#
#
-# 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
# 02110-1301, USA.
-"""Script to show add a new node to the cluster
+"""Inter-node RPC library.
"""
-# pylint: disable-msg=C0103
+# 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
+# if they need to start using instance attributes
+# R0904: Too many public methods
-import os
+import logging
+import zlib
+import base64
+import pycurl
+import threading
-from twisted.internet.pollreactor import PollReactor
-
-class ReReactor(PollReactor):
- """A re-startable Reactor implementation.
-
- """
- def run(self, installSignalHandlers=1):
- """Custom run method.
-
- This is customized run that, before calling Reactor.run, will
- reinstall the shutdown events and re-create the threadpool in case
- these are not present (as will happen on the second run of the
- reactor).
-
- """
- if not 'shutdown' in self._eventTriggers:
- # the shutdown queue has been killed, we are most probably
- # at the second run, thus recreate the queue
- self.addSystemEventTrigger('during', 'shutdown', self.crash)
- self.addSystemEventTrigger('during', 'shutdown', self.disconnectAll)
- if self.threadpool is not None and self.threadpool.joined == 1:
- # in case the threadpool has been stopped, re-start it
- # and add a trigger to stop it at reactor shutdown
- self.threadpool.start()
- self.addSystemEventTrigger('during', 'shutdown', self.threadpool.stop)
-
- return PollReactor.run(self, installSignalHandlers)
-
-
-import twisted.internet.main
-twisted.internet.main.installReactor(ReReactor())
-
-from twisted.spread import pb
-from twisted.internet import reactor
-from twisted.cred import credentials
-from OpenSSL import SSL, crypto
-
-from ganeti import logger
from ganeti import utils
-from ganeti import errors
-from ganeti import constants
from ganeti import objects
+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
-class NodeController:
- """Node-handling class.
+# Special module generated at build time
+from ganeti import _generated_rpc
- For each node that we speak with, we create an instance of this
- class, so that we have a safe place to store the details of this
- individual call.
+# pylint has a bug here, doesn't see this import
+import ganeti.http.client # pylint: disable=W0611
- """
- def __init__(self, parent, node):
- self.parent = parent
- self.node = node
- def _check_end(self):
- """Stop the reactor if we got all the results.
+# Timeout for connecting to nodes (seconds)
+_RPC_CONNECT_TIMEOUT = 5
- """
- if len(self.parent.results) == len(self.parent.nc):
- reactor.stop()
+_RPC_CLIENT_HEADERS = [
+ "Content-type: %s" % http.HTTP_APP_JSON,
+ "Expect:",
+ ]
- def cb_call(self, obj):
- """Callback for successful connect.
+# 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
- If the connect and login sequence succeeded, we proceed with
- making the actual call.
+#: Special value to describe an offline host
+_OFFLINE = object()
- """
- deferred = obj.callRemote(self.parent.procedure, self.parent.args)
- deferred.addCallbacks(self.cb_done, self.cb_err2)
- def cb_done(self, result):
- """Callback for successful call.
+def Init():
+ """Initializes the module-global HTTP client manager.
- When we receive the result from a call, we check if it was an
- error and if so we raise a generic RemoteError (we can't pass yet
- the actual exception over). If there was no error, we store the
- result.
-
- """
- tb, self.parent.results[self.node] = result
- self._check_end()
- if tb:
- raise errors.RemoteError("Remote procedure error calling %s on %s:"
- "\n%s" % (self.parent.procedure,
- self.node,
- tb))
-
- def cb_err1(self, reason):
- """Error callback for unsuccessful connect.
-
- """
- logger.Error("caller_connect: could not connect to remote host %s,"
- " reason %s" % (self.node, reason))
- self.parent.results[self.node] = False
- self._check_end()
-
- def cb_err2(self, reason):
- """Error callback for unsuccessful call.
-
- This is when the call didn't return anything, not even an error,
- or when it time out, etc.
-
- """
- logger.Error("caller_call: could not call %s on node %s,"
- " reason %s" % (self.parent.procedure, self.node, reason))
- self.parent.results[self.node] = False
- self._check_end()
-
-
-class MirrorContextFactory:
- """Certificate verifier factory.
-
- This factory creates contexts that verify if the remote end has a
- specific certificate (i.e. our own certificate).
-
- The checks we do are that the PEM dump of the certificate is the
- same as our own and (somewhat redundantly) that the SHA checksum is
- the same.
+ Must be called before using any RPC function and while exactly one thread is
+ running.
