X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/5c54b832dc21858aa283f95b9e7d5c6abb521d79..468c5f773ae4d7b7e10e575afe71fbdedfd988ad:/lib/cmdlib.py diff --git a/lib/cmdlib.py b/lib/cmdlib.py index 4fc15dd..8c8f4f7 100644 --- a/lib/cmdlib.py +++ b/lib/cmdlib.py @@ -37,6 +37,7 @@ from ganeti import logger from ganeti import utils from ganeti import errors from ganeti import hypervisor +from ganeti import locking from ganeti import config from ganeti import constants from ganeti import objects @@ -49,22 +50,27 @@ class LogicalUnit(object): """Logical Unit base class. Subclasses must follow these rules: - - implement CheckPrereq which also fills in the opcode instance - with all the fields (even if as None) + - implement ExpandNames + - implement CheckPrereq - implement Exec - implement BuildHooksEnv - redefine HPATH and HTYPE - - optionally redefine their run requirements (REQ_CLUSTER, - REQ_MASTER); note that all commands require root permissions + - optionally redefine their run requirements: + REQ_MASTER: the LU needs to run on the master node + REQ_WSSTORE: the LU needs a writable SimpleStore + REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively + + Note that all commands require root permissions. """ HPATH = None HTYPE = None _OP_REQP = [] - REQ_CLUSTER = True REQ_MASTER = True + REQ_WSSTORE = False + REQ_BGL = True - def __init__(self, processor, op, cfg, sstore): + def __init__(self, processor, op, context, sstore): """Constructor for LogicalUnit. This needs to be overriden in derived classes in order to check op @@ -73,8 +79,10 @@ class LogicalUnit(object): """ self.proc = processor self.op = op - self.cfg = cfg + self.cfg = context.cfg self.sstore = sstore + self.context = context + self.needed_locks = None self.__ssh = None for attr_name in self._OP_REQP: @@ -82,15 +90,15 @@ class LogicalUnit(object): if attr_val is None: raise errors.OpPrereqError("Required parameter '%s' missing" % attr_name) - if self.REQ_CLUSTER: - if not cfg.IsCluster(): - raise errors.OpPrereqError("Cluster not initialized yet," - " use 'gnt-cluster init' first.") - if self.REQ_MASTER: - master = sstore.GetMasterNode() - if master != utils.HostInfo().name: - raise errors.OpPrereqError("Commands must be run on the master" - " node %s" % master) + + if not self.cfg.IsCluster(): + raise errors.OpPrereqError("Cluster not initialized yet," + " use 'gnt-cluster init' first.") + if self.REQ_MASTER: + master = sstore.GetMasterNode() + if master != utils.HostInfo().name: + raise errors.OpPrereqError("Commands must be run on the master" + " node %s" % master) def __GetSSH(self): """Returns the SshRunner object @@ -102,6 +110,46 @@ class LogicalUnit(object): ssh = property(fget=__GetSSH) + def ExpandNames(self): + """Expand names for this LU. + + This method is called before starting to execute the opcode, and it should + update all the parameters of the opcode to their canonical form (e.g. a + short node name must be fully expanded after this method has successfully + completed). This way locking, hooks, logging, ecc. can work correctly. + + LUs which implement this method must also populate the self.needed_locks + member, as a dict with lock levels as keys, and a list of needed lock names + as values. Rules: + - Use an empty dict if you don't need any lock + - If you don't need any lock at a particular level omit that level + - Don't put anything for the BGL level + - If you want all locks at a level use None as a value + (this reflects what LockSet does, and will be replaced before + CheckPrereq with the full list of nodes that have been locked) + + Examples: + # Acquire all nodes and one instance + self.needed_locks = { + locking.LEVEL_NODE: None, + locking.LEVEL_INSTANCES: ['instance1.example.tld'], + } + # Acquire just two nodes + self.needed_locks = { + locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'], + } + # Acquire no locks + self.needed_locks = {} # No, you can't leave it to the default value None + + """ + # The implementation of this method is mandatory only if the new LU is + # concurrent, so that old LUs don't need to be changed all at the same + # time. + if self.REQ_BGL: + self.needed_locks = {} # Exclusive LUs don't need locks. + else: + raise NotImplementedError + def CheckPrereq(self): """Check prerequisites for this LU. @@ -114,9 +162,7 @@ class LogicalUnit(object): not fulfilled. Its return value is ignored. This method should also update all the parameters of the opcode to - their canonical form; e.g. a short node name must be fully - expanded after this method has successfully completed (so that - hooks, logging, etc. work correctly). + their canonical form if it hasn't been done by ExpandNames before. """ raise NotImplementedError @@ -183,23 +229,6 @@ class NoHooksLU(LogicalUnit): HTYPE = None -def _AddHostToEtcHosts(hostname): - """Wrapper around utils.SetEtcHostsEntry. - - """ - hi = utils.HostInfo(name=hostname) - utils.SetEtcHostsEntry(constants.ETC_HOSTS, hi.ip, hi.name, [hi.ShortName()]) - - -def _RemoveHostFromEtcHosts(hostname): - """Wrapper around utils.RemoveEtcHostsEntry. - - """ - hi = utils.HostInfo(name=hostname) - utils.RemoveEtcHostsEntry(constants.ETC_HOSTS, hi.name) - utils.RemoveEtcHostsEntry(constants.ETC_HOSTS, hi.ShortName()) - - def _GetWantedNodes(lu, nodes): """Returns list of checked and expanded node names. @@ -323,85 +352,6 @@ def _BuildInstanceHookEnvByObject(instance, override=None): return _BuildInstanceHookEnv(**args) -def _HasValidVG(vglist, vgname): - """Checks if the volume group list is valid. - - A non-None return value means there's an error, and the return value - is the error message. - - """ - vgsize = vglist.get(vgname, None) - if vgsize is None: - return "volume group '%s' missing" % vgname - elif vgsize < 20480: - return ("volume group '%s' too small (20480MiB required, %dMib found)" % - (vgname, vgsize)) - return None - - -def _InitSSHSetup(node): - """Setup the SSH configuration for the cluster. - - - This generates a dsa keypair for root, adds the pub key to the - permitted hosts and adds the hostkey to its own known hosts. - - Args: - node: the name of this host as a fqdn - - """ - priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS) - - for name in priv_key, pub_key: - if os.path.exists(name): - utils.CreateBackup(name) - utils.RemoveFile(name) - - result = utils.RunCmd(["ssh-keygen", "-t", "dsa", - "-f", priv_key, - "-q", "-N", ""]) - if result.failed: - raise errors.OpExecError("Could not generate ssh keypair, error %s" % - result.output) - - f = open(pub_key, 'r') - try: - utils.AddAuthorizedKey(auth_keys, f.read(8192)) - finally: - f.close() - - -def _InitGanetiServerSetup(ss): - """Setup the necessary configuration for the initial node daemon. - - This creates the nodepass file containing the shared password for - the cluster and also generates the SSL certificate. - - """ - # Create pseudo random password - randpass = sha.new(os.urandom(64)).hexdigest() - # and write it into sstore - ss.SetKey(ss.SS_NODED_PASS, randpass) - - result = utils.RunCmd(["openssl", "req", "-new", "-newkey", "rsa:1024", - "-days", str(365*5), "-nodes", "-x509", - "-keyout", constants.SSL_CERT_FILE, - "-out", constants.SSL_CERT_FILE, "-batch"]) - if result.failed: - raise errors.OpExecError("could not generate server ssl cert, command" - " %s had exitcode %s and error message %s" % - (result.cmd, result.exit_code, result.output)) - - os.chmod(constants.SSL_CERT_FILE, 0400) - - result = utils.RunCmd([constants.NODE_INITD_SCRIPT, "restart"]) - - if result.failed: - raise errors.OpExecError("Could not start the node daemon, command %s" - " had exitcode %s and error %s" % - (result.cmd, result.exit_code, result.output)) - - def _CheckInstanceBridgesExist(instance): """Check that the brigdes needed by an instance exist. @@ -414,161 +364,6 @@ def _CheckInstanceBridgesExist(instance): (brlist, instance.primary_node)) -class LUInitCluster(LogicalUnit): - """Initialise the cluster. - - """ - HPATH = "cluster-init" - HTYPE = constants.HTYPE_CLUSTER - _OP_REQP = ["cluster_name", "hypervisor_type", "mac_prefix", - "def_bridge", "master_netdev", "file_storage_dir"] - REQ_CLUSTER = False - - def BuildHooksEnv(self): - """Build hooks env. - - Notes: Since we don't require a cluster, we must manually add - ourselves in the post-run node list. - - """ - env = {"OP_TARGET": self.op.cluster_name} - return env, [], [self.hostname.name] - - def CheckPrereq(self): - """Verify that the passed name is a valid one. - - """ - if config.ConfigWriter.IsCluster(): - raise errors.OpPrereqError("Cluster is already initialised") - - if self.op.hypervisor_type == constants.HT_XEN_HVM31: - if not os.path.exists(constants.VNC_PASSWORD_FILE): - raise errors.OpPrereqError("Please prepare the cluster VNC" - "password file %s" % - constants.VNC_PASSWORD_FILE) - - self.hostname = hostname = utils.HostInfo() - - if hostname.ip.startswith("127."): - raise errors.OpPrereqError("This host's IP resolves to the private" - " range (%s). Please fix DNS or %s." % - (hostname.ip, constants.ETC_HOSTS)) - - if not utils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT, - source=constants.LOCALHOST_IP_ADDRESS): - raise errors.OpPrereqError("Inconsistency: this host's name resolves" - " to %s,\nbut this ip address does not" - " belong to this host." - " Aborting." % hostname.ip) - - self.clustername = clustername = utils.HostInfo(self.op.cluster_name) - - if utils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT, - timeout=5): - raise errors.OpPrereqError("Cluster IP already active. Aborting.") - - secondary_ip = getattr(self.op, "secondary_ip", None) - if secondary_ip and not utils.IsValidIP(secondary_ip): - raise errors.OpPrereqError("Invalid secondary ip given") - if (secondary_ip and - secondary_ip != hostname.ip and - (not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT, - source=constants.LOCALHOST_IP_ADDRESS))): - raise errors.OpPrereqError("You gave %s as secondary IP," - " but it does not belong to this host." % - secondary_ip) - self.secondary_ip = secondary_ip - - if not hasattr(self.op, "vg_name"): - self.op.vg_name = None - # if vg_name not None, checks if volume group is valid - if self.op.vg_name: - vgstatus = _HasValidVG(utils.ListVolumeGroups(), self.op.vg_name) - if vgstatus: - raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if" - " you