#
#
-# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Google Inc.
+# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012 Google Inc.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
"""Node related commands"""
-# pylint: disable-msg=W0401,W0613,W0614,C0103
+# pylint: disable=W0401,W0613,W0614,C0103
# W0401: Wildcard import ganeti.cli
# W0613: Unused argument, since all functions follow the same API
# W0614: Unused import %s from wildcard import (since we need cli)
# C0103: Invalid name gnt-node
+import itertools
+
from ganeti.cli import *
from ganeti import cli
from ganeti import bootstrap
from ganeti import constants
from ganeti import errors
from ganeti import netutils
+from ganeti import pathutils
from cStringIO import StringIO
+from ganeti import confd
+from ganeti.confd import client as confd_client
#: default list of field for L{ListNodes}
_LIST_DEF_FIELDS = [
constants.OOB_POWER_CYCLE])
+_ENV_OVERRIDE = frozenset(["list"])
+
+
NONODE_SETUP_OPT = cli_option("--no-node-setup", default=True,
action="store_false", dest="node_setup",
help=("Do not make initial SSH setup on remote"
help=("Ignore the Node(s) offline status"
" (potentially DANGEROUS)"))
-OOB_TIMEOUT_OPT = cli_option("--oob-timeout", dest="oob_timeout", type="int",
- default=constants.OOB_TIMEOUT,
- help="Maximum time to wait for out-of-band helper")
def ConvertStorageType(user_storage_type):
"""Converts a user storage type to its internal name.
@param nodes: The nodes to setup
"""
- cmd = [constants.SETUP_SSH]
+
+ assert nodes, "Empty node list"
+
+ cmd = [pathutils.SETUP_SSH]
# Pass --debug|--verbose to the external script if set on our invocation
# --debug overrides --verbose
readd = opts.readd
try:
- output = cl.QueryNodes(names=[node], fields=['name', 'sip'],
+ output = cl.QueryNodes(names=[node], fields=["name", "sip", "master"],
use_locking=False)
- node_exists, sip = output[0]
+ node_exists, sip, is_master = output[0]
except (errors.OpPrereqError, errors.OpExecError):
node_exists = ""
sip = None
ToStderr("Node %s not in the cluster"
" - please retry without '--readd'", node)
return 1
+ if is_master:
+ ToStderr("Node %s is the master, cannot readd", node)
+ return 1
else:
if node_exists:
ToStderr("Node %s already in the cluster (as %s)"
sip = opts.secondary_ip
# read the cluster name from the master
- output = cl.QueryConfigValues(['cluster_name'])
+ output = cl.QueryConfigValues(["cluster_name"])
cluster_name = output[0]
if not readd and opts.node_setup:
bootstrap.SetupNodeDaemon(cluster_name, node, opts.ssh_key_check)
+ if opts.disk_state:
+ disk_state = utils.FlatToDict(opts.disk_state)
+ else:
+ disk_state = {}
+
+ hv_state = dict(opts.hv_state)
+
op = opcodes.OpNodeAdd(node_name=args[0], secondary_ip=sip,
readd=opts.readd, group=opts.nodegroup,
vm_capable=opts.vm_capable, ndparams=opts.ndparams,
- master_capable=opts.master_capable)
+ master_capable=opts.master_capable,
+ disk_state=disk_state,
+ hv_state=hv_state)
SubmitOpCode(op, opts=opts)
return GenericList(constants.QR_NODE, selected_fields, args, opts.units,
opts.separator, not opts.no_headers,
- format_override=fmtoverride, verbose=opts.verbose)
+ format_override=fmtoverride, verbose=opts.verbose,
+ force_filter=opts.force_filter)
def ListNodeFields(opts, args):
@return: the desired exit code
"""
- cl = GetClient()
- force = opts.force
+ if opts.dst_node is not None:
+ ToStderr("New secondary node given (disabling iallocator), hence evacuating"
+ " secondary instances only.")