"""
- isClient = 1
-
- def __init__(self):
- try:
- fd = open(constants.SSL_CERT_FILE, 'r')
- try:
- data = fd.read(16384)
- finally:
- fd.close()
- except EnvironmentError, err:
- raise errors.ConfigurationError("missing SSL certificate: %s" %
- str(err))
- self.mycert = crypto.load_certificate(crypto.FILETYPE_PEM, data)
- self.mypem = crypto.dump_certificate(crypto.FILETYPE_PEM, self.mycert)
- self.mydigest = self.mycert.digest('SHA')
-
- def verifier(self, conn, x509, errno, err_depth, retcode):
- """Certificate verify method.
-
- """
- if self.mydigest != x509.digest('SHA'):
- return False
- if crypto.dump_certificate(crypto.FILETYPE_PEM, x509) != self.mypem:
- return False
- return True
+ # 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"
- def getContext(self):
- """Context generator.
+ logging.info("Using PycURL %s", pycurl.version)
- """
- context = SSL.Context(SSL.TLSv1_METHOD)
- context.set_verify(SSL.VERIFY_PEER, self.verifier)
- return context
+ pycurl.global_init(pycurl.GLOBAL_ALL)
-class Client:
- """RPC Client class.
- This class, given a (remote) ethod 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 Shutdown():
+ """Stops the module-global HTTP client manager.
- One current bug is that generic failure is still signalled by
- 'False' result, which is not good. This overloading of values can
- cause bugs.
+ Must be called before quitting the program and while exactly one thread is
+ running.
"""
- result_set = False
- result = False
- allresult = []
-
- def __init__(self, procedure, args):
- ss = ssconf.SimpleStore()
- self.port = ss.GetNodeDaemonPort()
- self.nodepw = ss.GetNodeDaemonPassword()
- self.nc = {}
- self.results = {}
- self.procedure = procedure
- self.args = args
-
- #--- generic connector -------------
-
- def connect_list(self, node_list):
- """Add a list of nodes to the target nodes.
-
- """
- for node in node_list:
- self.connect(node)
+ pycurl.global_cleanup()
- def connect(self, connect_node):
- """Add a node to the target list.
- """
- factory = pb.PBClientFactory()
- self.nc[connect_node] = nc = NodeController(self, connect_node)
- reactor.connectSSL(connect_node, self.port, factory,
- MirrorContextFactory())
- #d = factory.getRootObject()
- d = factory.login(credentials.UsernamePassword("master_node", self.nodepw))
- d.addCallbacks(nc.cb_call, nc.cb_err1)
+def _ConfigRpcCurl(curl):
+ noded_cert = str(constants.NODED_CERT_FILE)
- def getresult(self):
- """Return the results of the call.
+ 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)
- """
- return self.results
- def run(self):
- """Wrapper over reactor.run().
+def RunWithRPC(fn):
+ """RPC-wrapper decorator.
- This function simply calls reactor.run() if we have any requests
- queued, otherwise it does nothing.
-
- """
- if self.nc:
- reactor.run()
-
-
-def call_volume_list(node_list, vg_name):
- """Gets the logical volumes present in a given volume group.
-
- This is a multi-node call.
+ 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.
"""
- c = Client("volume_list", [vg_name])
- c.connect_list(node_list)
- c.run()
- return c.getresult()
-
-
-def call_vg_list(node_list):
- """Gets the volume group list.
-
- This is a multi-node call.
-
- """
- c = Client("vg_list", [])
- c.connect_list(node_list)
- c.run()
- return c.getresult()
-
-
-def call_bridges_exist(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.
+ def wrapper(*args, **kwargs):
+ Init()
+ try:
+ return fn(*args, **kwargs)
+ finally:
+ Shutdown()
+ return wrapper
- """
- c = Client("bridges_exist", [bridges_list])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+def _Compress(data):
+ """Compresses a string for transport over RPC.
-def call_instance_start(node, instance, extra_args):
- """Stars an instance.
+ Small amounts of data are not compressed.
- This is a single-node call.
+ @type data: str
+ @param data: Data
+ @rtype: tuple
+ @return: Encoded data to send
"""
- c = Client("instance_start", [instance.ToDict(), extra_args])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ # 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)))
-def call_instance_shutdown(node, instance):
- """Stops an instance.
- This is a single-node call.
-
- """
- c = Client("instance_shutdown", [instance.ToDict()])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+class RpcResult(object):
+ """RPC Result class.
+ This class holds an RPC result. It is needed since in multi-node
+ calls we can't raise an exception just because one one out of many
+ failed, and therefore we use this class to encapsulate the result.
-def call_instance_os_add(node, inst, osdev, swapdev):
- """Installs an OS on the given instance.
-
- This is a single-node call.
+ @ivar data: the data payload, for successful results, or None
+ @ivar call: the name of the RPC call
+ @ivar node: the name of the node to which we made the call
+ @ivar offline: whether the operation failed because the node was
+ offline, as opposed to actual failure; offline=True will always
+ imply failed=True, in order to allow simpler checking if
+ the user doesn't care about the exact failure mode
+ @ivar fail_msg: the error message if the call failed
"""
- params = [inst.ToDict(), osdev, swapdev]
- c = Client("instance_os_add", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
-
-
-def call_instance_run_rename(node, inst, old_name, osdev, swapdev):
- """Run the OS rename script for an instance.