are not using lvm" % vgstatus) - - self.op.file_storage_dir = os.path.normpath(self.op.file_storage_dir) - - if not os.path.isabs(self.op.file_storage_dir): - raise errors.OpPrereqError("The file storage directory you have is" - " not an absolute path.") - - if not os.path.exists(self.op.file_storage_dir): - try: - os.makedirs(self.op.file_storage_dir, 0750) - except OSError, err: - raise errors.OpPrereqError("Cannot create file storage directory" - " '%s': %s" % - (self.op.file_storage_dir, err)) - - if not os.path.isdir(self.op.file_storage_dir): - raise errors.OpPrereqError("The file storage directory '%s' is not" - " a directory." % self.op.file_storage_dir) - - if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", - self.op.mac_prefix): - raise errors.OpPrereqError("Invalid mac prefix given '%s'" % - self.op.mac_prefix) - - if self.op.hypervisor_type not in constants.HYPER_TYPES: - raise errors.OpPrereqError("Invalid hypervisor type given '%s'" % - self.op.hypervisor_type) - - result = utils.RunCmd(["ip", "link", "show", "dev", self.op.master_netdev]) - if result.failed: - raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" % - (self.op.master_netdev, - result.output.strip())) - - if not (os.path.isfile(constants.NODE_INITD_SCRIPT) and - os.access(constants.NODE_INITD_SCRIPT, os.X_OK)): - raise errors.OpPrereqError("Init.d script '%s' missing or not" - " executable." % constants.NODE_INITD_SCRIPT) - - def Exec(self, feedback_fn): - """Initialize the cluster. - - """ - clustername = self.clustername - hostname = self.hostname - - # set up the simple store - self.sstore = ss = ssconf.SimpleStore() - ss.SetKey(ss.SS_HYPERVISOR, self.op.hypervisor_type) - ss.SetKey(ss.SS_MASTER_NODE, hostname.name) - ss.SetKey(ss.SS_MASTER_IP, clustername.ip) - ss.SetKey(ss.SS_MASTER_NETDEV, self.op.master_netdev) - ss.SetKey(ss.SS_CLUSTER_NAME, clustername.name) - ss.SetKey(ss.SS_FILE_STORAGE_DIR, self.op.file_storage_dir) - ss.SetKey(ss.SS_CONFIG_VERSION, constants.CONFIG_VERSION) - - # set up the inter-node password and certificate - _InitGanetiServerSetup(ss) - - # start the master ip - rpc.call_node_start_master(hostname.name) - - # set up ssh config and /etc/hosts - f = open(constants.SSH_HOST_RSA_PUB, 'r') - try: - sshline = f.read() - finally: - f.close() - sshkey = sshline.split(" ")[1] - - _AddHostToEtcHosts(hostname.name) - _InitSSHSetup(hostname.name) - - # init of cluster config file - self.cfg = cfgw = config.ConfigWriter() - cfgw.InitConfig(hostname.name, hostname.ip, self.secondary_ip, - sshkey, self.op.mac_prefix, - self.op.vg_name, self.op.def_bridge) - - ssh.WriteKnownHostsFile(cfgw, ss, constants.SSH_KNOWN_HOSTS_FILE) - - class LUDestroyCluster(NoHooksLU): """Logical unit for destroying the cluster. @@ -650,7 +445,8 @@ class LUVerifyCluster(LogicalUnit): (node,)) bad = True else: - vgstatus = _HasValidVG(vglist, self.cfg.GetVGName()) + vgstatus = utils.CheckVolumeGroupSize(vglist, self.cfg.GetVGName(), + constants.MIN_VG_SIZE) if vgstatus: feedback_fn(" - ERROR: %s on node %s" % (vgstatus, node)) bad = True @@ -1100,6 +896,7 @@ class LURenameCluster(LogicalUnit): HPATH = "cluster-rename" HTYPE = constants.HTYPE_CLUSTER _OP_REQP = ["name"] + REQ_WSSTORE = True def BuildHooksEnv(self): """Build hooks env. @@ -1126,8 +923,7 @@ class LURenameCluster(LogicalUnit): raise errors.OpPrereqError("Neither the name nor the IP address of the" " cluster has changed") if new_ip != old_ip: - result = utils.RunCmd(["fping", "-q", new_ip]) - if not result.failed: + if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT): raise errors.OpPrereqError("The given cluster IP address (%s) is" " reachable on the network. Aborting." % new_ip) @@ -1229,7 +1025,8 @@ class LUSetClusterParams(LogicalUnit): node_list = self.cfg.GetNodeList() vglist = rpc.call_vg_list(node_list) for node in node_list: - vgstatus = _HasValidVG(vglist[node], self.op.vg_name) + vgstatus = utils.CheckVolumeGroupSize(vglist[node], self.op.vg_name, + constants.MIN_VG_SIZE) if vgstatus: raise errors.OpPrereqError("Error on node '%s': %s" % (node, vgstatus)) @@ -1296,15 +1093,7 @@ def _WaitForSync(cfgw, instance, proc, oneshot=False, unlock=False): if done or oneshot: break - if unlock: - #utils.Unlock('cmd') - pass - try: - time.sleep(min(60, max_time)) - finally: - if unlock: - #utils.Lock('cmd') - pass + time.sleep(min(60, max_time)) if done: proc.LogInfo("Instance %s's disks are in sync." % instance.name) @@ -1432,7 +1221,7 @@ class LURemoveNode(LogicalUnit): """Build hooks env. This doesn't run on the target node in the pre phase as a failed - node would not allows itself to run. + node would then be impossible to remove. """ env = { @@ -1486,13 +1275,13 @@ class LURemoveNode(LogicalUnit): rpc.call_node_leave_cluster(node.name) - self.ssh.Run(node.name, 'root', "%s stop" % constants.NODE_INITD_SCRIPT) - logger.Info("Removing node %s from config" % node.name) self.cfg.RemoveNode(node.name) + # Remove the node from the Ganeti Lock Manager + self.context.glm.remove(locking.LEVEL_NODE, node.name) - _RemoveHostFromEtcHosts(node.name) + utils.RemoveHostFromEtcHosts(node.name) class LUQueryNodes(NoHooksLU): @@ -1516,7 +1305,7 @@ class LUQueryNodes(NoHooksLU): _CheckOutputFields(static=["name", "pinst_cnt", "sinst_cnt", "pinst_list", "sinst_list", - "pip", "sip"], + "pip", "sip", "tags"], dynamic=self.dynamic_fields, selected=self.op.output_fields) @@ -1587,6 +1376,8 @@ class LUQueryNodes(NoHooksLU): val = node.primary_ip elif field == "sip": val = node.secondary_ip + elif field == "tags": + val = list(node.GetTags()) elif field in self.dynamic_fields: val = live_data[node.name].get(field, None) else: @@ -1768,11 +1559,6 @@ class LUAddNode(LogicalUnit): primary_ip=primary_ip, secondary_ip=secondary_ip) - if self.sstore.GetHypervisorType() == constants.HT_XEN_HVM31: - if not os.path.exists(constants.VNC_PASSWORD_FILE): - raise errors.OpPrereqError("Cluster VNC password file %s missing" % - constants.VNC_PASSWORD_FILE) - def Exec(self, feedback_fn): """Adds the new node to the cluster. @@ -1780,46 +1566,7 @@ class LUAddNode(LogicalUnit): new_node = self.new_node node = new_node.name - # set up inter-node password and certificate and restarts the node daemon - gntpass = self.sstore.GetNodeDaemonPassword() - if not re.match('^[a-zA-Z0-9.]{1,64}$', gntpass): - raise errors.OpExecError("ganeti password corruption detected") - f = open(constants.SSL_CERT_FILE) - try: - gntpem = f.read(8192) - finally: - f.close() - # in the base64 pem encoding, neither '!' nor '.' are valid chars, - # so we use this to detect an invalid certificate; as long as the - # cert doesn't contain this, the here-document will be correctly - # parsed by the shell sequence below - if re.search('^!EOF\.', gntpem, re.MULTILINE): - raise errors.OpExecError("invalid PEM encoding in the SSL certificate") - if not gntpem.endswith("\n"): - raise errors.OpExecError("PEM must end with newline") - logger.Info("copy cluster pass to %s and starting the node daemon" % node) - - # and then connect with ssh to set password and start ganeti-noded - # note that all the below variables are sanitized at this point, - # either by being constants or by the checks above - ss = self.sstore - mycommand = ("umask 077 && " - "echo '%s' > '%s' && " - "cat > '%s' << '!EOF.' && \n" - "%s!EOF.\n%s restart" % - (gntpass, ss.KeyToFilename(ss.SS_NODED_PASS), - constants.SSL_CERT_FILE, gntpem, - constants.NODE_INITD_SCRIPT)) - - result = self.ssh.Run(node, 'root', mycommand, batch=False, ask_key=True) - if result.failed: - raise errors.OpExecError("Remote command on node %s, error: %s," - " output: %s" % - (node, result.fail_reason, result.output)) - # check connectivity - time.sleep(4) - result = rpc.call_version([node])[node] if result: if constants.PROTOCOL_VERSION == result: @@ -1854,7 +1601,7 @@ class LUAddNode(LogicalUnit): raise errors.OpExecError("Cannot transfer ssh keys to the new node") # Add node to our /etc/hosts, and add key to known_hosts - _AddHostToEtcHosts(new_node.name) + utils.AddHostToEtcHosts(new_node.name) if new_node.secondary_ip != new_node.primary_ip: if not rpc.call_node_tcp_ping(new_node.name, @@ -1866,12 +1613,22 @@ class LUAddNode(LogicalUnit): " you gave (%s). Please fix and re-run this" " command." % new_node.secondary_ip) - success, msg = self.ssh.VerifyNodeHostname(node) - if not success: - raise errors.OpExecError("Node '%s' claims it has a different hostname" - " than the one the resolver gives: %s." - " Please fix and re-run this command." % - (node, msg)) + node_verify_list = [self.sstore.GetMasterNode()] + node_verify_param = { + 'nodelist': [node], + # TODO: do a node-net-test as well? + } + + result = rpc.call_node_verify(node_verify_list, node_verify_param) + for verifier in node_verify_list: + if not result[verifier]: + raise errors.OpExecError("Cannot communicate with %s's node daemon" + " for remote verification" % verifier) + if result[verifier]['nodelist']: + for failed in result[verifier]['nodelist']: + feedback_fn("ssh/hostname verification failed %s -> %s" % + (verifier, result[verifier]['nodelist'][failed])) + raise errors.OpExecError("ssh/hostname verification failed.") # Distribute updated /etc/hosts and known_hosts to all nodes, # including the node just added @@ -1890,16 +1647,19 @@ class LUAddNode(LogicalUnit): logger.Error("copy of file %s to node %s failed" % (fname, to_node)) - to_copy = ss.GetFileList() + to_copy = self.sstore.GetFileList() if self.sstore.GetHypervisorType() == constants.HT_XEN_HVM31: to_copy.append(constants.VNC_PASSWORD_FILE) for fname in to_copy: - if not self.ssh.CopyFileToNode(node, fname): + result = rpc.call_upload_file([node], fname) + if not result[node]: logger.Error("could not copy file %s to node %s" % (fname, node)) if not self.op.readd: logger.Info("adding node %s to cluster.conf" % node) self.cfg.AddNode(new_node) + # Add the new node to the Ganeti Lock Manager + self.context.glm.add(locking.LEVEL_NODE, node) class LUMasterFailover(LogicalUnit): @@ -1911,6 +1671,7 @@ class LUMasterFailover(LogicalUnit): HPATH = "master-failover" HTYPE = constants.HTYPE_CLUSTER REQ_MASTER = False + REQ_WSSTORE = True _OP_REQP = [] def BuildHooksEnv(self): @@ -1999,50 +1760,12 @@ class