node_maintenance(argv, on=True):
action = ["-v", "on"] if on else ["-D"]
cluster_nodes = utils.getNodesFromPacemaker()
nodes = []
failed_count = 0
if "--all" in utils.pcs_options:
nodes = cluster_nodes
elif argv:
for node in argv:
if node not in cluster_nodes:
utils.err(
"Node '{0}' does not appear to exist in "
"configuration".format(node),
False
)
failed_count += 1
else:
nodes.append(node)
else:
nodes.append("")
if failed_count > 0:
sys.exit(1)
for node in nodes:
node_attr = ["-N", node] if node else []
output, retval = utils.run(
["crm_attribute", "-t", "nodes", "-n", "maintenance"] + action +
node_attr
)
if retval != 0:
node_name = ("node '{0}'".format(node)) if argv else "current node"
failed_count += 1
if on:
utils.err(
"Unable to put {0} to maintenance mode: {1}".format(
node_name, output
),
False
)
else:
utils.err(
"Unable to remove {0} from maintenance mode: {1}".format(
node_name, output
),
False
)
if failed_count > 0:
sys.exit(1)
def xml_status(dummy_lib, argv, modifiers):
"""
Options:
* -f - CIB file
"""
modifiers.ensure_only_supported("-f")
if argv:
raise CmdLineInputError()
(output, retval) = utils.run(["crm_mon", "-1", "-r", "-X"],
ignore_stderr=True)
if (retval != 0):
utils.err("running crm_mon, is pacemaker running?")
print(output.rstrip())
def stonith_fence(argv):
if len(argv) != 1:
utils.err("must specify one (and only one) node to fence")
node = argv.pop(0)
if "--off" in utils.pcs_options:
args = ["stonith_admin", "-F", node]
else:
args = ["stonith_admin", "-B", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to fence '%s'\n" % node + output)
else:
print("Node: %s fenced" % node)
def stonith_fence(argv):
if len(argv) != 1:
utils.err("must specify one (and only one) node to fence")
node = argv.pop(0)
if "--off" in utils.pcs_options:
args = ["stonith_admin", "-F", node]
else:
args = ["stonith_admin", "-B", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to fence '%s'\n" % node + output)
else:
print("Node: %s fenced" % node)
def kill_local_cluster_services():
all_cluster_daemons = [
# Daemons taken from cluster-clean script in pacemaker
"pacemaker-attrd",
"pacemaker-based",
"pacemaker-controld",
"pacemaker-execd",
"pacemaker-fenced",
"pacemaker-remoted",
"pacemaker-schedulerd",
"pacemakerd",
"dlm_controld",
"gfs_controld",
# Corosync daemons
"corosync-qdevice",
"corosync",
]
return utils.run(["killall", "-9"] + all_cluster_daemons)
def cluster_status(argv):
(output, retval) = utils.run(["crm_mon", "-1", "-r"])
if (retval != 0):
utils.err("cluster is not currently running on this node")
first_empty_line = False
print("Cluster Status:")
for line in output.splitlines():
if line == "":
if first_empty_line:
break
first_empty_line = True
continue
else:
print("",line)
if not utils.usefile and utils.hasCorosyncConf():
print()
print_pcsd_daemon_status()
def stonith_fence(dummy_lib, argv, modifiers):
"""
Options:
* --off - use off action of fence agent
"""
modifiers.ensure_only_supported("--off")
if len(argv) != 1:
utils.err("must specify one (and only one) node to fence")
node = argv.pop(0)
if modifiers.get("--off"):
args = ["stonith_admin", "-F", node]
else:
args = ["stonith_admin", "-B", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to fence '%s'\n" % node + output)
else:
print("Node: %s fenced" % node)
def cluster_status(argv):
(output, retval) = utils.run(["crm_mon", "-1", "-r"])
if (retval != 0):
utils.err("cluster is not currently running on this node")
first_empty_line = False
print("Cluster Status:")
for line in output.splitlines():
if line == "":
if first_empty_line:
break
first_empty_line = True
continue
else:
print("", line)
if not utils.usefile and utils.hasCorosyncConf():
print()
print_pcsd_daemon_status()
def node_maintenance(argv, on=True):
action = ["-v", "on"] if on else ["-D"]
cluster_nodes = utils.getNodesFromPacemaker()
nodes = []
failed_count = 0
if "--all" in utils.pcs_options:
nodes = cluster_nodes
elif argv:
for node in argv:
if node not in cluster_nodes:
utils.err(
"Node '{0}' does not appear to exist in "
"configuration".format(node), False)
