X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/81a49123c7c29581eae76fb458c5f3eff43a252d..087ed2edee08da7bd3c4872cabde13c57585ca5a:/lib/config.py?ds=sidebyside diff --git a/lib/config.py b/lib/config.py index d7a1e9f..91621b9 100644 --- a/lib/config.py +++ b/lib/config.py @@ -35,6 +35,7 @@ import os import tempfile import random import logging +import time from ganeti import errors from ganeti import locking @@ -79,6 +80,7 @@ class ConfigWriter: self._cfg_file = cfg_file self._temporary_ids = set() self._temporary_drbds = {} + self._temporary_macs = set() # 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 @@ -110,11 +112,12 @@ class ConfigWriter: byte2 = random.randrange(0, 256) byte3 = random.randrange(0, 256) mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3) - if mac not in all_macs: + if mac not in all_macs and mac not in self._temporary_macs: break retries -= 1 else: raise errors.ConfigurationError("Can't generate unique MAC") + self._temporary_macs.add(mac) return mac @locking.ssynchronized(_config_lock, shared=1) @@ -126,7 +129,7 @@ class ConfigWriter: """ all_macs = self._AllMACs() - return mac in all_macs + return mac in all_macs or mac in self._temporary_macs @locking.ssynchronized(_config_lock, shared=1) def GenerateDRBDSecret(self): @@ -227,6 +230,36 @@ class ConfigWriter: return result + def _CheckDiskIDs(self, disk, l_ids, p_ids): + """Compute duplicate disk IDs + + @type disk: L{objects.Disk} + @param disk: the disk at which to start searching + @type l_ids: list + @param l_ids: list of current logical ids + @type p_ids: list + @param p_ids: list of current physical ids + @rtype: list + @return: a list of error messages + + """ + result = [] + if disk.logical_id is not None: + if disk.logical_id in l_ids: + result.append("duplicate logical id %s" % str(disk.logical_id)) + else: + l_ids.append(disk.logical_id) + if disk.physical_id is not None: + if disk.physical_id in p_ids: + result.append("duplicate physical id %s" % str(disk.physical_id)) + else: + p_ids.append(disk.physical_id) + + if disk.children: + for child in disk.children: + result.extend(self._CheckDiskIDs(child, l_ids, p_ids)) + return result + def _UnlockedVerifyConfig(self): """Verify function. @@ -239,6 +272,22 @@ class ConfigWriter: seen_macs = [] ports = {} data = self._config_data + seen_lids = [] + seen_pids = [] + + # global cluster checks + if not data.cluster.enabled_hypervisors: + result.append("enabled hypervisors list doesn't have any entries") + invalid_hvs = set(data.cluster.enabled_hypervisors) - constants.HYPER_TYPES + if invalid_hvs: + result.append("enabled hypervisors contains invalid entries: %s" % + invalid_hvs) + + if data.cluster.master_node not in data.nodes: + result.append("cluster has invalid primary node '%s'" % + data.cluster.master_node) + + # per-instance checks for instance_name in data.instances: instance = data.instances[instance_name] if instance.primary_node not in data.nodes: @@ -273,6 +322,7 @@ class ConfigWriter: for idx, disk in enumerate(instance.disks): result.extend(["instance '%s' disk %d error: %s" % (instance.name, idx, msg) for msg in disk.Verify()]) + result.extend(self._CheckDiskIDs(disk, seen_lids, seen_pids)) # cluster-wide pool of free ports for free_port in data.cluster.tcpudp_port_pool: @@ -304,6 +354,14 @@ class ConfigWriter: result.append("Not enough master candidates: actual %d, target %d" % (mc_now, mc_max)) + # node checks + for node in data.nodes.values(): + if [node.master_candidate, node.drained, node.offline].count(True) > 1: + result.append("Node %s state is invalid: master_candidate=%s," + " drain=%s, offline=%s" % + (node.name, node.master_candidate, node.drain, + node.offline)) + # drbd minors check d_map, duplicates = self._UnlockedComputeDRBDMap() for node, minor, instance_a, instance_b in duplicates: @@ -431,8 +489,8 @@ class ConfigWriter: 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)): + node_a, node_b, _, minor_a, minor_b = disk.logical_id[:5] + for node, port in ((node_a, minor_a), (node_b, minor_b)): assert node in used, ("Node '%s' of instance '%s' not found" " in node list" % (node, instance_name)) if port in used[node]: @@ -615,7 +673,7 @@ class ConfigWriter: """Get the hypervisor type for this cluster. """ - return self._config_data.cluster.default_hypervisor + return self._config_data.cluster.enabled_hypervisors[0] @locking.ssynchronized(_config_lock, shared=1) def