LUQueryClusterInfo(NoHooksLU): "export_version": constants.EXPORT_VERSION, "master": self.sstore.GetMasterNode(), "architecture": (platform.architecture()[0], platform.machine()), + "hypervisor_type": self.sstore.GetHypervisorType(), } return result -class LUClusterCopyFile(NoHooksLU): - """Copy file to cluster. - - """ - _OP_REQP = ["nodes", "filename"] - - def CheckPrereq(self): - """Check prerequisites. - - It should check that the named file exists and that the given list - of nodes is valid. - - """ - if not os.path.exists(self.op.filename): - raise errors.OpPrereqError("No such filename '%s'" % self.op.filename) - - self.nodes = _GetWantedNodes(self, self.op.nodes) - - def Exec(self, feedback_fn): - """Copy a file from master to some nodes. - - Args: - opts - class with options as members - args - list containing a single element, the file name - Opts used: - nodes - list containing the name of target nodes; if empty, all nodes - - """ - filename = self.op.filename - - myname = utils.HostInfo().name - - for node in self.nodes: - if node == myname: - continue - if not self.ssh.CopyFileToNode(node, filename): - logger.Error("Copy of file %s to node %s failed" % (filename, node)) - - class LUDumpClusterConfig(NoHooksLU): """Return a text-representation of the cluster-config. @@ -2062,38 +1785,6 @@ class LUDumpClusterConfig(NoHooksLU): return self.cfg.DumpConfig() -class LURunClusterCommand(NoHooksLU): - """Run a command on some nodes. - - """ - _OP_REQP = ["command", "nodes"] - - def CheckPrereq(self): - """Check prerequisites. - - It checks that the given list of nodes is valid. - - """ - self.nodes = _GetWantedNodes(self, self.op.nodes) - - def Exec(self, feedback_fn): - """Run a command on some nodes. - - """ - # put the master at the end of the nodes list - master_node = self.sstore.GetMasterNode() - if master_node in self.nodes: - self.nodes.remove(master_node) - self.nodes.append(master_node) - - data = [] - for node in self.nodes: - result = self.ssh.Run(node, "root", self.op.command) - data.append((node, result.output, result.exit_code)) - - return data - - class LUActivateInstanceDisks(NoHooksLU): """Bring up an instance's disks. @@ -2603,9 +2294,7 @@ class LURenameInstance(LogicalUnit): new_name) if not getattr(self.op, "ignore_ip", False): - command = ["fping", "-q", name_info.ip] - result = utils.RunCmd(command) - if not result.failed: + if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT): raise errors.OpPrereqError("IP %s of instance %s already in use" % (name_info.ip, new_name)) @@ -2713,6 +2402,8 @@ class LURemoveInstance(LogicalUnit): logger.Info("removing instance %s out of cluster config" % instance.name) self.cfg.RemoveInstance(instance.name) + # Remove the new instance from the Ganeti Lock Manager + self.context.glm.remove(locking.LEVEL_INSTANCE, instance.name) class LUQueryInstances(NoHooksLU): @@ -2731,7 +2422,7 @@ class LUQueryInstances(NoHooksLU): _CheckOutputFields(static=["name", "os", "pnode", "snodes", "admin_state", "admin_ram", "disk_template", "ip", "mac", "bridge", - "sda_size", "sdb_size", "vcpus"], + "sda_size", "sdb_size", "vcpus", "tags"], dynamic=self.dynamic_fields, selected=self.op.output_fields) @@ -2824,6 +2515,8 @@ class LUQueryInstances(NoHooksLU): val = disk.size elif field == "vcpus": val = instance.vcpus + elif field == "tags": + val = list(instance.GetTags()) else: raise errors.ParameterError(field) iout.append(val) @@ -2872,7 +2565,7 @@ class LUFailoverInstance(LogicalUnit): secondary_nodes = instance.secondary_nodes if not secondary_nodes: raise errors.ProgrammerError("no secondary node but using " - "DT_REMOTE_RAID1 template") + "a mirrored disk template") target_node = secondary_nodes[0] # check memory requirements on the secondary node @@ -2902,7 +2595,7 @@ class LUFailoverInstance(LogicalUnit): feedback_fn("* checking disk consistency between source and target") for dev in instance.disks: - # for remote_raid1, these are md over drbd + # for drbd, these are drbd over lvm if not _CheckDiskConsistency(self.cfg, dev, target_node, False): if instance.status == "up" and not self.op.ignore_consistency: raise errors.OpExecError("Disk %s is degraded on target node," @@ -2927,7 +2620,7 @@ class LUFailoverInstance(LogicalUnit): instance.primary_node = target_node # distribute new instance config to the other nodes - self.cfg.AddInstance(instance) + self.cfg.Update(instance) # Only start the instance if it's marked as up if instance.status == "up": @@ -3011,22 +2704,6 @@ def _GenerateUniqueNames(cfg, exts): return results -def _GenerateMDDRBDBranch(cfg, primary, secondary, size, names): - """Generate a drbd device complete with its children. - - """ - port = cfg.AllocatePort() - vgname = cfg.GetVGName() - dev_data = objects.Disk(dev_type=constants.LD_LV, size=size, - logical_id=(vgname, names[0])) - dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128, - logical_id=(vgname, names[1])) - drbd_dev = objects.Disk(dev_type=constants.LD_DRBD7, size=size, - logical_id = (primary, secondary, port), - children = [dev_data, dev_meta]) - return drbd_dev - - def _GenerateDRBD8Branch(cfg, primary, secondary, size, names, iv_name): """Generate a drbd8 device complete with its children. @@ -3294,7 +2971,8 @@ class LUCreateInstance(LogicalUnit): """ # set