+ opts.secondary_only = True
+ opts.primary_only = False
+
+ if opts.secondary_only and opts.primary_only:
+ raise errors.OpPrereqError("Only one of the --primary-only and"
+ " --secondary-only options can be passed",
+ errors.ECODE_INVAL)
+ elif opts.primary_only:
+ mode = constants.NODE_EVAC_PRI
+ elif opts.secondary_only:
+ mode = constants.NODE_EVAC_SEC
+ else:
+ mode = constants.NODE_EVAC_ALL
+
+ # Determine affected instances
+ fields = []
- dst_node = opts.dst_node
- iallocator = opts.iallocator
+ if not opts.secondary_only:
+ fields.append("pinst_list")
+ if not opts.primary_only:
+ fields.append("sinst_list")
- op = opcodes.OpNodeEvacStrategy(nodes=args,
- iallocator=iallocator,
- remote_node=dst_node)
+ cl = GetClient()
+
+ result = cl.QueryNodes(names=args, fields=fields, use_locking=False)
+ instances = set(itertools.chain(*itertools.chain(*itertools.chain(result))))
- result = SubmitOpCode(op, cl=cl, opts=opts)
- if not result:
- # no instances to migrate
- ToStderr("No secondary instances on node(s) %s, exiting.",
+ if not instances:
+ # No instances to evacuate
+ ToStderr("No instances to evacuate on node(s) %s, exiting.",
utils.CommaJoin(args))
return constants.EXIT_SUCCESS
- if not force and not AskUser("Relocate instance(s) %s from node(s) %s?" %
- (",".join("'%s'" % name[0] for name in result),
- utils.CommaJoin(args))):
+ if not (opts.force or
+ AskUser("Relocate instance(s) %s from node(s) %s?" %
+ (utils.CommaJoin(utils.NiceSort(instances)),
+ utils.CommaJoin(args)))):
return constants.EXIT_CONFIRMATION
+ # Evacuate node
+ op = opcodes.OpNodeEvacuate(node_name=args[0], mode=mode,
+ remote_node=opts.dst_node,
+ iallocator=opts.iallocator,
+ early_release=opts.early_release)
+ result = SubmitOrSend(op, opts, cl=cl)
+
+ # Keep track of submitted jobs
jex = JobExecutor(cl=cl, opts=opts)
- for row in result:
- iname = row[0]
- node = row[1]
- ToStdout("Will relocate instance %s to node %s", iname, node)
- op = opcodes.OpInstanceReplaceDisks(instance_name=iname,
- remote_node=node, disks=[],
- mode=constants.REPLACE_DISK_CHG,
- early_release=opts.early_release)
- jex.QueueJob(iname, op)
+
+ for (status, job_id) in result[constants.JOB_IDS_KEY]:
+ jex.AddJobId(None, status, job_id)
+
results = jex.GetResults()
bad_cnt = len([row for row in results if not row[0]])
if bad_cnt == 0:
- ToStdout("All %d instance(s) failed over successfully.", len(results))
+ ToStdout("All instances evacuated successfully.")
