X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/64d3bd527390e9b67e512213c33e7cb1c9779bd2..f12eadb345d80b51c0aa667b0e7cf62d97925448:/lib/config.py diff --git a/lib/config.py b/lib/config.py index efd24d4..9256bb9 100644 --- a/lib/config.py +++ b/lib/config.py @@ -38,25 +38,29 @@ import logging from ganeti import errors from ganeti import locking -from ganeti import logger from ganeti import utils from ganeti import constants from ganeti import rpc from ganeti import objects from ganeti import serializer -from ganeti import ssconf _config_lock = locking.SharedLock() -def ValidateConfig(): - sstore = ssconf.SimpleStore() +def _ValidateConfig(data): + """Verifies that a configuration objects looks valid. - if sstore.GetConfigVersion() != constants.CONFIG_VERSION: + This only verifies the version of the configuration. + + @raise errors.ConfigurationError: if the version differs from what + we expect + + """ + if data.version != constants.CONFIG_VERSION: raise errors.ConfigurationError("Cluster configuration version" " mismatch, got %s instead of %s" % - (sstore.GetConfigVersion(), + (data.version, constants.CONFIG_VERSION)) @@ -68,20 +72,20 @@ class ConfigWriter: self.write_count = 0 self._lock = _config_lock self._config_data = None - self._config_time = None - self._config_size = None - self._config_inode = None self._offline = offline if cfg_file is None: self._cfg_file = constants.CLUSTER_CONF_FILE else: self._cfg_file = cfg_file self._temporary_ids = set() + self._temporary_drbds = {} # Note: in order to prevent errors when resolving our name in # _DistributeConfig, we compute it here once and reuse it; it's # better to raise an error before starting to modify the config # file than after it was modified self._my_hostname = utils.HostInfo().name + self._last_cluster_serial = -1 + self._OpenConfig() # this method needs to be static, so that we can call it on the class @staticmethod @@ -98,7 +102,6 @@ class ConfigWriter: This should check the current instances for duplicates. """ - self._OpenConfig() prefix = self._config_data.cluster.mac_prefix all_macs = self._AllMACs() retries = 64 @@ -122,15 +125,31 @@ class ConfigWriter: check for potential collisions elsewhere. """ - self._OpenConfig() all_macs = self._AllMACs() return mac in all_macs + @locking.ssynchronized(_config_lock, shared=1) + def GenerateDRBDSecret(self): + """Generate a DRBD secret. + + This checks the current disks for duplicates. + + """ + all_secrets = self._AllDRBDSecrets() + retries = 64 + while retries > 0: + secret = utils.GenerateSecret() + if secret not in all_secrets: + break + retries -= 1 + else: + raise errors.ConfigurationError("Can't generate unique DRBD secret") + return secret + def _ComputeAllLVs(self): """Compute the list of all LVs. """ - self._OpenConfig() lvnames = set() for instance in self._config_data.instances.values(): node_data = instance.MapLVsByNode() @@ -145,13 +164,13 @@ class ConfigWriter: This checks the current node, instances and disk names for duplicates. - Args: - - exceptions: a list with some other names which should be checked - for uniqueness (used for example when you want to get - more than one id at one time without adding each one in - turn to the config file + @param exceptions: a list with some other names which should be checked + for uniqueness (used for example when you want to get + more than one id at one time without adding each one in + turn to the config file) - Returns: the unique id as a string + @rtype: string + @return: the unique id """ existing = set() @@ -175,9 +194,10 @@ class ConfigWriter: def _AllMACs(self): """Return all MACs present in the config. - """ - self._OpenConfig() + @rtype: list + @return: the list of all MACs + """ result = [] for instance in self._config_data.instances.values(): for nic in instance.nics: @@ -185,14 +205,39 @@ class ConfigWriter: return result - @locking.ssynchronized(_config_lock, shared=1) - def VerifyConfig(self): - """Stub verify function. + def _AllDRBDSecrets(self): + """Return all DRBD secrets present in the config. + + @rtype: list + @return: the list of all DRBD secrets + """ - self._OpenConfig() + def