optional parameters to none if they don't exist for attr in ["kernel_path", "initrd_path", "hvm_boot_order", "pnode", - "iallocator"]: + "iallocator", "hvm_acpi", "hvm_pae", "hvm_cdrom_image_path", + "vnc_bind_address"]: if not hasattr(self.op, attr): setattr(self.op, attr, None) @@ -3458,7 +3136,7 @@ class LUCreateInstance(LogicalUnit): info = nodeinfo.get(node, None) if not info: raise errors.OpPrereqError("Cannot get current information" - " from node '%s'" % nodeinfo) + " from node '%s'" % node) vg_free = info.get('vg_free', None) if not isinstance(vg_free, int): raise errors.OpPrereqError("Can't compute free disk space on" @@ -3484,6 +3162,31 @@ class LUCreateInstance(LogicalUnit): " destination node '%s'" % (self.op.bridge, pnode.name)) + # memory check on primary node + if self.op.start: + _CheckNodeFreeMemory(self.cfg, self.pnode.name, + "creating instance %s" % self.op.instance_name, + self.op.mem_size) + + # hvm_cdrom_image_path verification + if self.op.hvm_cdrom_image_path is not None: + if not os.path.isabs(self.op.hvm_cdrom_image_path): + raise errors.OpPrereqError("The path to the HVM CDROM image must" + " be an absolute path or None, not %s" % + self.op.hvm_cdrom_image_path) + if not os.path.isfile(self.op.hvm_cdrom_image_path): + raise errors.OpPrereqError("The HVM CDROM image must either be a" + " regular file or a symlink pointing to" + " an existing regular file, not %s" % + self.op.hvm_cdrom_image_path) + + # vnc_bind_address verification + if self.op.vnc_bind_address is not None: + if not utils.IsValidIP(self.op.vnc_bind_address): + raise errors.OpPrereqError("given VNC bind address '%s' doesn't look" + " like a valid IP address" % + self.op.vnc_bind_address) + if self.op.start: self.instance_status = 'up' else: @@ -3511,6 +3214,9 @@ class LUCreateInstance(LogicalUnit): else: network_port = None + if self.op.vnc_bind_address is None: + self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS + # this is needed because os.path.join does not accept None arguments if self.op.file_storage_dir is None: string_file_storage_dir = "" @@ -3542,6 +3248,10 @@ class LUCreateInstance(LogicalUnit): kernel_path=self.op.kernel_path, initrd_path=self.op.initrd_path, hvm_boot_order=self.op.hvm_boot_order, + hvm_acpi=self.op.hvm_acpi, + hvm_pae=self.op.hvm_pae, + hvm_cdrom_image_path=self.op.hvm_cdrom_image_path, + vnc_bind_address=self.op.vnc_bind_address, ) feedback_fn("* creating instance disks...") @@ -3552,6 +3262,8 @@ class LUCreateInstance(LogicalUnit): feedback_fn("adding instance %s to cluster config" % instance) self.cfg.AddInstance(iobj) + # Add the new instance to the Ganeti Lock Manager + self.context.glm.add(locking.LEVEL_INSTANCE, instance) if self.op.wait_for_sync: disk_abort = not _WaitForSync(self.cfg, iobj, self.proc) @@ -3566,6 +3278,8 @@ class LUCreateInstance(LogicalUnit): if disk_abort: _RemoveDisks(iobj, self.cfg) self.cfg.RemoveInstance(iobj.name) + # Remove the new instance from the Ganeti Lock Manager + self.context.glm.remove(locking.LEVEL_INSTANCE, iobj.name) raise errors.OpExecError("There are some degraded disks for" " this instance") @@ -3751,13 +3465,6 @@ class LUReplaceDisks(LogicalUnit): # replacement as for drbd7 (no different port allocated) raise errors.OpPrereqError("Same secondary given, cannot execute" " replacement") - # the user gave the current secondary, switch to - # 'no-replace-secondary' mode for drbd7 - remote_node = None - if (instance.disk_template == constants.DT_REMOTE_RAID1 and - self.op.mode != constants.REPLACE_DISK_ALL): - raise errors.OpPrereqError("Template 'remote_raid1' only allows all" - " disks replacement, not individual ones") if instance.disk_template == constants.DT_DRBD8: if (self.op.mode == constants.REPLACE_DISK_ALL and remote_node is not None): @@ -3789,101 +3496,6 @@ class LUReplaceDisks(LogicalUnit): (name, instance.name)) self.op.remote_node = remote_node - def _ExecRR1(self, feedback_fn): - """Replace the disks of an instance. - - """ - instance = self.instance - iv_names = {} - # start of work - if self.op.remote_node is None: - remote_node = self.sec_node - else: - remote_node = self.op.remote_node - cfg = self.cfg - for dev in instance.disks: - size = dev.size - lv_names = [".%s_%s" % (dev.iv_name, suf) for suf in ["data", "meta"]] - names = _GenerateUniqueNames(cfg, lv_names) - new_drbd = _GenerateMDDRBDBranch(cfg, instance.primary_node, - remote_node, size, names) - iv_names[dev.iv_name] = (dev, dev.children[0], new_drbd) - logger.Info("adding new mirror component on secondary for %s" % - dev.iv_name) - #HARDCODE - if not _CreateBlockDevOnSecondary(cfg, remote_node, instance, - new_drbd, False, - _GetInstanceInfoText(instance)): - raise errors.OpExecError("Failed to create new component on secondary" - " node %s. Full abort, cleanup manually!" % - remote_node) - - logger.Info("adding new mirror component on primary") - #HARDCODE - if not _CreateBlockDevOnPrimary(cfg, instance.primary_node, - instance, new_drbd, - _GetInstanceInfoText(instance)): - # remove secondary dev - cfg.SetDiskID(new_drbd, remote_node) - rpc.call_blockdev_remove(remote_node, new_drbd) - raise errors.OpExecError("Failed to create