GetHostKey(self): @@ -644,10 +702,19 @@ class ConfigWriter: all_lvs = instance.MapLVsByNode() logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs) + all_macs = self._AllMACs() + for nic in instance.nics: + if nic.mac in all_macs: + raise errors.ConfigurationError("Cannot add instance %s:" + " MAC address '%s' already in use." % (instance.name, nic.mac)) + instance.serial_no = 1 + instance.ctime = instance.mtime = time.time() self._config_data.instances[instance.name] = instance self._config_data.cluster.serial_no += 1 self._UnlockedReleaseDRBDMinors(instance.name) + for nic in instance.nics: + self._temporary_macs.discard(nic.mac) self._WriteConfig() def _SetInstanceStatus(self, instance_name, status): @@ -664,6 +731,7 @@ class ConfigWriter: if instance.admin_up != status: instance.admin_up = status instance.serial_no += 1 + instance.mtime = time.time() self._WriteConfig() @locking.ssynchronized(_config_lock) @@ -745,7 +813,7 @@ class ConfigWriter: self._config_data.instances.keys()) def _UnlockedGetInstanceInfo(self, instance_name): - """Returns informations about an instance. + """Returns information about an instance. This function is for internal use, when the config lock is already held. @@ -757,9 +825,9 @@ class ConfigWriter: @locking.ssynchronized(_config_lock, shared=1) def GetInstanceInfo(self, instance_name): - """Returns informations about an instance. + """Returns information about an instance. - It takes the information from the configuration file. Other informations of + It takes the information from the configuration file. Other information of an instance are taken from the live systems. @param instance_name: name of the instance, e.g. @@ -776,7 +844,7 @@ class ConfigWriter: """Get the configuration of all instances. @rtype: dict - @returns: dict of (instance, instance_info), where instance_info is what + @return: dict of (instance, instance_info), where instance_info is what would GetInstanceInfo return for the node """ @@ -795,6 +863,7 @@ class ConfigWriter: logging.info("Adding node %s to configuration" % node.name) node.serial_no = 1 + node.ctime = node.mtime = time.time() self._config_data.nodes[node.name] = node self._config_data.cluster.serial_no += 1 self._WriteConfig() @@ -894,16 +963,20 @@ class ConfigWriter: for node in self._UnlockedGetNodeList()]) return my_dict - def _UnlockedGetMasterCandidateStats(self): + def _UnlockedGetMasterCandidateStats(self, exceptions=None): """Get the number of current and maximum desired and possible candidates. + @type exceptions: list + @param exceptions: if passed, list of nodes that should be ignored @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: + for node in self._config_data.nodes.values(): + if exceptions and node.name in exceptions: + continue + if not (node.offline or node.drained): mc_max += 1 if node.master_candidate: mc_now += 1 @@ -911,16 +984,18 @@ class ConfigWriter: return (mc_now, mc_max) @locking.ssynchronized(_config_lock, shared=1) - def GetMasterCandidateStats(self): + def GetMasterCandidateStats(self, exceptions=None): """Get the number of current and maximum possible candidates. This is just a wrapper over L{_UnlockedGetMasterCandidateStats}. + @type exceptions: list + @param exceptions: if passed, list of nodes that should be ignored @rtype: tuple @return: tuple of (current, max) """ - return self._UnlockedGetMasterCandidateStats() + return self._UnlockedGetMasterCandidateStats(exceptions) @locking.ssynchronized(_config_lock) def MaintainCandidatePool(self): @@ -939,7 +1014,7 @@ class ConfigWriter: if mc_now >= mc_max: break node = self._config_data.nodes[name] - if node.master_candidate or node.offline: + if node.master_candidate or node.offline or node.drained: continue mod_list.append(node) node.master_candidate = True @@ -960,19 +1035,18 @@ class ConfigWriter: """ self._config_data.serial_no += 1 + self._config_data.mtime = time.time() def _OpenConfig(self): """Read the config data from disk. """ - f = open(self._cfg_file, 'r') + raw_data = utils.ReadFile(self._cfg_file) + try: - try: - data = objects.ConfigData.FromDict(serializer.Load(f.read())) - except Exception, err: - raise errors.ConfigurationError(err) - finally: - f.close() + data = objects.ConfigData.FromDict(serializer.Load(raw_data)) + except Exception, err: + raise errors.ConfigurationError(err) # Make sure the configuration has the right version _ValidateConfig(data) @@ -1015,10 +1089,12 @@ class ConfigWriter: result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file, address_list=addr_list) - for node in node_list: - if not result[node]: - logging.error("copy of file %s to node %s failed", - self._cfg_file, node) + for to_node, to_result in result.items(): + msg = to_result.fail_msg + if msg: + msg = ("Copy of file %s to node %s failed: %s" % + (self._cfg_file, to_node, msg)) + logging.error(msg) bad = True return not bad @@ -1035,16 +1111,9 @@ class ConfigWriter: destination = self._cfg_file self._BumpSerialNo() txt = serializer.Dump(self._config_data.ToDict()) - dir_name, file_name = os.path.split(destination) - fd, name = tempfile.mkstemp('.newconfig', file_name, dir_name) - f = os.fdopen(fd, 'w') - try: - f.write(txt) - os.fsync(f.fileno()) - finally: - f.close() - # we don't need to do os.close(fd) as f.close() did it - os.rename(name, destination) + + utils.WriteFile(destination, data=txt) + self.write_count += 1 # and redistribute the config file to master candidates @@ -1053,8 +1122,14 @@ class ConfigWriter: # 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()) + result = rpc.RpcRunner.call_write_ssconf_files(\ + self._UnlockedGetNodeList(), + self._UnlockedGetSsconfValues()) + for nname, nresu in result.items(): + msg = nresu.fail_msg + if msg: + logging.warning("Error while uploading ssconf files to" + " node %s: %s", nname, msg) self._last_cluster_serial = self._config_data.cluster.serial_no def _UnlockedGetSsconfValues(self): @@ -1069,54 +1144,41 @@ class ConfigWriter: instance_names = utils.NiceSort(self._UnlockedGetInstanceList()) node_names = utils.NiceSort(self._UnlockedGetNodeList()) node_info = [self._UnlockedGetNodeInfo(name) for name in node_names] + node_pri_ips = ["%s %s" % (ninfo.name, ninfo.primary_ip) + for ninfo in node_info] + node_snd_ips = ["%s %s" % (ninfo.name, ninfo.secondary_ip) + for ninfo in node_info] instance_data = fn(instance_names) off_data = fn(node.name for node in node_info if node.offline) on_data = fn(node.name for node in node_info if not node.offline) mc_data = fn(node.name for node in node_info if node.master_candidate) + mc_ips_data = fn(node.primary_ip for node in node_info + if node.master_candidate) node_data = fn(node_names) + node_pri_ips_data = fn(node_pri_ips) + node_snd_ips_data = fn(node_snd_ips) cluster = self._config_data.cluster + cluster_tags = fn(cluster.GetTags()) return { constants.SS_CLUSTER_NAME: cluster.cluster_name, + constants.SS_CLUSTER_TAGS: cluster_tags, constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir, constants.SS_MASTER_CANDIDATES: mc_data, + constants.SS_MASTER_CANDIDATES_IPS: mc_ips_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_NODE_PRIMARY_IPS: node_pri_ips_data, + constants.SS_NODE_SECONDARY_IPS: node_snd_ips_data, constants.SS_OFFLINE_NODES: off_data, constants.SS_ONLINE_NODES: on_data, constants.SS_INSTANCE_LIST: instance_data, constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION, } - @locking.ssynchronized(_config_lock) - 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. - - @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={}, - serial_no=1) - self._WriteConfig() - @locking.ssynchronized(_config_lock, shared=1) def GetVGName(self): """Return the volume group name. @@ -1134,13 +1196,6 @@ class ConfigWriter: self._WriteConfig() @locking.ssynchronized(_config_lock, shared=1) - def GetDefBridge(self): - """Return the default bridge. - - """ - return self._config_data.cluster.default_bridge - - @locking.ssynchronized(_config_lock, shared=1) def GetMACPrefix(self): """Return the mac prefix. @@ -1149,7 +1204,7 @@ class ConfigWriter: @locking.ssynchronized(_config_lock, shared=1) def GetClusterInfo(self): - """Returns informations about the cluster + """Returns information about the cluster @rtype: L{objects.Cluster} @return: the cluster object @@ -1190,12 +1245,16 @@ class ConfigWriter: raise errors.ConfigurationError("Configuration updated since object" " has been read or unknown object") target.serial_no += 1 + target.mtime = now = time.time() if update_serial: # for node updates, we need to increase the cluster serial too self._config_data.cluster.serial_no += 1 + self._config_data.cluster.mtime = now if isinstance(target, objects.Instance): self._UnlockedReleaseDRBDMinors(target.name) + for nic in target.nics: + self._temporary_macs.discard(nic.mac) self._WriteConfig()