X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/168c1de23b6a8b62ebd5933d3df73810c2ae5ee7..c4929a8bcca4a43dc6434394a91a8ea67d854844:/lib/rpc.py diff --git a/lib/rpc.py b/lib/rpc.py index 6de365b..0bedb3c 100644 --- a/lib/rpc.py +++ b/lib/rpc.py @@ -1,7 +1,7 @@ # # -# Copyright (C) 2006, 2007 Google Inc. +# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 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 @@ -23,7 +23,7 @@ """ -# pylint: disable-msg=C0103,R0201,R0904 +# pylint: disable=C0103,R0201,R0904 # C0103: Invalid name, since call_ are not valid # R0201: Method could be a function, we keep all rpcs instance methods # as not to change them back and forth between static/instance methods @@ -34,6 +34,8 @@ import os import logging import zlib import base64 +import pycurl +import threading from ganeti import utils from ganeti import objects @@ -41,41 +43,116 @@ from ganeti import http from ganeti import serializer from ganeti import constants from ganeti import errors +from ganeti import netutils +from ganeti import ssconf +from ganeti import runtime +from ganeti import compat +from ganeti import rpc_defs + +# Special module generated at build time +from ganeti import _generated_rpc # pylint has a bug here, doesn't see this import -import ganeti.http.client # pylint: disable-msg=W0611 +import ganeti.http.client # pylint: disable=W0611 + + +# Timeout for connecting to nodes (seconds) +_RPC_CONNECT_TIMEOUT = 5 +_RPC_CLIENT_HEADERS = [ + "Content-type: %s" % http.HTTP_APP_JSON, + "Expect:", + ] -# Module level variable -_http_manager = None +# Various time constants for the timeout table +_TMO_URGENT = 60 # one minute +_TMO_FAST = 5 * 60 # five minutes +_TMO_NORMAL = 15 * 60 # 15 minutes +_TMO_SLOW = 3600 # one hour +_TMO_4HRS = 4 * 3600 +_TMO_1DAY = 86400 + +#: Special value to describe an offline host +_OFFLINE = object() def Init(): """Initializes the module-global HTTP client manager. - Must be called before using any RPC function. + Must be called before using any RPC function and while exactly one thread is + running. """ - global _http_manager # pylint: disable-msg=W0603 - - assert not _http_manager, "RPC module initialized more than once" + # curl_global_init(3) and curl_global_cleanup(3) must be called with only + # one thread running. This check is just a safety measure -- it doesn't + # cover all cases. + assert threading.activeCount() == 1, \ + "Found more than one active thread when initializing pycURL" - http.InitSsl() + logging.info("Using PycURL %s", pycurl.version) - _http_manager = http.client.HttpClientManager() + pycurl.global_init(pycurl.GLOBAL_ALL) def Shutdown(): """Stops the module-global HTTP client manager. - Must be called before quitting the program. + Must be called before quitting the program and while exactly one thread is + running. + + """ + pycurl.global_cleanup() + + +def _ConfigRpcCurl(curl): + noded_cert = str(constants.NODED_CERT_FILE) + + curl.setopt(pycurl.FOLLOWLOCATION, False) + curl.setopt(pycurl.CAINFO, noded_cert) + curl.setopt(pycurl.SSL_VERIFYHOST, 0) + curl.setopt(pycurl.SSL_VERIFYPEER, True) + curl.setopt(pycurl.SSLCERTTYPE, "PEM") + curl.setopt(pycurl.SSLCERT, noded_cert) + curl.setopt(pycurl.SSLKEYTYPE, "PEM") + curl.setopt(pycurl.SSLKEY, noded_cert) + curl.setopt(pycurl.CONNECTTIMEOUT, _RPC_CONNECT_TIMEOUT) + + +def RunWithRPC(fn): + """RPC-wrapper decorator. + + When applied to a function, it runs it with the RPC system + initialized, and it shutsdown the system afterwards. This means the + function must be called without RPC being initialized. """ - global _http_manager # pylint: disable-msg=W0603 + def wrapper(*args, **kwargs): + Init() + try: + return fn(*args, **kwargs) + finally: + Shutdown() + return wrapper + - if _http_manager: - _http_manager.Shutdown() - _http_manager = None +def _Compress(data): + """Compresses a string for transport over RPC. + + Small amounts of data are not compressed. + + @type data: str + @param data: Data + @rtype: tuple + @return: Encoded data to send + + """ + # Small amounts of data are not compressed + if len(data) < 512: + return (constants.RPC_ENCODING_NONE, data) + + # Compress with zlib and encode in base64 + return (constants.RPC_ENCODING_ZLIB_BASE64, + base64.b64encode(zlib.compress(data, 3))) class RpcResult(object): @@ -125,12 +202,9 @@ class RpcResult(object): self.fail_msg = None self.payload = data[1] - assert hasattr(self, "call") - assert hasattr(self, "data") - assert hasattr(self, "fail_msg") - assert hasattr(self, "node") - assert hasattr(self, "offline") - assert hasattr(self, "payload") + for attr_name in ["call", "data", "fail_msg", + "node", "offline", "payload"]: + assert hasattr(self, attr_name), "Missing attribute %s" % attr_name @staticmethod def _EnsureErr(val): @@ -160,1032 +234,449 @@ class RpcResult(object): else: ec = errors.OpExecError if ecode is not None: - args = (msg, prereq) + args = (msg, ecode) else: args = (msg, ) - raise ec(*args) # pylint: disable-msg=W0142 - + raise ec(*args) # pylint: disable=W0142 -class Client: - """RPC Client class. - This class, given a (remote) method name, a list of parameters and a - list of nodes, will contact (in parallel) all nodes, and return a - dict of results (key: node name, value: result). +def _SsconfResolver(node_list, + ssc=ssconf.SimpleStore, + nslookup_fn=netutils.Hostname.GetIP): + """Return addresses for given node names. - One current bug is that generic failure is still signaled