volume on primary!" - " Full abort, cleanup manually!!") - - # the device exists now - # call the primary node to add the mirror to md - logger.Info("adding new mirror component to md") - if not rpc.call_blockdev_addchildren(instance.primary_node, dev, - [new_drbd]): - logger.Error("Can't add mirror compoment to md!") - cfg.SetDiskID(new_drbd, remote_node) - if not rpc.call_blockdev_remove(remote_node, new_drbd): - logger.Error("Can't rollback on secondary") - cfg.SetDiskID(new_drbd, instance.primary_node) - if not rpc.call_blockdev_remove(instance.primary_node, new_drbd): - logger.Error("Can't rollback on primary") - raise errors.OpExecError("Full abort, cleanup manually!!") - - dev.children.append(new_drbd) - cfg.AddInstance(instance) - - # this can fail as the old devices are degraded and _WaitForSync - # does a combined result over all disks, so we don't check its - # return value - _WaitForSync(cfg, instance, self.proc, unlock=True) - - # so check manually all the devices - for name in iv_names: - dev, child, new_drbd = iv_names[name] - cfg.SetDiskID(dev, instance.primary_node) - is_degr = rpc.call_blockdev_find(instance.primary_node, dev)[5] - if is_degr: - raise errors.OpExecError("MD device %s is degraded!" % name) - cfg.SetDiskID(new_drbd, instance.primary_node) - is_degr = rpc.call_blockdev_find(instance.primary_node, new_drbd)[5] - if is_degr: - raise errors.OpExecError("New drbd device %s is degraded!" % name) - - for name in iv_names: - dev, child, new_drbd = iv_names[name] - logger.Info("remove mirror %s component" % name) - cfg.SetDiskID(dev, instance.primary_node) - if not rpc.call_blockdev_removechildren(instance.primary_node, - dev, [child]): - logger.Error("Can't remove child from mirror, aborting" - " *this device cleanup*.\nYou need to cleanup manually!!") - continue - - for node in child.logical_id[:2]: - logger.Info("remove child device on %s" % node) - cfg.SetDiskID(child, node) - if not rpc.call_blockdev_remove(node, child): - logger.Error("Warning: failed to remove device from node %s," - " continuing operation." % node) - - dev.children.remove(child) - - cfg.AddInstance(instance) - def _ExecD8DiskOnly(self, feedback_fn): """Replace a disk on the primary or secondary for dbrd8. @@ -4225,16 +3837,110 @@ class LUReplaceDisks(LogicalUnit): """ instance = self.instance - if instance.disk_template == constants.DT_REMOTE_RAID1: - fn = self._ExecRR1 - elif instance.disk_template == constants.DT_DRBD8: + + # Activate the instance disks if we're replacing them on a down instance + if instance.status == "down": + op = opcodes.OpActivateInstanceDisks(instance_name=instance.name) + self.proc.ChainOpCode(op) + + if instance.disk_template == constants.DT_DRBD8: if self.op.remote_node is None: fn = self._ExecD8DiskOnly else: fn = self._ExecD8Secondary else: raise errors.ProgrammerError("Unhandled disk replacement case") - return fn(feedback_fn) + + ret = fn(feedback_fn) + + # Deactivate the instance disks if we're replacing them on a down instance + if instance.status == "down": + op = opcodes.OpDeactivateInstanceDisks(instance_name=instance.name) + self.proc.ChainOpCode(op) + + return ret + + +class LUGrowDisk(LogicalUnit): + """Grow a disk of an instance. + + """ + HPATH = "disk-grow" + HTYPE = constants.HTYPE_INSTANCE + _OP_REQP = ["instance_name", "disk", "amount"] + + def BuildHooksEnv(self): + """Build hooks env. + + This runs on the master, the primary and all the secondaries. + + """ + env = { + "DISK": self.op.disk, + "AMOUNT": self.op.amount, + } + env.update(_BuildInstanceHookEnvByObject(self.instance)) + nl = [ + self.sstore.GetMasterNode(), + self.instance.primary_node, + ] + return env, nl, nl + + def CheckPrereq(self): + """Check prerequisites. + + This checks that the instance is in the cluster. + + """ + instance = self.cfg.GetInstanceInfo( + self.cfg.ExpandInstanceName(self.op.instance_name)) + if instance is None: + raise errors.OpPrereqError("Instance '%s' not known" % + self.op.instance_name) + self.instance = instance + self.op.instance_name = instance.name + + if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8): + raise errors.OpPrereqError("Instance's disk layout does not support" + " growing.") + + if instance.FindDisk(self.op.disk) is None: + raise errors.OpPrereqError("Disk '%s' not found for instance '%s'" % + (self.op.disk, instance.name)) + + nodenames = [instance.primary_node] + list(instance.secondary_nodes) + nodeinfo = rpc.call_node_info(nodenames, self.cfg.GetVGName()) + for node in nodenames: + info = nodeinfo.get(node, None) + if not info: + raise errors.OpPrereqError("Cannot get current information" + " from node '%s'" % node) + vg_free = info.get('vg_free', None) + if not isinstance(vg_free, int): + raise errors.OpPrereqError("Can't compute free disk space on" + " node %s" % node) + if self.op.amount > info['vg_free']: + raise errors.OpPrereqError("Not enough disk space on target node %s:" + " %d MiB available, %d MiB required" % + (node, info['vg_free'], self.op.amount)) + + def Exec(self, feedback_fn): + """Execute disk grow. + + """ + instance = self.instance + disk = instance.FindDisk(self.op.disk) + for node in (instance.secondary_nodes + (instance.primary_node,)): + self.cfg.SetDiskID(disk, node) + result = rpc.call_blockdev_grow(node, disk, self.op.amount) + if not result or not isinstance(result, tuple) or len(result) != 2: + raise errors.OpExecError("grow request failed to node %s" % node) + elif not result[0]: + raise errors.OpExecError("grow request failed to node %s: %s" % + (node, result[1])) + disk.RecordGrow(self.op.amount) + self.cfg.Update(instance) + return class LUQueryInstanceData(NoHooksLU): @@ -4330,13 +4036,24 @@ class LUQueryInstanceData(NoHooksLU): "memory": instance.memory, "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics], "disks": disks, - "network_port": instance.network_port, "vcpus": instance.vcpus, - "kernel_path": instance.kernel_path, - "initrd_path": instance.initrd_path, - "hvm_boot_order": instance.hvm_boot_order, } + htkind = self.sstore.GetHypervisorType() + if htkind == constants.HT_XEN_PVM30: + idict["kernel_path"] = instance.kernel_path + idict["initrd_path"] = instance.initrd_path + + if htkind == constants.HT_XEN_HVM31: + idict["hvm_boot_order"] = instance.hvm_boot_order + idict["hvm_acpi"] = instance.hvm_acpi + idict["hvm_pae"] = instance.hvm_pae + idict["hvm_cdrom_image_path"] = instance.hvm_cdrom_image_path + + if htkind in constants.HTS_REQ_PORT: + idict["vnc_bind_address"] = instance.vnc_bind_address + idict["network_port"] = instance.network_port + result[instance.name] = idict return result @@ -4394,9 +4111,15 @@ class LUSetInstanceParams(LogicalUnit): self.kernel_path = getattr(self.op, "kernel_path", None) self.initrd_path = getattr(self.op, "initrd_path", None) self.hvm_boot_order = getattr(self.op, "hvm_boot_order", None) - all_params = [self.mem, self.vcpus, self.ip, self.bridge, self.mac, - self.kernel_path, self.initrd_path, self.hvm_boot_order] - if all_params.count(None) == len(all_params): + self.hvm_acpi = getattr(self.op, "hvm_acpi", None) + self.hvm_pae = getattr(self.op, "hvm_pae", None) + self.hvm_cdrom_image_path = getattr(self.op, "hvm_cdrom_image_path", None) + self.vnc_bind_address = getattr(self.op, "vnc_bind_address", None) + all_parms = [self.mem, self.vcpus, self.ip, self.bridge, self.mac, + self.kernel_path, self.initrd_path, self.hvm_boot_order, + self.hvm_acpi, self.hvm_pae, self.hvm_cdrom_image_path, + self.vnc_bind_address] + if all_parms.count(None) == len(all_parms): raise errors.OpPrereqError("No changes submitted") if self.mem is not None: try: @@ -4455,6 +4178,25 @@ class LUSetInstanceParams(LogicalUnit): " must be one or more of [acdn]" " or 'default'") + # hvm_cdrom_image_path verification + if self.op.hvm_cdrom_image_path is not None: + if not os.path.isabs(self.op.hvm_cdrom_image_path): + raise errors.OpPrereqError("The path to the HVM CDROM image must" + " be an absolute path or None, not %s" % + self.op.hvm_cdrom_image_path) + if not os.path.isfile(self.op.hvm_cdrom_image_path): + raise errors.OpPrereqError("The HVM CDROM image must either be a" + " regular file or a symlink pointing to" + " an existing regular file, not %s" % + self.op.hvm_cdrom_image_path) + + # vnc_bind_address verification + if self.op.vnc_bind_address is not None: + if not utils.IsValidIP(self.op.vnc_bind_address): + raise errors.OpPrereqError("given VNC bind address '%s' doesn't look" + " like a valid IP address" % + self.op.vnc_bind_address) + instance = self.cfg.GetInstanceInfo( self.cfg.ExpandInstanceName(self.op.instance_name)) if instance is None: @@ -4498,6 +4240,18 @@ class LUSetInstanceParams(LogicalUnit): else: instance.hvm_boot_order = self.hvm_boot_order result.append(("hvm_boot_order", self.hvm_boot_order)) + if self.hvm_acpi: + instance.hvm_acpi = self.hvm_acpi + result.append(("hvm_acpi", self.hvm_acpi)) + if self.hvm_pae: + instance.hvm_pae = self.hvm_pae + result.append(("hvm_pae", self.hvm_pae)) + if self.hvm_cdrom_image_path: + instance.hvm_cdrom_image_path = self.hvm_cdrom_image_path + result.append(("hvm_cdrom_image_path", self.hvm_cdrom_image_path)) + if self.vnc_bind_address: + instance.vnc_bind_address = self.vnc_bind_address + result.append(("vnc_bind_address", self.vnc_bind_address)) self.cfg.AddInstance(instance) @@ -4839,22 +4593,31 @@ class LUDelTags(TagsLU): class LUTestDelay(NoHooksLU): """Sleep for a specified amount of time. - This LU sleeps on the master and/or nodes for a specified amoutn of + This LU sleeps on the master and/or nodes for a specified amount of time. """ _OP_REQP = ["duration", "on_master", "on_nodes"] + REQ_BGL = False - def CheckPrereq(self): - """Check prerequisites. + def ExpandNames(self): + """Expand names and set required locks. - This checks that we have a good list of nodes and/or the duration - is valid. + This expands the node list, if any. """ - + self.needed_locks = {} if self.op.on_nodes: + # _GetWantedNodes can be used here, but is not always appropriate to use + # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for + # more information. self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes) + self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes + + def CheckPrereq(self): + """Check prerequisites. + + """ def Exec(self, feedback_fn): """Do the actual sleep.