rcode = constants.EXIT_SUCCESS
else:
- ToStdout("There were errors during the failover:\n"
- "%d error(s) out of %d instance(s).", bad_cnt, len(results))
+ ToStdout("There were %s errors during the evacuation.", bad_cnt)
rcode = constants.EXIT_FAILURE
+
return rcode
jex = JobExecutor(cl=cl, opts=opts)
for iname in pinst:
op = opcodes.OpInstanceFailover(instance_name=iname,
- ignore_consistency=opts.ignore_consistency)
+ ignore_consistency=opts.ignore_consistency,
+ iallocator=opts.iallocator)
jex.QueueJob(iname, op)
results = jex.GetResults()
bad_cnt = len([row for row in results if not row[0]])
selected_fields = ["name", "pinst_list"]
result = cl.QueryNodes(names=args, fields=selected_fields, use_locking=False)
- node, pinst = result[0]
+ ((node, pinst), ) = result
if not pinst:
ToStdout("No primary instances on node %s, exiting." % node)
pinst = utils.NiceSort(pinst)
- if not force and not AskUser("Migrate instance(s) %s?" %
- (",".join("'%s'" % name for name in pinst))):
- return 2
+ if not (force or
+ AskUser("Migrate instance(s) %s?" %
+ utils.CommaJoin(utils.NiceSort(pinst)))):
+ return constants.EXIT_CONFIRMATION
# this should be removed once --non-live is deprecated
if not opts.live and opts.migration_mode is not None:
mode = constants.HT_MIGRATION_NONLIVE
else:
mode = opts.migration_mode
- op = opcodes.OpNodeMigrate(node_name=args[0], mode=mode)
- SubmitOpCode(op, cl=cl, opts=opts)
+
+ op = opcodes.OpNodeMigrate(node_name=args[0], mode=mode,
+ iallocator=opts.iallocator,
+ target_node=opts.dst_node,
+ allow_runtime_changes=opts.allow_runtime_chgs,
+ ignore_ipolicy=opts.ignore_ipolicy)
+
+ result = SubmitOrSend(op, opts, cl=cl)
+
+ # Keep track of submitted jobs
+ jex = JobExecutor(cl=cl, opts=opts)
+
+ for (status, job_id) in result[constants.JOB_IDS_KEY]:
+ jex.AddJobId(None, status, job_id)
+
+ results = jex.GetResults()
+ bad_cnt = len([row for row in results if not row[0]])
+ if bad_cnt == 0:
+ ToStdout("All instances migrated successfully.")
+ rcode = constants.EXIT_SUCCESS
+ else:
+ ToStdout("There were %s errors during the node migration.", bad_cnt)
+ rcode = constants.EXIT_FAILURE
+
+ return rcode
def ShowNodeConfig(opts, args):
return 2
op = opcodes.OpNodePowercycle(node_name=node, force=opts.force)
- result = SubmitOpCode(op, opts=opts)
+ result = SubmitOrSend(op, opts)
if result:
ToStderr(result)
return 0
opcodelist.append(opcodes.OpOobCommand(node_names=args,
command=oob_command,
ignore_status=opts.ignore_status,
- timeout=opts.oob_timeout))
+ timeout=opts.oob_timeout,
+ power_delay=opts.power_delay))
cli.SetGenericOpcodeOpts(opcodelist, opts)
storage_type=storage_type,
name=volume_name,
changes=changes)
- SubmitOpCode(op, opts=opts)
+ SubmitOrSend(op, opts)
else:
ToStderr("No changes to perform, exiting.")
storage_type=storage_type,
name=volume_name,
ignore_consistency=opts.ignore_consistency)
- SubmitOpCode(op, opts=opts)
+ SubmitOrSend(op, opts)
def SetNodeParams(opts, args):
all_changes = [opts.master_candidate, opts.drained, opts.offline,
opts.master_capable, opts.vm_capable, opts.secondary_ip,
opts.ndparams]
- if all_changes.count(None) == len(all_changes):
+ if (all_changes.count(None) == len(all_changes) and
+ not (opts.hv_state or opts.disk_state)):
ToStderr("Please give at least one of the parameters.")
return 1
+ if opts.disk_state:
+ disk_state = utils.FlatToDict(opts.disk_state)
+ else:
+ disk_state = {}
+
+ hv_state = dict(opts.hv_state)
+
op = opcodes.OpNodeSetParams(node_name=args[0],
master_candidate=opts.master_candidate,
offline=opts.offline,
force=opts.force,
ndparams=opts.ndparams,
auto_promote=opts.auto_promote,
- powered=opts.node_powered)
+ powered=opts.node_powered,
+ hv_state=hv_state,
+ disk_state=disk_state)
# even if here we process the result, we allow submit only
result = SubmitOrSend(op, opts)
return 0
+class ReplyStatus(object):
+ """Class holding a reply status for synchronous confd clients.
+
+ """
+ def __init__(self):
+ self.failure = True
+ self.answer = False
+
+
+def ListDrbd(opts, args):
+ """Modifies a node.
+
+ @param opts: the command line options selected by the user
+ @type args: list
+ @param args: should contain only one element, the node name
+ @rtype: int
+ @return: the desired exit code
+
+ """
+ if len(args) != 1:
+ ToStderr("Please give one (and only one) node.")