-
- This is a single-node call.
+ def __init__(self, data=None, failed=False, offline=False,
+ call=None, node=None):
+ self.offline = offline
+ self.call = call
+ self.node = node
- """
- params = [inst.ToDict(), old_name, osdev, swapdev]
- c = Client("instance_run_rename", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ if offline:
+ self.fail_msg = "Node is marked offline"
+ self.data = self.payload = None
+ elif failed:
+ self.fail_msg = self._EnsureErr(data)
+ self.data = self.payload = None
+ else:
+ self.data = data
+ if not isinstance(self.data, (tuple, list)):
+ self.fail_msg = ("RPC layer error: invalid result type (%s)" %
+ type(self.data))
+ self.payload = None
+ elif len(data) != 2:
+ self.fail_msg = ("RPC layer error: invalid result length (%d), "
+ "expected 2" % len(self.data))
+ self.payload = None
+ elif not self.data[0]:
+ self.fail_msg = self._EnsureErr(self.data[1])
+ self.payload = None
+ else:
+ # finally success
+ self.fail_msg = None
+ self.payload = data[1]
+
+ 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):
+ """Helper to ensure we return a 'True' value for error."""
+ if val:
+ return val
+ else:
+ return "No error information"
+ def Raise(self, msg, prereq=False, ecode=None):
+ """If the result has failed, raise an OpExecError.
-def call_instance_info(node, instance):
- """Returns information about a single instance.
+ This is used so that LU code doesn't have to check for each
+ result, but instead can call this function.
- This is a single-node call.
+ """
+ if not self.fail_msg:
+ return
- """
- c = Client("instance_info", [instance])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ if not msg: # one could pass None for default message
+ msg = ("Call '%s' to node '%s' has failed: %s" %
+ (self.call, self.node, self.fail_msg))
+ else:
+ msg = "%s: %s" % (msg, self.fail_msg)
+ if prereq:
+ ec = errors.OpPrereqError
+ else:
+ ec = errors.OpExecError
+ if ecode is not None:
+ args = (msg, ecode)
+ else:
+ args = (msg, )
+ raise ec(*args) # pylint: disable=W0142
-def call_all_instances_info(node_list):
- """Returns information about all instances on a given node.
+def _SsconfResolver(node_list, _,
+ ssc=ssconf.SimpleStore,
+ nslookup_fn=netutils.Hostname.GetIP):
+ """Return addresses for given node names.
- This is a single-node call.
+ @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
"""
- c = Client("all_instances_info", [])
- c.connect_list(node_list)
- c.run()
- return c.getresult()
-
+ ss = ssc()
+ iplist = ss.GetNodePrimaryIPList()
+ family = ss.GetPrimaryIPFamily()
+ ipmap = dict(entry.split() for entry in iplist)
-def call_instance_list(node_list):
- """Returns the list of running instances on a given node.
+ result = []
+ for node in node_list:
+ ip = ipmap.get(node)
+ if ip is None:
+ ip = nslookup_fn(node, family=family)
+ result.append((node, ip))
- This is a single-node call.
+ return result
- """
- c = Client("instance_list", [])
- c.connect_list(node_list)
- c.run()
- return c.getresult()
+class _StaticResolver:
+ def __init__(self, addresses):
+ """Initializes this class.
-def call_node_tcp_ping(node, source, target, port, timeout, live_port_needed):
- """Do a TcpPing on the remote node
+ """
+ self._addresses = addresses
- This is a single-node call.
- """
- c = Client("node_tcp_ping", [source, target, port, timeout,
- live_port_needed])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ def __call__(self, hosts, _):
+ """Returns static addresses for hosts.
+ """
+ assert len(hosts) == len(self._addresses)
+ return zip(hosts, self._addresses)
-def call_node_info(node_list, vg_name):
- """Return node information.
- This will return memory information and volume group size and free
- space.
+def _CheckConfigNode(name, node, accept_offline_node):
+ """Checks if a node is online.
- This is a multi-node call.
+ @type name: string
+ @param name: Node name
+ @type node: L{objects.Node} or None
+ @param node: Node object
"""
- c = Client("node_info", [vg_name])
- c.connect_list(node_list)
- c.run()
- retux = c.getresult()
-
- for node_name in retux:
- ret = retux.get(node_name, False)
- if type(ret) != dict:
- logger.Error("could not connect to node %s" % (node_name))
- ret = {}
-
- utils.CheckDict(ret,
- { 'memory_total' : '-',
- 'memory_dom0' : '-',
- 'memory_free' : '-',
- 'vg_size' : 'node_unreachable',
- 'vg_free' : '-' },
- "call_node_info",
- )
- return retux
+ if node is None:
+ # Depend on DNS for name resolution
+ ip = name
+ elif node.offline and not accept_offline_node:
+ ip = _OFFLINE
+ else:
+ ip = node.primary_ip
+ return (name, ip)
-def call_node_add(node, dsa, dsapub, rsa, rsapub, ssh, sshpub):
- """Add a node to the cluster.