helper(disk, result): + """Recursively gather secrets from this disk.""" + if disk.dev_type == constants.DT_DRBD8: + result.append(disk.logical_id[5]) + if disk.children: + for child in disk.children: + helper(child, result) + + result = [] + for instance in self._config_data.instances.values(): + for disk in instance.disks: + helper(disk, result) + + return result + def _UnlockedVerifyConfig(self): + """Verify function. + + @rtype: list + @return: a list of error messages; a non-empty list signifies + configuration errors + + """ result = [] seen_macs = [] + ports = {} data = self._config_data for instance_name in data.instances: instance = data.instances[instance_name] @@ -209,8 +254,77 @@ class ConfigWriter: (instance_name, idx, nic.mac)) else: seen_macs.append(nic.mac) + + # gather the drbd ports for duplicate checks + for dsk in instance.disks: + if dsk.dev_type in constants.LDS_DRBD: + tcp_port = dsk.logical_id[2] + if tcp_port not in ports: + ports[tcp_port] = [] + ports[tcp_port].append((instance.name, "drbd disk %s" % dsk.iv_name)) + # gather network port reservation + net_port = getattr(instance, "network_port", None) + if net_port is not None: + if net_port not in ports: + ports[net_port] = [] + ports[net_port].append((instance.name, "network port")) + + # instance disk verify + for idx, disk in enumerate(instance.disks): + result.extend(["instance '%s' disk %d error: %s" % + (instance.name, idx, msg) for msg in disk.Verify()]) + + # cluster-wide pool of free ports + for free_port in data.cluster.tcpudp_port_pool: + if free_port not in ports: + ports[free_port] = [] + ports[free_port].append(("cluster", "port marked as free")) + + # compute tcp/udp duplicate ports + keys = ports.keys() + keys.sort() + for pnum in keys: + pdata = ports[pnum] + if len(pdata) > 1: + txt = ", ".join(["%s/%s" % val for val in pdata]) + result.append("tcp/udp port %s has duplicates: %s" % (pnum, txt)) + + # highest used tcp port check + if keys: + if keys[-1] > data.cluster.highest_used_port: + result.append("Highest used port mismatch, saved %s, computed %s" % + (data.cluster.highest_used_port, keys[-1])) + + if not data.nodes[data.cluster.master_node].master_candidate: + result.append("Master node is not a master candidate") + + # master candidate checks + mc_now, mc_max = self._UnlockedGetMasterCandidateStats() + if mc_now < mc_max: + result.append("Not enough master candidates: actual %d, target %d" % + (mc_now, mc_max)) + + # drbd minors check + d_map, duplicates = self._UnlockedComputeDRBDMap() + for node, minor, instance_a, instance_b in duplicates: + result.append("DRBD minor %d on node %s is assigned twice to instances" + " %s and %s" % (minor, node, instance_a, instance_b)) + return result + @locking.ssynchronized(_config_lock, shared=1) + def VerifyConfig(self): + """Verify function. + + This is just a wrapper over L{_UnlockedVerifyConfig}. + + @rtype: list + @return: a list of error messages; a non-empty list signifies + configuration errors + + """ + return self._UnlockedVerifyConfig() + def _UnlockedSetDiskID(self, disk, node_name): """Convert the unique ID to the ID needed on the target nodes. @@ -229,8 +343,8 @@ class ConfigWriter: if disk.logical_id is None and disk.physical_id is not None: return - if disk.dev_type in constants.LDS_DRBD: - pnode, snode, port = disk.logical_id + if disk.dev_type == constants.LD_DRBD8: + pnode, snode, port, pminor, sminor, secret = disk.logical_id if node_name not in (pnode, snode): raise errors.ConfigurationError("DRBD device not knowing node %s" % node_name) @@ -239,12 +353,12 @@ class ConfigWriter: if pnode_info is None or snode_info is None: raise errors.ConfigurationError("Can't find primary or secondary node" " for %s" % str(disk)) + p_data = (pnode_info.secondary_ip, port) + s_data = (snode_info.secondary_ip, port) if pnode == node_name: - disk.physical_id = (pnode_info.secondary_ip, port, - snode_info.secondary_ip, port) + disk.physical_id = p_data + s_data + (pminor, secret) else: # it must be secondary, we tested above - disk.physical_id = (snode_info.secondary_ip, port, - pnode_info.secondary_ip, port) + disk.physical_id = s_data + p_data + (sminor, secret) else: disk.physical_id = disk.logical_id return @@ -270,7 +384,6 @@ class ConfigWriter: if not isinstance(port, int): raise errors.ProgrammerError("Invalid type passed for port") - self._OpenConfig() self._config_data.cluster.tcpudp_port_pool.add(port) self._WriteConfig() @@ -279,7 +392,6 @@ class ConfigWriter: """Returns a copy of the current port list. """ - self._OpenConfig() return self._config_data.cluster.tcpudp_port_pool.copy() @locking.ssynchronized(_config_lock) @@ -291,8 +403,6 @@ class ConfigWriter: highest_used_port). """ - self._OpenConfig() - # If there are TCP/IP ports configured, we use them first. if self._config_data.cluster.tcpudp_port_pool: port = self._config_data.cluster.tcpudp_port_pool.pop() @@ -307,15 +417,214 @@ class ConfigWriter: self._WriteConfig() return port + def _UnlockedComputeDRBDMap(self): + """Compute the used DRBD minor/nodes. + + @rtype: (dict, list) + @return: dictionary of node_name: dict of minor: instance_name; + the returned dict will have all the nodes in it (even if with + an empty list), and a list of duplicates; if the duplicates + list is not empty, the configuration is corrupted and its caller + should raise an exception + + """ + def _AppendUsedPorts(instance_name, disk, used): + duplicates = [] + if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5: + nodeA, nodeB, dummy, minorA, minorB = disk.logical_id[:5] + for node, port in ((nodeA, minorA), (nodeB, minorB)): + assert node in used, ("Node '%s' of instance '%s' not found" + " in node list" % (node, instance_name)) + if port in used[node]: + duplicates.append((node, port, instance_name, used[node][port])) + else: + used[node][port] = instance_name + if disk.children: + for child in disk.children: + duplicates.extend(_AppendUsedPorts(instance_name, child, used)) + return duplicates + + duplicates = [] + my_dict = dict((node, {}) for node in self._config_data.nodes) + for instance in self._config_data.instances.itervalues(): + for disk in instance.disks: + duplicates.extend(_AppendUsedPorts(instance.name, disk, my_dict)) + for (node, minor), instance in self._temporary_drbds.iteritems(): + if minor in my_dict[node] and my_dict[node][minor] != instance: + duplicates.append((node, minor, instance, my_dict[node][minor])) + else: + my_dict[node][minor] = instance + return my_dict, duplicates + + @locking.ssynchronized(_config_lock) + def ComputeDRBDMap(self): + """Compute the used DRBD minor/nodes. + + This is just a wrapper over L{_UnlockedComputeDRBDMap}. + + @return: dictionary of node_name: dict of minor: instance_name; + the returned dict will have all the nodes in it (even if with + an empty list). + + """ + d_map, duplicates = self._UnlockedComputeDRBDMap() + if duplicates: + raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" % + str(duplicates)) + return d_map + + @locking.ssynchronized(_config_lock) + def AllocateDRBDMinor(self, nodes, instance): + """Allocate a drbd minor. + + The free minor will be automatically computed from the existing + devices. A node can be given multiple times in order to allocate + multiple minors. The result is the list of minors, in the same + order as the passed nodes. + + @type instance: string + @param instance: the instance for which we allocate minors + + """ + assert isinstance(instance, basestring), \ + "Invalid argument '%s' passed to AllocateDRBDMinor" % instance + + d_map, duplicates = self._UnlockedComputeDRBDMap() + if duplicates: + raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" % + str(duplicates)) + result = [] + for nname in nodes: + ndata = d_map[nname] + if not ndata: + # no minors used, we can start at 0 + result.append(0) + ndata[0] = instance + self._temporary_drbds[(nname, 0)] = instance + continue + keys = ndata.keys() + keys.sort() + ffree = utils.FirstFree(keys) + if ffree is None: + # return the next minor + # TODO: implement high-limit check + minor = keys[-1] + 1 + else: + minor = ffree + # double-check minor against current instances + assert minor not in d_map[nname], \ + ("Attempt to reuse allocated DRBD minor %d on node %s," + " already allocated to instance %s" % + (minor, nname, d_map[nname][minor])) + ndata[minor] = instance + # double-check minor against reservation + r_key = (nname, minor) + assert r_key not in self._temporary_drbds, \ + ("Attempt to reuse reserved DRBD minor %d on node %s," + " reserved for instance %s" % + (minor, nname, self._temporary_drbds[r_key])) + self._temporary_drbds[r_key] = instance + result.append(minor) + logging.debug("Request to allocate drbd minors, input: %s, returning %s", + nodes, result) + return result + + def _UnlockedReleaseDRBDMinors(self, instance): + """Release temporary drbd minors allocated for a given instance. + + @type instance: string + @param instance: the instance for which temporary minors should be + released + + """ + assert isinstance(instance, basestring), \ + "Invalid argument passed to ReleaseDRBDMinors" + for key, name in self._temporary_drbds.items(): + if name == instance: + del self._temporary_drbds[key] + + @locking.ssynchronized(_config_lock) + def ReleaseDRBDMinors(self, instance): + """Release temporary drbd minors allocated for a given instance. + + This should be called on the error paths, on the success paths + it's automatically called by the ConfigWriter add and update + functions. + + This function is just a wrapper over L{_UnlockedReleaseDRBDMinors}. + + @type instance: string + @param instance: the instance for which temporary minors should be + released + + """ + self._UnlockedReleaseDRBDMinors(instance) + + @locking.ssynchronized(_config_lock, shared=1) + def GetConfigVersion(self): + """Get the configuration version. + + @return: Config version + + """ + return self._config_data.version + + @locking.ssynchronized(_config_lock, shared=1) + def GetClusterName(self): + """Get cluster name. + + @return: Cluster name + + """ + return self._config_data.cluster.cluster_name + + @locking.ssynchronized(_config_lock, shared=1) + def GetMasterNode(self): + """Get the hostname of the master node for this cluster. + + @return: Master hostname + + """ + return self._config_data.cluster.master_node + + @locking.ssynchronized(_config_lock, shared=1) + def GetMasterIP(self): + """Get the IP of the master node for this cluster. + + @return: Master IP + + """ + return self._config_data.cluster.master_ip + + @locking.ssynchronized(_config_lock, shared=1) + def GetMasterNetdev(self): + """Get the master network device for this cluster. + + """ + return self._config_data.cluster.master_netdev + + @locking.ssynchronized(_config_lock, shared=1) + def GetFileStorageDir(self): + """Get the file storage dir for this cluster. + + """ + return self._config_data.cluster.file_storage_dir + + @locking.ssynchronized(_config_lock, shared=1) + def GetHypervisorType(self): + """Get the hypervisor type for this cluster. + + """ + return self._config_data.cluster.default_hypervisor + @locking.ssynchronized(_config_lock, shared=1) def GetHostKey(self): """Return the rsa hostkey from the config. - Args: None + @rtype: string + @return: the rsa hostkey - Returns: rsa hostkey """ - self._OpenConfig() return self._config_data.cluster.rsahostkeypub @locking.ssynchronized(_config_lock) @@ -324,36 +633,36 @@ class ConfigWriter: This should be used after creating a new instance. - Args: - instance: the instance object + @type instance: L{objects.Instance} + @param instance: the instance object + """ if not isinstance(instance, objects.Instance): raise errors.ProgrammerError("Invalid type passed to AddInstance") if instance.disk_template != constants.DT_DISKLESS: all_lvs = instance.MapLVsByNode() - logger.Info("Instance '%s' DISK_LAYOUT: %s" % (instance.name, all_lvs)) + logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs) - self._OpenConfig() + instance.serial_no = 1 self._config_data.instances[instance.name] = instance + self._UnlockedReleaseDRBDMinors(instance.name) self._WriteConfig() def _SetInstanceStatus(self, instance_name, status): """Set the instance's status to a given value. """ - if status not in ("up", "down"): - raise errors.ProgrammerError("Invalid status '%s' passed to" - " ConfigWriter._SetInstanceStatus()" % - status) - self._OpenConfig() + assert isinstance(status, bool), \ + "Invalid status '%s' passed to