+ return constants.EXIT_FAILURE
+
+ if not constants.ENABLE_CONFD:
+ ToStderr("Error: this command requires confd support, but it has not"
+ " been enabled at build time.")
+ return constants.EXIT_FAILURE
+
+ if not constants.HS_CONFD:
+ ToStderr("Error: this command requires the Haskell version of confd,"
+ " but it has not been enabled at build time.")
+ return constants.EXIT_FAILURE
+
+ status = ReplyStatus()
+
+ def ListDrbdConfdCallback(reply):
+ """Callback for confd queries"""
+ if reply.type == confd_client.UPCALL_REPLY:
+ answer = reply.server_reply.answer
+ reqtype = reply.orig_request.type
+ if reqtype == constants.CONFD_REQ_NODE_DRBD:
+ if reply.server_reply.status != constants.CONFD_REPL_STATUS_OK:
+ ToStderr("Query gave non-ok status '%s': %s" %
+ (reply.server_reply.status,
+ reply.server_reply.answer))
+ status.failure = True
+ return
+ if not confd.HTNodeDrbd(answer):
+ ToStderr("Invalid response from server: expected %s, got %s",
+ confd.HTNodeDrbd, answer)
+ status.failure = True
+ else:
+ status.failure = False
+ status.answer = answer
+ else:
+ ToStderr("Unexpected reply %s!?", reqtype)
+ status.failure = True
+
+ node = args[0]
+ hmac = utils.ReadFile(pathutils.CONFD_HMAC_KEY)
+ filter_callback = confd_client.ConfdFilterCallback(ListDrbdConfdCallback)
+ counting_callback = confd_client.ConfdCountingCallback(filter_callback)
+ cf_client = confd_client.ConfdClient(hmac, [constants.IP4_ADDRESS_LOCALHOST],
+ counting_callback)
+ req = confd_client.ConfdClientRequest(type=constants.CONFD_REQ_NODE_DRBD,
+ query=node)
+
+ def DoConfdRequestReply(req):
+ counting_callback.RegisterQuery(req.rsalt)
+ cf_client.SendRequest(req, async=False)
+ while not counting_callback.AllAnswered():
+ if not cf_client.ReceiveReply():
+ ToStderr("Did not receive all expected confd replies")
+ break
+
+ DoConfdRequestReply(req)
+
+ if status.failure:
+ return constants.EXIT_FAILURE
+
+ fields = ["node", "minor", "instance", "disk", "role", "peer"]
+ if opts.no_headers:
+ headers = None
+ else:
+ headers = {"node": "Node", "minor": "Minor", "instance": "Instance",
+ "disk": "Disk", "role": "Role", "peer": "PeerNode"}
+
+ data = GenerateTable(separator=opts.separator, headers=headers,
+ fields=fields, data=sorted(status.answer),
+ numfields=["minor"])
+ for line in data:
+ ToStdout(line)
+
+ return constants.EXIT_SUCCESS
+
commands = {
- 'add': (
+ "add": (
AddNode, [ArgHost(min=1, max=1)],
[SECONDARY_IP_OPT, READD_OPT, NOSSH_KEYCHECK_OPT, NODE_FORCE_JOIN_OPT,
NONODE_SETUP_OPT, VERBOSE_OPT, NODEGROUP_OPT, PRIORITY_OPT,
- CAPAB_MASTER_OPT, CAPAB_VM_OPT, NODE_PARAMS_OPT],
+ CAPAB_MASTER_OPT, CAPAB_VM_OPT, NODE_PARAMS_OPT, HV_STATE_OPT,
+ DISK_STATE_OPT],
"[-s ip] [--readd] [--no-ssh-key-check] [--force-join]"