by - 'False' result, which is not good. This overloading of values can - cause bugs. + @type node_list: list + @param node_list: List of node names + @type ssc: class + @param ssc: SimpleStore class that is used to obtain node->ip mappings + @type nslookup_fn: callable + @param nslookup_fn: function use to do NS lookup + @rtype: list of tuple; (string, string) + @return: List of tuples containing node name and IP address """ - def __init__(self, procedure, body, port): - self.procedure = procedure - self.body = body - self.port = port - self.nc = {} - - self._ssl_params = \ - http.HttpSslParams(ssl_key_path=constants.NODED_CERT_FILE, - ssl_cert_path=constants.NODED_CERT_FILE) - - def ConnectList(self, node_list, address_list=None): - """Add a list of nodes to the target nodes. + ss = ssc() + iplist = ss.GetNodePrimaryIPList() + family = ss.GetPrimaryIPFamily() + ipmap = dict(entry.split() for entry in iplist) - @type node_list: list - @param node_list: the list of node names to connect - @type address_list: list or None - @keyword address_list: either None or a list with node addresses, - which must have the same length as the node list + result = [] + for node in node_list: + ip = ipmap.get(node) + if ip is None: + ip = nslookup_fn(node, family=family) + result.append((node, ip)) - """ - if address_list is None: - address_list = [None for _ in node_list] - else: - assert len(node_list) == len(address_list), \ - "Name and address lists should have the same length" - for node, address in zip(node_list, address_list): - self.ConnectNode(node, address) + return result - def ConnectNode(self, name, address=None): - """Add a node to the target list. - @type name: str - @param name: the node name - @type address: str - @keyword address: the node address, if known +class _StaticResolver: + def __init__(self, addresses): + """Initializes this class. """ - if address is None: - address = name + self._addresses = addresses - self.nc[name] = \ - http.client.HttpClientRequest(address, self.port, http.HTTP_PUT, - "/%s" % self.procedure, - post_data=self.body, - ssl_params=self._ssl_params, - ssl_verify_peer=True) - - def GetResults(self): - """Call nodes and return results. - - @rtype: list - @return: List of RPC results + def __call__(self, hosts): + """Returns static addresses for hosts. """ - assert _http_manager, "RPC module not initialized" + assert len(hosts) == len(self._addresses) + return zip(hosts, self._addresses) - _http_manager.ExecRequests(self.nc.values()) - results = {} +def _CheckConfigNode(name, node): + """Checks if a node is online. - for name, req in self.nc.iteritems(): - if req.success and req.resp_status_code == http.HTTP_OK: - results[name] = RpcResult(data=serializer.LoadJson(req.resp_body), - node=name, call=self.procedure) - continue + @type name: string + @param name: Node name + @type node: L{objects.Node} or None + @param node: Node object - # TODO: Better error reporting - if req.error: - msg = req.error - else: - msg = req.resp_body + """ + if node is None: + # Depend on DNS for name resolution + ip = name + elif node.offline: + ip = _OFFLINE + else: + ip = node.primary_ip + return (name, ip) - logging.error("RPC error in %s from node %s: %s", - self.procedure, name, msg) - results[name] = RpcResult(data=msg, failed=True, node=name, - call=self.procedure) - return results +def _NodeConfigResolver(single_node_fn, all_nodes_fn, hosts): + """Calculate node addresses using configuration. + """ + # Special case for single-host lookups + if len(hosts) == 1: + (name, ) = hosts + return [_CheckConfigNode(name, single_node_fn(name))] + else: + all_nodes = all_nodes_fn() + return [_CheckConfigNode(name, all_nodes.get(name, None)) + for name in hosts] -class RpcRunner(object): - """RPC runner class""" - def __init__(self, cfg): - """Initialized the rpc runner. +class _RpcProcessor: + def __init__(self, resolver, port, lock_monitor_cb=None): + """Initializes this class. - @type cfg: C{config.ConfigWriter} - @param cfg: the configuration object that will be used to get data - about the cluster + @param resolver: callable accepting a list of hostnames, returning a list + of tuples containing name and IP address (IP address can be the name or + the special value L{_OFFLINE} to mark offline machines) + @type port: int + @param port: TCP port + @param lock_monitor_cb: Callable for registering with lock monitor """ - self._cfg = cfg - self.port = utils.GetDaemonPort(constants.NODED) + self._resolver = resolver + self._port = port + self._lock_monitor_cb = lock_monitor_cb - def _InstDict(self, instance, hvp=None, bep=None): - """Convert the given instance to a dict. - - This is done via the instance's ToDict() method and additionally - we fill the hvparams with the cluster defaults. - - @type instance: L{objects.Instance} - @param instance: an Instance object - @type hvp: dict or None - @param hvp: a dictionary with overridden hypervisor parameters - @type bep: dict or None - @param bep: a dictionary with overridden backend parameters - @rtype: dict - @return: the instance dict, with the hvparams filled with the - cluster defaults + @staticmethod + def _PrepareRequests(hosts, port, procedure, body, read_timeout): + """Prepares requests by sorting offline hosts into separate list. """ - idict = instance.ToDict() - cluster = self._cfg.GetClusterInfo() - idict["hvparams"] = cluster.FillHV(instance) - if hvp is not None: - idict["hvparams"].update(hvp) - idict["beparams"] = cluster.FillBE(instance) - if