failed_count += 1
else:
nodes.append(node)
else:
nodes.append("")
if failed_count > 0:
sys.exit(1)
for node in nodes:
node_attr = ["-N", node] if node else []
output, retval = utils.run(
["crm_attribute", "-t", "nodes", "-n", "maintenance"] + action +
node_attr)
if retval != 0:
node_name = ("node '{0}'".format(node)) if argv else "current node"
failed_count += 1
if on:
utils.err(
"Unable to put {0} to maintenance mode: {1}".format(
node_name, output), False)
else:
utils.err(
"Unable to remove {0} from maintenance mode: {1}".format(
node_name, output), False)
if failed_count > 0:
sys.exit(1)
def stonith_fence(lib, argv, modifiers):
"""
Options:
* --off - use off action of fence agent
"""
del lib
modifiers.ensure_only_supported("--off")
if len(argv) != 1:
utils.err("must specify one (and only one) node to fence")
node = argv.pop(0)
if modifiers.get("--off"):
args = ["stonith_admin", "-F", node]
else:
args = ["stonith_admin", "-B", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to fence '%s'\n" % node + output)
else:
print("Node: %s fenced" % node)
def stonith_confirm(argv, skip_question=False):
if len(argv) != 1:
utils.err("must specify one (and only one) node to confirm fenced")
node = argv.pop(0)
if not skip_question and "--force" not in utils.pcs_options:
answer = utils.get_terminal_input(
("WARNING: If node {node} is not powered off or it does" +
" have access to shared resources, data corruption and/or" +
" cluster failure may occur. Are you sure you want to" +
" continue? [y/N] ").format(node=node))
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
args = ["stonith_admin", "-C", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to confirm fencing of node '%s'\n" % node + output)
else:
print("Node: %s confirmed fenced" % node)
def pcs(
cib_file, args, corosync_conf_opt=None, mock_settings=None
):
"""
Run pcs with -f on specified file
Return tuple with:
shell stdoutdata
shell returncode
"""
if mock_settings is None:
mock_settings = {}
arg_split = args.split()
arg_split_temp = []
in_quote = False
for arg in arg_split:
if in_quote:
arg_split_temp[-1] = arg_split_temp[-1] + " " + arg.replace("'", "")
if arg.find("'") != -1:
in_quote = False
else:
arg_split_temp.append(arg.replace("'", ""))
if arg.find("'") != -1 and not (arg[0] == "'" and arg[-1] == "'"):
in_quote = True
env_mock_settings_prefix = "PCS.SETTINGS."
env = {
"{}{}".format(env_mock_settings_prefix, option): value
for option, value in mock_settings.items()
}
if test_installed:
env["PCS_TEST.TEST_INSTALLED"] = "1"
cmd = [__pcs_location] + arg_split_temp
if cib_file:
cmd.extend(["-f", cib_file])
if corosync_conf_opt:
cmd.extend(["--corosync_conf", corosync_conf_opt])
return utils.run(cmd, env_extend=env)
def stonith_confirm(lib, argv, modifiers):
"""
Options:
* --force - do not warn user
"""
del lib
modifiers.ensure_only_supported("--force")
if len(argv) != 1:
utils.err("must specify one (and only one) node to confirm fenced")
node = argv.pop(0)
if not utils.get_continue_confirmation_or_force(
f"If node '{node}' is not powered off or it does have access to shared "
"resources, data corruption and/or cluster failure may occur",
modifiers.get("--force"),
):
return
args = ["stonith_admin", "-C", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to confirm fencing of node '%s'\n" % node + output)
else:
print("Node: %s confirmed fenced" % node)
def stonith_confirm(argv, skip_question=False):
if len(argv) != 1:
utils.err("must specify one (and only one) node to confirm fenced")
node = argv.pop(0)
if not skip_question and "--force" not in utils.pcs_options:
answer = utils.get_terminal_input(
(
"WARNING: If node {node} is not powered off or it does"
+ " have access to shared resources, data corruption and/or"
+ " cluster failure may occur. Are you sure you want to"
+ " continue? [y/N] "
).format(node=node)
)
if answer.lower() not in ["y", "yes"]:
print("Canceled")
return
args = ["stonith_admin", "-C", node]
output, retval = utils.run(args)
if retval != 0:
utils.err("unable to confirm fencing of node '%s'\n" % node + output)