" [--no-node-setup] [--verbose]"
" <node_name>",
"Add a node to the cluster"),
- 'evacuate': (
- EvacuateNode, [ArgNode(min=1)],
+ "evacuate": (
+ EvacuateNode, ARGS_ONE_NODE,
[FORCE_OPT, IALLOCATOR_OPT, NEW_SECONDARY_OPT, EARLY_RELEASE_OPT,
- PRIORITY_OPT],
- "[-f] {-I <iallocator> | -n <dst>} <node>",
- "Relocate the secondary instances from a node"
- " to other nodes (only for instances with drbd disk template)"),
- 'failover': (
- FailoverNode, ARGS_ONE_NODE, [FORCE_OPT, IGNORE_CONSIST_OPT, PRIORITY_OPT],
+ PRIORITY_OPT, PRIMARY_ONLY_OPT, SECONDARY_ONLY_OPT, SUBMIT_OPT],
+ "[-f] {-I <iallocator> | -n <dst>} [-p | -s] [options...] <node>",
+ "Relocate the primary and/or secondary instances from a node"),
+ "failover": (
+ FailoverNode, ARGS_ONE_NODE, [FORCE_OPT, IGNORE_CONSIST_OPT,
+ IALLOCATOR_OPT, PRIORITY_OPT],
"[-f] <node>",
"Stops the primary instances on a node and start them on their"
" secondary node (only for instances with drbd disk template)"),
- 'migrate': (
+ "migrate": (
MigrateNode, ARGS_ONE_NODE,
- [FORCE_OPT, NONLIVE_OPT, MIGRATION_MODE_OPT, PRIORITY_OPT],
+ [FORCE_OPT, NONLIVE_OPT, MIGRATION_MODE_OPT, DST_NODE_OPT,
+ IALLOCATOR_OPT, PRIORITY_OPT, IGNORE_IPOLICY_OPT,
+ NORUNTIME_CHGS_OPT, SUBMIT_OPT],
"[-f] <node>",
"Migrate all the primary instance on a node away from it"
" (only for instances of type drbd)"),
- 'info': (
+ "info": (
ShowNodeConfig, ARGS_MANY_NODES, [],
"[<node_name>...]", "Show information about the node(s)"),
- 'list': (
+ "list": (
ListNodes, ARGS_MANY_NODES,
- [NOHDR_OPT, SEP_OPT, USEUNITS_OPT, FIELDS_OPT, VERBOSE_OPT],
+ [NOHDR_OPT, SEP_OPT, USEUNITS_OPT, FIELDS_OPT, VERBOSE_OPT,
+ FORCE_FILTER_OPT],
"[nodes...]",
"Lists the nodes in the cluster. The available fields can be shown using"
" the \"list-fields\" command (see the man page for details)."
[NOHDR_OPT, SEP_OPT],
"[fields...]",
"Lists all available fields for nodes"),
- 'modify': (
+ "modify": (
SetNodeParams, ARGS_ONE_NODE,
[FORCE_OPT, SUBMIT_OPT, MC_OPT, DRAINED_OPT, OFFLINE_OPT,
CAPAB_MASTER_OPT, CAPAB_VM_OPT, SECONDARY_IP_OPT,
AUTO_PROMOTE_OPT, DRY_RUN_OPT, PRIORITY_OPT, NODE_PARAMS_OPT,
- NODE_POWERED_OPT],
+ NODE_POWERED_OPT, HV_STATE_OPT, DISK_STATE_OPT],
"<node_name>", "Alters the parameters of a node"),
- 'powercycle': (
+ "powercycle": (
PowercycleNode, ARGS_ONE_NODE,
- [FORCE_OPT, CONFIRM_OPT, DRY_RUN_OPT, PRIORITY_OPT],
+ [FORCE_OPT, CONFIRM_OPT, DRY_RUN_OPT, PRIORITY_OPT, SUBMIT_OPT],
"<node_name>", "Tries to forcefully powercycle a node"),
- 'power': (
+ "power": (
PowerNode,
[ArgChoice(min=1, max=1, choices=_LIST_POWER_COMMANDS),
ArgNode()],
[SUBMIT_OPT, AUTO_PROMOTE_OPT, PRIORITY_OPT, IGNORE_STATUS_OPT,
- FORCE_OPT, NOHDR_OPT, SEP_OPT, OOB_TIMEOUT_OPT],