bep is not None: - idict["beparams"].update(bep) - for nic in idict["nics"]: - nic['nicparams'] = objects.FillDict( - cluster.nicparams[constants.PP_DEFAULT], - nic['nicparams']) - return idict - - def _ConnectList(self, client, node_list, call): - """Helper for computing node addresses. - - @type client: L{ganeti.rpc.Client} - @param client: a C{Client} instance - @type node_list: list - @param node_list: the node list we should connect - @type call: string - @param call: the name of the remote procedure call, for filling in - correctly any eventual offline nodes' results + results = {} + requests = {} - """ - all_nodes = self._cfg.GetAllNodesInfo() - name_list = [] - addr_list = [] - skip_dict = {} - for node in node_list: - if node in all_nodes: - if all_nodes[node].offline: - skip_dict[node] = RpcResult(node=node, offline=True, call=call) - continue - val = all_nodes[node].primary_ip + for (name, ip) in hosts: + if ip is _OFFLINE: + # Node is marked as offline + results[name] = RpcResult(node=name, offline=True, call=procedure) else: - val = None - addr_list.append(val) - name_list.append(node) - if name_list: - client.ConnectList(name_list, address_list=addr_list) - return skip_dict - - def _ConnectNode(self, client, node, call): - """Helper for computing one node's address. - - @type client: L{ganeti.rpc.Client} - @param client: a C{Client} instance - @type node: str - @param node: the node we should connect - @type call: string - @param call: the name of the remote procedure call, for filling in - correctly any eventual offline nodes' results + requests[name] = \ + http.client.HttpClientRequest(str(ip), port, + http.HTTP_PUT, str("/%s" % procedure), + headers=_RPC_CLIENT_HEADERS, + post_data=body, + read_timeout=read_timeout, + nicename="%s/%s" % (name, procedure), + curl_config_fn=_ConfigRpcCurl) - """ - node_info = self._cfg.GetNodeInfo(node) - if node_info is not None: - if node_info.offline: - return RpcResult(node=node, offline=True, call=call) - addr = node_info.primary_ip - else: - addr = None - client.ConnectNode(node, address=addr) - - def _MultiNodeCall(self, node_list, procedure, args): - """Helper for making a multi-node call - - """ - body = serializer.DumpJson(args, indent=False) - c = Client(procedure, body, self.port) - skip_dict = self._ConnectList(c, node_list, procedure) - skip_dict.update(c.GetResults()) - return skip_dict - - @classmethod - def _StaticMultiNodeCall(cls, node_list, procedure, args, - address_list=None): - """Helper for making a multi-node static call - - """ - body = serializer.DumpJson(args, indent=False) - c = Client(procedure, body, utils.GetDaemonPort(constants.NODED)) - c.ConnectList(node_list, address_list=address_list) - return c.GetResults() - - def _SingleNodeCall(self, node, procedure, args): - """Helper for making a single-node call - - """ - body = serializer.DumpJson(args, indent=False) - c = Client(procedure, body, self.port) - result = self._ConnectNode(c, node, procedure) - if result is None: - # we did connect, node is not offline - result = c.GetResults()[node] - return result - - @classmethod - def _StaticSingleNodeCall(cls, node, procedure, args): - """Helper for making a single-node static call - - """ - body = serializer.DumpJson(args, indent=False) - c = Client(procedure, body, utils.GetDaemonPort(constants.NODED)) - c.ConnectNode(node) - return c.GetResults()[node] + return (results, requests) @staticmethod - def _Compress(data): - """Compresses a string for transport over RPC. - - Small amounts of data are not compressed. - - @type data: str - @param data: Data - @rtype: tuple - @return: Encoded data to send - - """ - # Small amounts of data are not compressed - if len(data) < 512: - return (constants.RPC_ENCODING_NONE, data) - - # Compress with zlib and encode in base64 - return (constants.RPC_ENCODING_ZLIB_BASE64, - base64.b64encode(zlib.compress(data, 3))) - - # - # Begin RPC calls - # - - def call_lv_list(self, node_list, vg_name): - """Gets the logical volumes present in a given volume group. - - This is a multi-node call. - - """ - return self._MultiNodeCall(node_list, "lv_list", [vg_name]) - - def call_vg_list(self, node_list): - """Gets the volume group list. - - This is a multi-node call. - - """ - return self._MultiNodeCall(node_list, "vg_list", []) - - def call_storage_list(self, node_list, su_name, su_args, name, fields): - """Get list of storage units. - - This is a multi-node call. - - """ - return self._MultiNodeCall(node_list, "storage_list", - [su_name, su_args, name, fields]) - - def call_storage_modify(self, node, su_name, su_args, name, changes): - """Modify a storage unit. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "storage_modify", - [su_name, su_args, name, changes]) - - def call_storage_execute(self, node, su_name, su_args, name, op): - """Executes an operation on a storage unit. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "storage_execute", - [su_name, su_args, name, op]) - - def call_bridges_exist(self, node, bridges_list): - """Checks if a node has all the bridges given. - - This method checks if all bridges given in the bridges_list are - present on the remote node, so that an instance that uses interfaces - on those bridges can be started. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "bridges_exist", [bridges_list]) - - def call_instance_start(self, node, instance, hvp, bep): - """Starts an instance. - - This is a single-node call. - - """ - idict = self._InstDict(instance, hvp=hvp, bep=bep) - return self._SingleNodeCall(node, "instance_start", [idict]) - - def call_instance_shutdown(self, node, instance, timeout): - """Stops an instance. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "instance_shutdown", - [self._InstDict(instance), timeout]) - - def call_migration_info(self, node, instance): - """Gather the information necessary to prepare an instance migration. - - This is a single-node call. - - @type node: string - @param node: the node on which the instance is currently running - @type instance: C{objects.Instance} - @param instance: the instance definition - - """ - return self._SingleNodeCall(node, "migration_info", - [self._InstDict(instance)]) - - def call_accept_instance(self, node, instance, info, target): - """Prepare a node to accept an instance. - - This is a single-node call. - - @type node: string - @param node: the target node for the migration - @type instance: C{objects.Instance} - @param instance: the instance definition - @type info: opaque/hypervisor specific (string/data) - @param info: result for the call_migration_info call - @type target: string - @param target: target hostname (usually ip address) (on the node itself) - - """ - return self._SingleNodeCall(node, "accept_instance", - [self._InstDict(instance), info, target]) - - def call_finalize_migration(self, node, instance, info, success): - """Finalize any target-node migration specific operation. - - This is called both in case of a successful migration and in case of error - (in which case it should abort the migration). - - This is a single-node call. - - @type node: string - @param node: the target node for the migration - @type instance: C{objects.Instance} - @param instance: the instance definition - @type info: opaque/hypervisor specific (string/data) - @param info: result for the call_migration_info call - @type success: boolean - @param success: whether the migration was a success or a failure - - """ - return self._SingleNodeCall(node, "finalize_migration", - [self._InstDict(instance), info, success]) - - def call_instance_migrate(self, node, instance, target, live): - """Migrate an instance. - - This is a single-node call. - - @type node: string - @param node: the node on which the instance is currently running - @type instance: C{objects.Instance} - @param instance: the instance definition - @type target: string - @param target: the target node name - @type live: boolean - @param live: whether the migration should be done live or not (the - interpretation of this parameter is left to the hypervisor) - - """ - return self._SingleNodeCall(node, "instance_migrate", - [self._InstDict(instance), target, live]) - - def call_instance_reboot(self, node, inst, reboot_type, shutdown_timeout): - """Reboots an instance. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "instance_reboot", - [self._InstDict(inst), reboot_type, - shutdown_timeout]) - - def call_instance_os_add(self, node, inst, reinstall, debug): - """Installs an OS on the given instance. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "instance_os_add", - [self._InstDict(inst), reinstall, debug]) - - def call_instance_run_rename(self, node, inst, old_name, debug): - """Run the OS rename script for an instance. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "instance_run_rename", - [self._InstDict(inst), old_name, debug]) - - def call_instance_info(self, node, instance, hname): - """Returns information about a single instance. - - This is a single-node call. - - @type node: list - @param node: the list of nodes to query - @type instance: string - @param instance: the instance name - @type hname: string - @param hname: the hypervisor type of the instance - - """ - return self._SingleNodeCall(node, "instance_info", [instance, hname]) - - def call_instance_migratable(self, node, instance): - """Checks whether the given instance can be migrated. - - This is a single-node call. - - @param node: the node to query - @type instance: L{objects.Instance} - @param instance: the instance to check - - - """ - return self._SingleNodeCall(node, "instance_migratable", - [self._InstDict(instance)]) - - def call_all_instances_info(self, node_list, hypervisor_list): - """Returns information about all instances on the given nodes. - - This is a multi-node call. - - @type node_list: list - @param node_list: the list of nodes to query - @type hypervisor_list: list - @param hypervisor_list: the hypervisors to query for instances - - """ - return self._MultiNodeCall(node_list, "all_instances_info", - [hypervisor_list]) - - def call_instance_list(self, node_list, hypervisor_list): - """Returns the list of running instances on a given node. - - This is a multi-node call. - - @type node_list: list - @param node_list: the list of nodes to query - @type hypervisor_list: list - @param hypervisor_list: the hypervisors to query for instances - - """ - return self._MultiNodeCall(node_list, "instance_list", [hypervisor_list]) - - def call_node_tcp_ping(self, node, source, target, port, timeout, - live_port_needed): - """Do a TcpPing on the remote node - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "node_tcp_ping", - [source, target, port, timeout, - live_port_needed]) - - def call_node_has_ip_address(self, node, address): - """Checks if a node has the given IP address. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "node_has_ip_address", [address]) - - def call_node_info(self, node_list, vg_name, hypervisor_type): - """Return node information. - - This will return memory information and volume group size and free - space. - - This is a multi-node call. - - @type node_list: list - @param node_list: the list of nodes to query - @type vg_name: C{string} - @param vg_name: the name of the volume group to ask for disk space - information - @type hypervisor_type: C{str} - @param hypervisor_type: the name of the hypervisor to ask for - memory information - - """ - return self._MultiNodeCall(node_list, "node_info", - [vg_name, hypervisor_type]) - - def call_node_add(self, node, dsa, dsapub, rsa, rsapub, ssh, sshpub): - """Add a node to the cluster. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "node_add", - [dsa, dsapub, rsa, rsapub, ssh, sshpub]) - - def call_node_verify(self, node_list, checkdict, cluster_name): - """Request verification of given parameters. - - This is a multi-node call. - - """ - return self._MultiNodeCall(node_list, "node_verify", - [checkdict, cluster_name]) - - @classmethod - def call_node_start_master(cls, node, start_daemons, no_voting): - """Tells a node to activate itself as a master. - - This is a single-node call. - - """ - return cls._StaticSingleNodeCall(node, "node_start_master", - [start_daemons, no_voting]) - - @classmethod - def call_node_stop_master(cls, node, stop_daemons): - """Tells a node to demote itself from master status. - - This is a single-node call. - - """ - return cls._StaticSingleNodeCall(node, "node_stop_master", [stop_daemons]) - - @classmethod - def call_master_info(cls, node_list): - """Query master info. - - This is a multi-node call. - - """ - # TODO: should this method query down nodes? - return cls._StaticMultiNodeCall(node_list, "master_info", []) - - @classmethod - def call_version(cls, node_list): - """Query node version. - - This is a multi-node call. - - """ - return cls._StaticMultiNodeCall(node_list, "version", []) - - def call_blockdev_create(self, node, bdev, size, owner, on_primary, info): - """Request creation of a given block device. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "blockdev_create", - [bdev.ToDict(), size, owner, on_primary, info]) - - def call_blockdev_remove(self, node, bdev): - """Request removal of a given block device. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "blockdev_remove", [bdev.ToDict()]) - - def call_blockdev_rename(self, node, devlist): - """Request rename of the given block devices. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "blockdev_rename", - [(d.ToDict(), uid) for d, uid in devlist]) - - def call_blockdev_assemble(self, node, disk, owner, on_primary): - """Request assembling of a given block device. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "blockdev_assemble", - [disk.ToDict(), owner, on_primary]) - - def call_blockdev_shutdown(self, node, disk): - """Request shutdown of a given block device. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "blockdev_shutdown", [disk.ToDict()]) - - def call_blockdev_addchildren(self, node, bdev, ndevs): - """Request adding a list of children to a (mirroring) device. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "blockdev_addchildren", - [bdev.ToDict(), - [disk.ToDict() for disk in ndevs]]) - - def call_blockdev_removechildren(self, node, bdev, ndevs): - """Request removing a list of children from a (mirroring) device. - - This is a single-node call. - - """ - return self._SingleNodeCall(node, "blockdev_removechildren", - [bdev.ToDict(), - [disk.ToDict() for disk in ndevs]]) - - def call_blockdev_getmirrorstatus(self, node, disks): - """Request status of a (mirroring) device. - - This is a single-node call. - - """ - result = self._SingleNodeCall(node, "blockdev_getmirrorstatus", - [dsk.ToDict() for dsk in disks]) - if not result.fail_msg: - result.payload = [objects.BlockDevStatus.FromDict(i) - for i in result.payload] - return result - - def call_blockdev_find(self, node, disk): - """Request identification of a given block device. - - This is a single-node call. - - """ - result = self._SingleNodeCall(node, "blockdev_find", [disk.ToDict()]) - if not result.fail_msg and result.payload is not None: - result.payload = objects.BlockDevStatus.FromDict(result.payload) - return result - - def call_blockdev_close(self, node, instance_name, disks): - """Closes the given block devices. - - This is a single-node call. - - """ - params = [instance_name, [cf.ToDict() for cf in disks]] - return self._SingleNodeCall(node, "blockdev_close", params) - - def call_blockdev_getsizes(self, node, disks): - """Returns the size of the given disks. - - This is a single-node call. - - """ - params = [[cf.ToDict() for cf in disks]] - return self._SingleNodeCall(node, "blockdev_getsize", params) - - def call_drbd_disconnect_net(self, node_list, nodes_ip, disks): - """Disconnects the network of the given drbd devices. - - This is a multi-node call. - - """ - return self._MultiNodeCall(node_list, "drbd_disconnect_net", - [nodes_ip, [cf.ToDict() for cf in disks]]) - - def call_drbd_attach_net(self, node_list, nodes_ip, - disks, instance_name, multimaster): - """Disconnects the given drbd devices. - - This is a multi-node call. + def _CombineResults(results, requests, procedure): + """Combines pre-computed results for offline hosts with actual call results. """ - return self._MultiNodeCall(node_list, "drbd_attach_net", - [nodes_ip, [cf.ToDict() for