else:
print("Node: %s confirmed fenced" % node)
def skip_unless_pacemaker_supports_systemd():
output, dummy_retval = utils.run(["pacemakerd", "--features"])
return skipUnless(
"systemd" in output,
"Pacemaker does not support systemd resources"
)
def skip_unless_pacemaker_supports_systemd():
output, dummy_retval = utils.run(["pacemakerd", "--features"])
return skipUnless("systemd" in output,
"Pacemaker does not support systemd resources")
def is_corosyc_running():
if utils.is_systemctl():
dummy_output, retval = utils.run(["systemctl", "status", "corosync.service"])
else:
dummy_output, retval = utils.run(["service", "corosync", "status"])
return retval == 0
def is_pacemaker_running():
if utils.is_systemctl():
dummy_output, retval = utils.run(["systemctl", "status", "pacemaker.service"])
else:
dummy_output, retval = utils.run(["service", "pacemaker", "status"])
return retval == 0
def corosync_status():
(output, retval) = utils.run(["corosync-quorumtool", "-l"])
if retval != 0:
utils.err("corosync not running")
else:
print(output, end="")
def xml_status():
(output, retval) = utils.run(["crm_mon", "-1", "-r", "-X"])
if (retval != 0):
utils.err("running crm_mon, is pacemaker running?")
print(output, end="")
def is_service_running(service):
if utils.is_systemctl():
dummy_output, retval = utils.run(["systemctl", "status", service])
else:
dummy_output, retval = utils.run(["service", service, "status"])
return retval == 0
def corosync_status():
(output, retval) = utils.run(["corosync-quorumtool", "-l"])
if retval != 0:
utils.err("corosync not running")
else:
print(output.rstrip())
def xml_status():
(output, retval) = utils.run(["crm_mon", "-1", "-r", "-X"])
if (retval != 0):
utils.err("running crm_mon, is pacemaker running?")
print(output.rstrip())
def is_service_running(service):
if utils.is_systemctl():
dummy_output, retval = utils.run(["systemctl", "status", service])
else:
dummy_output, retval = utils.run(["service", service, "status"])
return retval == 0
def full_status(lib, argv, modifiers):
"""
Options:
* --hide-inactive - hide inactive resources
* --full - show full details, node attributes and failcount
* -f - CIB file, crm_mon accepts CIB_file environment variable
* --corosync_conf - file corocync.conf
* --request-timeout - HTTP timeout for node authorization check
"""
modifiers.ensure_only_supported(
"--hide-inactive",
"--full",
"-f",
"--corosync_conf",
"--request-timeout",
)
if argv:
raise CmdLineInputError()
if (modifiers.is_specified("--hide-inactive")
and modifiers.is_specified("--full")):
utils.err("you cannot specify both --hide-inactive and --full")
monitor_command = [
os.path.join(settings.pacemaker_binaries, "crm_mon"), "--one-shot"
]
if not modifiers.get("--hide-inactive"):
monitor_command.append('--inactive')
if modifiers.get("--full"):
monitor_command.extend(
["--show-detail", "--show-node-attributes", "--failcounts"])
# by default, pending and failed actions are displayed
# with --full, we display the whole history
if is_fence_history_supported():
monitor_command.append("--fence-history=3")
stdout, stderr, retval = utils.cmd_runner().run(monitor_command)
if retval != 0:
utils.err("cluster is not currently running on this node")
warnings = []
if stderr.strip():
for line in stderr.strip().splitlines():
if line.startswith("DEBUG: "):
if modifiers.get("--full"):
warnings.append(line)
else:
warnings.append(line)
warnings.extend(status_stonith_check(modifiers))
print("Cluster name: %s" % utils.getClusterName())
if warnings:
print()
print("WARNINGS:")
print("\n".join(warnings))
print()
print(stdout)
if modifiers.get("--full"):
tickets, retval = utils.run(["crm_ticket", "-L"])
if retval != 0:
print("WARNING: Unable to get information about tickets")
print()
elif tickets:
print("Tickets:")
print("\n".join(indent(tickets.split("\n"))))
if not (modifiers.is_specified("-f")
or modifiers.is_specified("--corosync_conf")):
# do this only if in live environment
if modifiers.get("--full"):
print_pcsd_daemon_status(lib, modifiers)
print()
utils.serviceStatus(" ")