SetInstanceStatus" % (status,) if instance_name not in self._config_data.instances: raise errors.ConfigurationError("Unknown instance '%s'" % instance_name) instance = self._config_data.instances[instance_name] - if instance.status != status: - instance.status = status + if instance.admin_up != status: + instance.admin_up = status + instance.serial_no += 1 self._WriteConfig() @locking.ssynchronized(_config_lock) @@ -361,15 +670,13 @@ class ConfigWriter: """Mark the instance status to up in the config. """ - self._SetInstanceStatus(instance_name, "up") + self._SetInstanceStatus(instance_name, True) @locking.ssynchronized(_config_lock) def RemoveInstance(self, instance_name): """Remove the instance from the configuration. """ - self._OpenConfig() - if instance_name not in self._config_data.instances: raise errors.ConfigurationError("Unknown instance '%s'" % instance_name) del self._config_data.instances[instance_name] @@ -384,7 +691,6 @@ class ConfigWriter: rename. """ - self._OpenConfig() if old_name not in self._config_data.instances: raise errors.ConfigurationError("Unknown instance '%s'" % old_name) inst = self._config_data.instances[old_name] @@ -408,7 +714,7 @@ class ConfigWriter: """Mark the status of an instance to down in the configuration. """ - self._SetInstanceStatus(instance_name, "down") + self._SetInstanceStatus(instance_name, False) def _UnlockedGetInstanceList(self): """Get the list of instances. @@ -416,16 +722,14 @@ class ConfigWriter: This function is for internal use, when the config lock is already held. """ - self._OpenConfig() return self._config_data.instances.keys() @locking.ssynchronized(_config_lock, shared=1) def GetInstanceList(self): """Get the list of instances. - Returns: - array of instances, ex. ['instance2.example.com','instance1.example.com'] - these contains all the instances, also the ones in Admin_down state + @return: array of instances, ex. ['instance2.example.com', + 'instance1.example.com'] """ return self._UnlockedGetInstanceList() @@ -435,8 +739,6 @@ class ConfigWriter: """Attempt to expand an incomplete instance name. """ - self._OpenConfig() - return utils.MatchNameComponent(short_name, self._config_data.instances.keys()) @@ -446,8 +748,6 @@ class ConfigWriter: This function is for internal use, when the config lock is already held. """ - self._OpenConfig() - if instance_name not in self._config_data.instances: return None @@ -460,11 +760,11 @@ class ConfigWriter: It takes the information from the configuration file. Other informations of an instance are taken from the live systems. - Args: - instance: name of the instance, ex instance1.example.com + @param instance_name: name of the instance, e.g. + I{instance1.example.com} - Returns: - the instance object + @rtype: L{objects.Instance} + @return: the instance object """ return self._UnlockedGetInstanceInfo(instance_name) @@ -486,14 +786,15 @@ class ConfigWriter: def AddNode(self, node): """Add a node to the configuration. - Args: - node: an object.Node instance + @type node: L{objects.Node} + @param node: a Node instance """ logging.info("Adding node %s to configuration" % node.name) - self._OpenConfig() + node.serial_no = 1 self._config_data.nodes[node.name] = node + self._config_data.cluster.serial_no += 1 self._WriteConfig() @locking.ssynchronized(_config_lock) @@ -503,11 +804,11 @@ class ConfigWriter: """ logging.info("Removing node %s from configuration" % node_name) - self._OpenConfig() if node_name not in self._config_data.nodes: raise errors.ConfigurationError("Unknown node '%s'" % node_name) del self._config_data.nodes[node_name] + self._config_data.cluster.serial_no += 1 self._WriteConfig() @locking.ssynchronized(_config_lock, shared=1) @@ -515,23 +816,21 @@ class ConfigWriter: """Attempt to expand an incomplete instance name. """ - self._OpenConfig() - return utils.MatchNameComponent(short_name, self._config_data.nodes.keys()) def _UnlockedGetNodeInfo(self, node_name): """Get the configuration of a node, as stored in the config. - This function is for internal use, when the config lock is already held. + This function is for internal use, when the config lock is already + held. - Args: node: nodename (tuple) of the node + @param node_name: the node name, e.g. I{node1.example.com} - Returns: the node object + @rtype: L{objects.Node} + @return: the node object """ - self._OpenConfig() - if node_name not in self._config_data.nodes: return None @@ -542,9 +841,12 @@ class ConfigWriter: def GetNodeInfo(self, node_name): """Get the configuration of a node, as stored in the config. - Args: node: nodename (tuple) of the node + This is just a locked wrapper over L{_UnlockedGetNodeInfo}. + + @param node_name: the node name, e.g. I{node1.example.com} - Returns: the node object + @rtype: L{objects.Node} + @return: the node object """ return self._UnlockedGetNodeInfo(node_name) @@ -552,10 +854,12 @@ class ConfigWriter: def _UnlockedGetNodeList(self): """Return the list of nodes which are in the configuration. - This function is for internal use, when the config lock is already held. + This function is for internal use, when the config lock is already + held. + + @rtype: list """ - self._OpenConfig() return self._config_data.nodes.keys() @@ -567,11 +871,20 @@ class ConfigWriter: return self._UnlockedGetNodeList() @locking.ssynchronized(_config_lock, shared=1) + def GetOnlineNodeList(self): + """Return the list of nodes which are online. + + """ + all_nodes = [self._UnlockedGetNodeInfo(node) + for node in self._UnlockedGetNodeList()] + return [node.name for node in all_nodes if not node.offline] + + @locking.ssynchronized(_config_lock, shared=1) def GetAllNodesInfo(self): """Get the configuration of all nodes. @rtype: dict - @returns: dict of (node, node_info), where node_info is what + @return: dict of (node, node_info), where node_info is what would GetNodeInfo return for the node """ @@ -579,42 +892,77 @@ class ConfigWriter: for node in self._UnlockedGetNodeList()]) return my_dict + def _UnlockedGetMasterCandidateStats(self): + """Get the number of current and maximum desired and possible candidates. + + @rtype: tuple + @return: tuple of (current, desired and possible) + + """ + mc_now = mc_max = 0 + for node in self._config_data.nodes.itervalues(): + if not node.offline: + mc_max += 1 + if node.master_candidate: + mc_now += 1 + mc_max = min(mc_max, self._config_data.cluster.candidate_pool_size) + return (mc_now, mc_max) + @locking.ssynchronized(_config_lock, shared=1) - def DumpConfig(self): - """Return the entire configuration of the cluster. + def GetMasterCandidateStats(self): + """Get the number of current and maximum possible candidates. + + This is just a wrapper over L{_UnlockedGetMasterCandidateStats}. + + @rtype: tuple + @return: tuple of (current, max) + """ - self._OpenConfig() - return self._config_data + return self._UnlockedGetMasterCandidateStats() + + @locking.ssynchronized(_config_lock) + def MaintainCandidatePool(self): + """Try to grow the candidate pool to the desired size. + + @rtype: list + @return: list with the adjusted nodes (L{objects.Node} instances) + + """ + mc_now, mc_max = self._UnlockedGetMasterCandidateStats() + mod_list = [] + if mc_now < mc_max: + node_list = self._config_data.nodes.keys() + random.shuffle(node_list) + for name in node_list: + if mc_now >= mc_max: + break + node = self._config_data.nodes[name] + if node.master_candidate or node.offline: + continue + mod_list.append(node) + node.master_candidate = True + node.serial_no += 1 + mc_now += 1 + if mc_now != mc_max: + # this should not happen + logging.warning("Warning: MaintainCandidatePool didn't manage to" + " fill the candidate pool (%d/%d)", mc_now, mc_max) + if mod_list: + self._config_data.cluster.serial_no += 1 + self._WriteConfig() + + return mod_list def _BumpSerialNo(self): """Bump up the serial number of the config. """ - self._config_data.cluster.serial_no += 1 + self._config_data.serial_no += 1 def _OpenConfig(self): """Read the config data from disk. - In case we already have configuration data and the config file has - the same mtime as when we read it, we skip the parsing of the - file, since de-serialisation could be slow. - """ - try: - st = os.stat(self._cfg_file) - except OSError, err: - raise errors.ConfigurationError("Can't stat config file: %s" % err) - if (self._config_data is not None and - self._config_time is not None and - self._config_time == st.st_mtime and - self._config_size == st.st_size and - self._config_inode == st.st_ino): - # data is current, so skip loading of config file - return - - # Make sure the configuration has the right version - ValidateConfig() - f = open(self._cfg_file, 'r') try: try: @@ -623,14 +971,18 @@ class ConfigWriter: raise errors.ConfigurationError(err) finally: f.close() + + # Make sure the configuration has the right version + _ValidateConfig(data) + if (not hasattr(data, 'cluster') or not hasattr(data.cluster, 'rsahostkeypub')): raise errors.ConfigurationError("Incomplete configuration" " (missing cluster.rsahostkeypub)") self._config_data = data - self._config_time = st.st_mtime - self._config_size = st.st_size - self._config_inode = st.st_ino + # reset the last serial as -1 so that the next write will cause + # ssconf update + self._last_cluster_serial = -1 def _DistributeConfig(self): """Distribute the configuration to the other nodes. @@ -642,19 +994,29 @@ class ConfigWriter: if self._offline: return True bad = False - nodelist = self._UnlockedGetNodeList() - myhostname = self._my_hostname - try: - nodelist.remove(myhostname) - except ValueError: - pass - - result = rpc.call_upload_file(nodelist, self._cfg_file) - for node in nodelist: + node_list = [] + addr_list = [] + myhostname = self._my_hostname + # we can skip checking whether _UnlockedGetNodeInfo returns None + # since the node list comes from _UnlocketGetNodeList, and we are + # called with the lock held, so no modifications should take place + # in between + for node_name in self._UnlockedGetNodeList(): + if node_name == myhostname: + continue + node_info = self._UnlockedGetNodeInfo(node_name) + if not node_info.master_candidate: + continue + node_list.append(node_info.name) + addr_list.append(node_info.primary_ip) + + result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file, + address_list=addr_list) + for node in node_list: if not result[node]: - logger.Error("copy of file %s to node %s failed" % - (self._cfg_file, node)) + logging.error("copy of file %s to node %s failed", + self._cfg_file, node) bad = True return not bad @@ -662,6 +1024,11 @@ class ConfigWriter: """Write the configuration data to persistent storage. """ + config_errors = self._UnlockedVerifyConfig() + if config_errors: + raise errors.ConfigurationError("Configuration data is not" + " consistent: %s" % + (", ".join(config_errors))) if destination is None: destination = self._cfg_file self._BumpSerialNo() @@ -677,48 +1044,70 @@ class ConfigWriter: # we don't need to do os.close(fd) as f.close() did it os.rename(name, destination) self.write_count += 1 - # re-set our cache as not to re-read the config file - try: - st = os.stat(destination) - except OSError, err: - raise errors.ConfigurationError("Can't stat config file: %s" % err) - self._config_time = st.st_mtime - self._config_size = st.st_size - self._config_inode = st.st_ino - # and redistribute the config file + + # and redistribute the config file to master candidates self._DistributeConfig() + # Write ssconf files on all nodes (including locally) + if self._last_cluster_serial < self._config_data.cluster.serial_no: + if not self._offline: + rpc.RpcRunner.call_write_ssconf_files(self._UnlockedGetNodeList(), + self._UnlockedGetSsconfValues()) + self._last_cluster_serial = self._config_data.cluster.serial_no + + def _UnlockedGetSsconfValues(self): + """Return the values needed by ssconf. + + @rtype: dict + @return: a dictionary with keys the ssconf names and values their + associated value + + """ + fn = "\n".join + node_names = utils.NiceSort(self._UnlockedGetNodeList()) + node_info = [self._UnlockedGetNodeInfo(name) for name in node_names] + + off_data = fn(node.name for node in node_info if node.offline) + mc_data = fn(node.name for node in node_info if node.master_candidate) + node_data = fn(node_names) + + cluster = self._config_data.cluster + return { + constants.SS_CLUSTER_NAME: cluster.cluster_name, + constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir, + constants.SS_MASTER_CANDIDATES: mc_data, + constants.SS_MASTER_IP: cluster.master_ip, + constants.SS_MASTER_NETDEV: cluster.master_netdev, + constants.SS_MASTER_NODE: cluster.master_node, + constants.SS_NODE_LIST: node_data, + constants.SS_OFFLINE_NODES: off_data, + constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION, + } + @locking.ssynchronized(_config_lock) - def InitConfig(self, node, primary_ip, secondary_ip, - hostkeypub, mac_prefix, vg_name, def_bridge): + def InitConfig(self, version, cluster_config, master_node_config): """Create the initial cluster configuration. It will contain the current node, which will also be the master - node, and no instances or operating systmes. - - Args: - node: the nodename of the initial node - primary_ip: the IP address of the current host - secondary_ip: the secondary IP of the current host or None - hostkeypub: the public hostkey of this host - - """ - hu_port = constants.FIRST_DRBD_PORT - 1 - globalconfig = objects.Cluster(serial_no=1, - rsahostkeypub=hostkeypub, - highest_used_port=hu_port, - mac_prefix=mac_prefix, - volume_group_name=vg_name, - default_bridge=def_bridge, - tcpudp_port_pool=set()) - if secondary_ip is None: - secondary_ip = primary_ip - nodeconfig = objects.Node(name=node, primary_ip=primary_ip, - secondary_ip=secondary_ip) - - self._config_data = objects.ConfigData(nodes={node: nodeconfig}, + node, and no instances. + + @type version: int + @param version: Configuration version + @type cluster_config: objects.Cluster + @param cluster_config: Cluster configuration + @type master_node_config: objects.Node + @param master_node_config: Master node configuration + + """ + nodes = { + master_node_config.name: master_node_config, + } + + self._config_data = objects.ConfigData(version=version, + cluster=cluster_config, + nodes=nodes, instances={}, - cluster=globalconfig) + serial_no=1) self._WriteConfig() @locking.ssynchronized(_config_lock, shared=1) @@ -726,7 +1115,6 @@ class ConfigWriter: """Return the volume group name. """ - self._OpenConfig() return self._config_data.cluster.volume_group_name @locking.ssynchronized(_config_lock) @@ -734,8 +1122,8 @@ class ConfigWriter: """Set the volume group name. """ - self._OpenConfig() self._config_data.cluster.volume_group_name = vg_name + self._config_data.cluster.serial_no += 1 self._WriteConfig() @locking.ssynchronized(_config_lock, shared=1) @@ -743,7 +1131,6 @@ class ConfigWriter: """Return the default bridge. """ - self._OpenConfig() return self._config_data.cluster.default_bridge @locking.ssynchronized(_config_lock, shared=1) @@ -751,19 +1138,16 @@ class ConfigWriter: """Return the mac prefix. """ - self._OpenConfig() return self._config_data.cluster.mac_prefix @locking.ssynchronized(_config_lock, shared=1) def GetClusterInfo(self): """Returns informations about the cluster - Returns: - the cluster object + @rtype: L{objects.Cluster} + @return: the cluster object """ - self._OpenConfig() - return self._config_data.cluster @locking.ssynchronized(_config_lock) @@ -776,14 +1160,20 @@ class ConfigWriter: that all modified objects will be saved, but the target argument is the one the caller wants to ensure that it's saved. + @param target: an instance of either L{objects.Cluster}, + L{objects.Node} or L{objects.Instance} which is existing in + the cluster + """ if self._config_data is None: raise errors.ProgrammerError("Configuration file not read," " cannot save.") + update_serial = False if isinstance(target, objects.Cluster): test = target == self._config_data.cluster elif isinstance(target, objects.Node): test = target in self._config_data.nodes.values() + update_serial = True elif isinstance(target, objects.Instance): test = target in self._config_data.instances.values() else: @@ -792,4 +1182,13 @@ class ConfigWriter: if not test: raise errors.ConfigurationError("Configuration updated since object" " has been read or unknown object") + target.serial_no += 1 + + if update_serial: + # for node updates, we need to increase the cluster serial too + self._config_data.cluster.serial_no += 1 + + if isinstance(target, objects.Instance): + self._UnlockedReleaseDRBDMinors(target.name) + self._WriteConfig()