cf in disks], - instance_name, multimaster]) - - def call_drbd_wait_sync(self, node_list, nodes_ip, disks): - """Waits for the synchronization of drbd devices is complete. - - This is a multi-node call. + for name, req in requests.items(): + if req.success and req.resp_status_code == http.HTTP_OK: + host_result = RpcResult(data=serializer.LoadJson(req.resp_body), + node=name, call=procedure) + else: + # TODO: Better error reporting + if req.error: + msg = req.error + else: + msg = req.resp_body - """ - return self._MultiNodeCall(node_list, "drbd_wait_sync", - [nodes_ip, [cf.ToDict() for cf in disks]]) + logging.error("RPC error in %s on node %s: %s", procedure, name, msg) + host_result = RpcResult(data=msg, failed=True, node=name, + call=procedure) - @classmethod - def call_upload_file(cls, node_list, file_name, address_list=None): - """Upload a file. + results[name] = host_result - The node will refuse the operation in case the file is not on the - approved file list. + return results - This is a multi-node call. + def __call__(self, hosts, procedure, body, read_timeout=None, + _req_process_fn=http.client.ProcessRequests): + """Makes an RPC request to a number of nodes. - @type node_list: list - @param node_list: the list of node names to upload to - @type file_name: str - @param file_name: the filename to upload - @type address_list: list or None - @keyword address_list: an optional list of node addresses, in order - to optimize the RPC speed + @type hosts: sequence + @param hosts: Hostnames + @type procedure: string + @param procedure: Request path + @type body: string + @param body: Request body + @type read_timeout: int or None + @param read_timeout: Read timeout for request """ - file_contents = utils.ReadFile(file_name) - data = cls._Compress(file_contents) - st = os.stat(file_name) - params = [file_name, data, st.st_mode, st.st_uid, st.st_gid, - st.st_atime, st.st_mtime] - return cls._StaticMultiNodeCall(node_list, "upload_file", params, - address_list=address_list) + assert read_timeout is not None, \ + "Missing RPC read timeout for procedure '%s'" % procedure - @classmethod - def call_write_ssconf_files(cls, node_list, values): - """Write ssconf files. + (results, requests) = \ + self._PrepareRequests(self._resolver(hosts), self._port, procedure, + str(body), read_timeout) - This is a multi-node call. + _req_process_fn(requests.values(), lock_monitor_cb=self._lock_monitor_cb) - """ - return cls._StaticMultiNodeCall(node_list, "write_ssconf_files", [values]) - - def call_os_diagnose(self, node_list): - """Request a diagnose of OS definitions. + assert not frozenset(results).intersection(requests) - This is a multi-node call. + return self._CombineResults(results, requests, procedure) - """ - return self._MultiNodeCall(node_list, "os_diagnose", []) - def call_os_get(self, node, name): - """Returns an OS definition. - - This is a single-node call. +class _RpcClientBase: + def __init__(self, resolver, encoder_fn, lock_monitor_cb=None): + """Initializes this class. """ - result = self._SingleNodeCall(node, "os_get", [name]) - if not result.fail_msg and isinstance(result.payload, dict): - result.payload = objects.OS.FromDict(result.payload) - return result - - def call_hooks_runner(self, node_list, hpath, phase, env): - """Call the hooks runner. + self._proc = _RpcProcessor(resolver, + netutils.GetDaemonPort(constants.NODED), + lock_monitor_cb=lock_monitor_cb) + self._encoder = compat.partial(self._EncodeArg, encoder_fn) - Args: - - op: the OpCode instance - - env: a dictionary with the environment - - This is a multi-node call. + @staticmethod + def _EncodeArg(encoder_fn, (argkind, value)): + """Encode argument. """ - params = [hpath, phase, env] - return self._MultiNodeCall(node_list, "hooks_runner", params) - - def call_iallocator_runner(self, node, name, idata): - """Call an iallocator on a remote node - - Args: - - name: the iallocator name - - input: the json-encoded input string + if argkind is None: + return value + else: + return encoder_fn(argkind)(value) - This is a single-node call. + def _Call(self, cdef, node_list, args): + """Entry point for automatically generated RPC wrappers. """ - return self._SingleNodeCall(node, "iallocator_runner", [name, idata]) - - def call_blockdev_grow(self, node, cf_bdev, amount): - """Request a snapshot of the given block device. + (procedure, _, timeout, argdefs, postproc_fn, _) = cdef - This is a single-node call. + if callable(timeout): + read_timeout = timeout(args) + else: + read_timeout = timeout - """ - return self._SingleNodeCall(node, "blockdev_grow", - [cf_bdev.ToDict(), amount]) + body = serializer.DumpJson(map(self._encoder, + zip(map(compat.snd, argdefs), args)), + indent=False) - def call_blockdev_export(self, node, cf_bdev, - dest_node, dest_path, cluster_name): - """Export a given disk to another node. + result = self._proc(node_list, procedure, body, read_timeout=read_timeout) - This is a single-node call. + if postproc_fn: + return dict(map(lambda (key, value): (key, postproc_fn(value)), + result.items())) + else: + return result - """ - return self._SingleNodeCall(node, "blockdev_export", - [cf_bdev.ToDict(), dest_node, dest_path, - cluster_name]) - def call_blockdev_snapshot(self, node, cf_bdev): - """Request a snapshot of the given block device. +def _ObjectToDict(value): + """Converts an object to a dictionary. - This is a single-node call. + @note: See L{objects}. - """ - return self._SingleNodeCall(node, "blockdev_snapshot", [cf_bdev.ToDict()]) - - def call_snapshot_export(self, node, snap_bdev, dest_node, instance, - cluster_name, idx, debug): - """Request the export of a given snapshot. + """ + return value.ToDict() - This is a single-node call. - """ - return self._SingleNodeCall(node, "snapshot_export", - [snap_bdev.ToDict(), dest_node, - self._InstDict(instance), cluster_name, - idx, debug]) +def _ObjectListToDict(value): + """Converts a list of L{objects} to dictionaries. - def call_finalize_export(self, node, instance, snap_disks): - """Request the completion of an export operation. + """ + return map(_ObjectToDict, value) - This writes the export config file, etc. - This is a single-node call. +def _EncodeNodeToDiskDict(value): + """Encodes a dictionary with node name as key and disk objects as values. - """ - flat_disks = [] - for disk in snap_disks: - if isinstance(disk, bool): - flat_disks.append(disk) - else: - flat_disks.append(disk.ToDict()) + """ + return dict((name, _ObjectListToDict(disks)) + for name, disks in value.items()) - return self._SingleNodeCall(node, "finalize_export", - [self._InstDict(instance), flat_disks]) - def call_export_info(self, node, path): - """Queries the export information in a given path. +def _PrepareFileUpload(filename): + """Loads a file and prepares it for an upload to nodes. - This is a single-node call. + """ + data = _Compress(utils.ReadFile(filename)) + st = os.stat(filename) + getents = runtime.GetEnts() + return [filename, data, st.st_mode, getents.LookupUid(st.st_uid), + getents.LookupGid(st.st_gid), st.st_atime, st.st_mtime] - """ - return self._SingleNodeCall(node, "export_info", [path]) - def call_instance_os_import(self, node, inst, src_node, src_images, - cluster_name, debug): - """Request the import of a backup into an instance. +def _PrepareFinalizeExportDisks(snap_disks): + """Encodes disks for finalizing export. - This is a single-node call. + """ + flat_disks = [] - """ - return self._SingleNodeCall(node, "instance_os_import", - [self._InstDict(inst), src_node, src_images, - cluster_name, debug]) + for disk in snap_disks: + if isinstance(disk, bool): + flat_disks.append(disk) + else: + flat_disks.append(disk.ToDict()) - def call_export_list(self, node_list): - """Gets the stored exports list. + return flat_disks - This is a multi-node call. - """ - return self._MultiNodeCall(node_list, "export_list", []) +def _EncodeImportExportIO((ieio, ieioargs)): + """Encodes import/export I/O information. - def call_export_remove(self, node, export): - """Requests removal of a given export. + """ + if ieio == constants.IEIO_RAW_DISK: + assert len(ieioargs) == 1 + return (ieio, (ieioargs[0].ToDict(), )) - This is a single-node call. + if ieio == constants.IEIO_SCRIPT: + assert len(ieioargs) == 2 + return (ieio, (ieioargs[0].ToDict(), ieioargs[1])) - """ - return self._SingleNodeCall(node, "export_remove", [export]) + return (ieio, ieioargs) - @classmethod - def call_node_leave_cluster(cls, node, modify_ssh_setup): - """Requests a node to clean the cluster information it has. - This will remove the configuration information from the ganeti data - dir. +def _EncodeBlockdevRename(value): + """Encodes information for renaming block devices. - This is a single-node call. + """ + return [(d.ToDict(), uid) for d, uid in value] - """ - return cls._StaticSingleNodeCall(node, "node_leave_cluster", - [modify_ssh_setup]) - def call_node_volumes(self, node_list): - """Gets all volumes on node(s). +#: Generic encoders +_ENCODERS = { + rpc_defs.ED_OBJECT_DICT: _ObjectToDict, + rpc_defs.ED_OBJECT_DICT_LIST: _ObjectListToDict, + rpc_defs.ED_NODE_TO_DISK_DICT: _EncodeNodeToDiskDict, + rpc_defs.ED_FILE_DETAILS: _PrepareFileUpload, + rpc_defs.ED_COMPRESS: _Compress, + rpc_defs.ED_FINALIZE_EXPORT_DISKS: _PrepareFinalizeExportDisks, + rpc_defs.ED_IMPEXP_IO: _EncodeImportExportIO, + rpc_defs.ED_BLOCKDEV_RENAME: _EncodeBlockdevRename, + } - This is a multi-node call. - """ - return self._MultiNodeCall(node_list, "node_volumes", []) +class RpcRunner(_RpcClientBase, + _generated_rpc.RpcClientDefault, + _generated_rpc.RpcClientBootstrap, + _generated_rpc.RpcClientConfig): + """RPC runner class. - def call_node_demote_from_mc(self, node): - """Demote a node from the master candidate role. + """ + def __init__(self, context): + """Initialized the RPC runner. - This is a single-node call. + @type context: C{masterd.GanetiContext} + @param context: Ganeti context """ - return self._SingleNodeCall(node, "node_demote_from_mc", []) + self._cfg = context.cfg + encoders = _ENCODERS.copy() - def call_node_powercycle(self, node, hypervisor): - """Tries to powercycle a node. + # Add encoders requiring configuration object + encoders.update({ + rpc_defs.ED_INST_DICT: self._InstDict, + rpc_defs.ED_INST_DICT_HVP_BEP: self._InstDictHvpBep, + rpc_defs.ED_INST_DICT_OSP: self._InstDictOsp, + }) - This is a single-node call. + # Resolver using configuration + resolver = compat.partial(_NodeConfigResolver, self._cfg.GetNodeInfo, + self._cfg.GetAllNodesInfo) - """ - return self._SingleNodeCall(node, "node_powercycle", [hypervisor]) + # Pylint doesn't recognize multiple inheritance properly, see + # and + # + # pylint: disable=W0233 + _RpcClientBase.__init__(self, resolver, encoders.get, + lock_monitor_cb=context.glm.AddToLockMonitor) + _generated_rpc.RpcClientConfig.__init__(self) + _generated_rpc.RpcClientBootstrap.