+ FORCE_OPT, NOHDR_OPT, SEP_OPT, OOB_TIMEOUT_OPT, POWER_DELAY_OPT],
"on|off|cycle|status [nodes...]",
"Change power state of node by calling out-of-band helper."),
- 'remove': (
+ "remove": (
RemoveNode, ARGS_ONE_NODE, [DRY_RUN_OPT, PRIORITY_OPT],
"<node_name>", "Removes a node from the cluster"),
- 'volumes': (
+ "volumes": (
ListVolumes, [ArgNode()],
[NOHDR_OPT, SEP_OPT, USEUNITS_OPT, FIELDS_OPT, PRIORITY_OPT],
"[<node_name>...]", "List logical volumes on node(s)"),
- 'list-storage': (
+ "list-storage": (
ListStorage, ARGS_MANY_NODES,
[NOHDR_OPT, SEP_OPT, USEUNITS_OPT, FIELDS_OPT, _STORAGE_TYPE_OPT,
PRIORITY_OPT],
"[<node_name>...]", "List physical volumes on node(s). The available"
" fields are (see the man page for details): %s." %
(utils.CommaJoin(_LIST_STOR_HEADERS))),
- 'modify-storage': (
+ "modify-storage": (
ModifyStorage,
[ArgNode(min=1, max=1),
ArgChoice(min=1, max=1, choices=_MODIFIABLE_STORAGE_TYPES),
ArgFile(min=1, max=1)],
- [ALLOCATABLE_OPT, DRY_RUN_OPT, PRIORITY_OPT],
+ [ALLOCATABLE_OPT, DRY_RUN_OPT, PRIORITY_OPT, SUBMIT_OPT],
"<node_name> <storage_type> <name>", "Modify storage volume on a node"),
- 'repair-storage': (
+ "repair-storage": (
RepairStorage,
[ArgNode(min=1, max=1),
ArgChoice(min=1, max=1, choices=_REPAIRABLE_STORAGE_TYPES),
ArgFile(min=1, max=1)],
- [IGNORE_CONSIST_OPT, DRY_RUN_OPT, PRIORITY_OPT],
+ [IGNORE_CONSIST_OPT, DRY_RUN_OPT, PRIORITY_OPT, SUBMIT_OPT],
"<node_name> <storage_type> <name>",
"Repairs a storage volume on a node"),
- 'list-tags': (
+ "list-tags": (
ListTags, ARGS_ONE_NODE, [],
"<node_name>", "List the tags of the given node"),
- 'add-tags': (
- AddTags, [ArgNode(min=1, max=1), ArgUnknown()], [TAG_SRC_OPT, PRIORITY_OPT],
+ "add-tags": (
+ AddTags, [ArgNode(min=1, max=1), ArgUnknown()],
+ [TAG_SRC_OPT, PRIORITY_OPT, SUBMIT_OPT],
"<node_name> tag...", "Add tags to the given node"),
- 'remove-tags': (
+ "remove-tags": (
RemoveTags, [ArgNode(min=1, max=1), ArgUnknown()],
- [TAG_SRC_OPT, PRIORITY_OPT],
+ [TAG_SRC_OPT, PRIORITY_OPT, SUBMIT_OPT],
"<node_name> tag...", "Remove tags from the given node"),
"health": (
Health, ARGS_MANY_NODES,
- [NOHDR_OPT, SEP_OPT, SUBMIT_OPT, PRIORITY_OPT, OOB_TIMEOUT_OPT],
+ [NOHDR_OPT, SEP_OPT, PRIORITY_OPT, OOB_TIMEOUT_OPT],
"[<node_name>...]", "List health of node(s) using out-of-band"),
+ "list-drbd": (
+ ListDrbd, ARGS_ONE_NODE,
+ [NOHDR_OPT, SEP_OPT],
+ "[<node_name>]", "Query the list of used DRBD minors on the given node"),
+ }
+
+#: dictionary with aliases for commands
+aliases = {
+ "show": "info",
}
def Main():
- return GenericMain(commands, override={"tag_type": constants.TAG_NODE})
+ return GenericMain(commands, aliases=aliases,
+ override={"tag_type": constants.TAG_NODE},
+ env_override=_ENV_OVERRIDE)