__init__(self) + _generated_rpc.RpcClientDefault.__init__(self) + def _InstDict(self, instance, hvp=None, bep=None, osp=None): + """Convert the given instance to a dict. - def call_test_delay(self, node_list, duration): - """Sleep for a fixed time on given node(s). + This is done via the instance's ToDict() method and additionally + we fill the hvparams with the cluster defaults. - This is a multi-node call. + @type instance: L{objects.Instance} + @param instance: an Instance object + @type hvp: dict or None + @param hvp: a dictionary with overridden hypervisor parameters + @type bep: dict or None + @param bep: a dictionary with overridden backend parameters + @type osp: dict or None + @param osp: a dictionary with overridden os parameters + @rtype: dict + @return: the instance dict, with the hvparams filled with the + cluster defaults """ - return self._MultiNodeCall(node_list, "test_delay", [duration]) - - def call_file_storage_dir_create(self, node, file_storage_dir): - """Create the given file storage directory. + idict = instance.ToDict() + cluster = self._cfg.GetClusterInfo() + idict["hvparams"] = cluster.FillHV(instance) + if hvp is not None: + idict["hvparams"].update(hvp) + idict["beparams"] = cluster.FillBE(instance) + if bep is not None: + idict["beparams"].update(bep) + idict["osparams"] = cluster.SimpleFillOS(instance.os, instance.osparams) + if osp is not None: + idict["osparams"].update(osp) + for nic in idict["nics"]: + nic['nicparams'] = objects.FillDict( + cluster.nicparams[constants.PP_DEFAULT], + nic['nicparams']) + return idict - This is a single-node call. + def _InstDictHvpBep(self, (instance, hvp, bep)): + """Wrapper for L{_InstDict}. """ - return self._SingleNodeCall(node, "file_storage_dir_create", - [file_storage_dir]) + return self._InstDict(instance, hvp=hvp, bep=bep) - def call_file_storage_dir_remove(self, node, file_storage_dir): - """Remove the given file storage directory. - - This is a single-node call. + def _InstDictOsp(self, (instance, osparams)): + """Wrapper for L{_InstDict}. """ - return self._SingleNodeCall(node, "file_storage_dir_remove", - [file_storage_dir]) - - def call_file_storage_dir_rename(self, node, old_file_storage_dir, - new_file_storage_dir): - """Rename file storage directory. + return self._InstDict(instance, osp=osparams) - This is a single-node call. - """ - return self._SingleNodeCall(node, "file_storage_dir_rename", - [old_file_storage_dir, new_file_storage_dir]) +class JobQueueRunner(_RpcClientBase, _generated_rpc.RpcClientJobQueue): + """RPC wrappers for job queue. - @classmethod - def call_jobqueue_update(cls, node_list, address_list, file_name, content): - """Update job queue. - - This is a multi-node call. + """ + def __init__(self, context, address_list): + """Initializes this class. """ - return cls._StaticMultiNodeCall(node_list, "jobqueue_update", - [file_name, cls._Compress(content)], - address_list=address_list) + if address_list is None: + resolver = _SsconfResolver + else: + # Caller provided an address list + resolver = _StaticResolver(address_list) - @classmethod - def call_jobqueue_purge(cls, node): - """Purge job queue. + _RpcClientBase.__init__(self, resolver, _ENCODERS.get, + lock_monitor_cb=context.glm.AddToLockMonitor) + _generated_rpc.RpcClientJobQueue.__init__(self) - This is a single-node call. - """ - return cls._StaticSingleNodeCall(node, "jobqueue_purge", []) +class BootstrapRunner(_RpcClientBase, _generated_rpc.RpcClientBootstrap): + """RPC wrappers for bootstrapping. - @classmethod - def call_jobqueue_rename(cls, node_list, address_list, rename): - """Rename a job queue file. - - This is a multi-node call. + """ + def __init__(self): + """Initializes this class. """ - return cls._StaticMultiNodeCall(node_list, "jobqueue_rename", rename, - address_list=address_list) + _RpcClientBase.__init__(self, _SsconfResolver, _ENCODERS.get) + _generated_rpc.RpcClientBootstrap.__init__(self) - @classmethod - def call_jobqueue_set_drain(cls, node_list, drain_flag): - """Set the drain flag on the queue. - This is a multi-node call. +class ConfigRunner(_RpcClientBase, _generated_rpc.RpcClientConfig): + """RPC wrappers for L{config}. - @type node_list: list - @param node_list: the list of nodes to query - @type drain_flag: bool - @param drain_flag: if True, will set the drain flag, otherwise reset it. + """ + def __init__(self, context, address_list): + """Initializes this class. """ - return cls._StaticMultiNodeCall(node_list, "jobqueue_set_drain", - [drain_flag]) - - def call_hypervisor_validate_params(self, node_list, hvname, hvparams): - """Validate the hypervisor params. - - This is a multi-node call. + if context: + lock_monitor_cb = context.glm.AddToLockMonitor + else: + lock_monitor_cb = None - @type node_list: list - @param node_list: the list of nodes to query - @type hvname: string - @param hvname: the hypervisor name - @type hvparams: dict - @param hvparams: the hypervisor parameters to be validated + if address_list is None: + resolver = _SsconfResolver + else: + # Caller provided an address list + resolver = _StaticResolver(address_list) - """ - cluster = self._cfg.GetClusterInfo() - hv_full = objects.FillDict(cluster.hvparams.get(hvname, {}), hvparams) - return self._MultiNodeCall(node_list, "hypervisor_validate_params", - [hvname, hv_full]) + _RpcClientBase.__init__(self, resolver, _ENCODERS.get, + lock_monitor_cb=lock_monitor_cb) + _generated_rpc.RpcClientConfig.__init__(self)