-
- This is a single-node call.
+def _NodeConfigResolver(single_node_fn, all_nodes_fn, hosts, opts):
+ """Calculate node addresses using configuration.
"""
- params = [dsa, dsapub, rsa, rsapub, ssh, sshpub]
- c = Client("node_add", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ accept_offline_node = (opts is rpc_defs.ACCEPT_OFFLINE_NODE)
+ assert accept_offline_node or opts is None, "Unknown option"
-def call_node_verify(node_list, checkdict):
- """Request verification of given parameters.
+ # Special case for single-host lookups
+ if len(hosts) == 1:
+ (name, ) = hosts
+ return [_CheckConfigNode(name, single_node_fn(name), accept_offline_node)]
+ else:
+ all_nodes = all_nodes_fn()
+ return [_CheckConfigNode(name, all_nodes.get(name, None),
+ accept_offline_node)
+ for name in hosts]
- This is a multi-node call.
- """
- c = Client("node_verify", [checkdict])
- c.connect_list(node_list)
- c.run()
- return c.getresult()
+class _RpcProcessor:
+ def __init__(self, resolver, port, lock_monitor_cb=None):
+ """Initializes this class.
+ @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
-def call_node_start_master(node):
- """Tells a node to activate itself as a master.
+ """
+ self._resolver = resolver
+ self._port = port
+ self._lock_monitor_cb = lock_monitor_cb
- This is a single-node call.
+ @staticmethod
+ def _PrepareRequests(hosts, port, procedure, body, read_timeout):
+ """Prepares requests by sorting offline hosts into separate list.
- """
- c = Client("node_start_master", [])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ @type body: dict
+ @param body: a dictionary with per-host body data
+ """
+ results = {}
+ requests = {}
+
+ assert isinstance(body, dict)
+ assert len(body) == len(hosts)
+ assert compat.all(isinstance(v, str) for v in body.values())
+ assert frozenset(map(compat.fst, hosts)) == frozenset(body.keys()), \
+ "%s != %s" % (hosts, body.keys())
+
+ for (name, ip) in hosts:
+ if ip is _OFFLINE:
+ # Node is marked as offline
+ results[name] = RpcResult(node=name, offline=True, call=procedure)
+ else:
+ requests[name] = \
+ http.client.HttpClientRequest(str(ip), port,
+ http.HTTP_POST, str("/%s" % procedure),
+ headers=_RPC_CLIENT_HEADERS,
+ post_data=body[name],
+ read_timeout=read_timeout,
+ nicename="%s/%s" % (name, procedure),
+ curl_config_fn=_ConfigRpcCurl)
+
+ return (results, requests)
+
+ @staticmethod
+ def _CombineResults(results, requests, procedure):
+ """Combines pre-computed results for offline hosts with actual call results.
-def call_node_stop_master(node):
- """Tells a node to demote itself from master status.
+ """
+ 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
+
+ logging.error("RPC error in %s on node %s: %s", procedure, name, msg)
+ host_result = RpcResult(data=msg, failed=True, node=name,
+ call=procedure)
+
+ results[name] = host_result
+
+ return results
+
+ def __call__(self, hosts, procedure, body, read_timeout, resolver_opts,
+ _req_process_fn=None):
+ """Makes an RPC request to a number of nodes.
+
+ @type hosts: sequence
+ @param hosts: Hostnames
+ @type procedure: string
+ @param procedure: Request path
+ @type body: dictionary
+ @param body: dictionary with request bodies per host
+ @type read_timeout: int or None
+ @param read_timeout: Read timeout for request
- This is a single-node call.
+ """
+ assert read_timeout is not None, \
+ "Missing RPC read timeout for procedure '%s'" % procedure
- """
- c = Client("node_stop_master", [])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ if _req_process_fn is None:
+ _req_process_fn = http.client.ProcessRequests
+ (results, requests) = \
+ self._PrepareRequests(self._resolver(hosts, resolver_opts), self._port,
+ procedure, body, read_timeout)
-def call_version(node_list):
- """Query node version.
+ _req_process_fn(requests.values(), lock_monitor_cb=self._lock_monitor_cb)
- This is a multi-node call.
+ assert not frozenset(results).intersection(requests)
- """
- c = Client("version", [])
- c.connect_list(node_list)
- c.run()
- return c.getresult()
+ return self._CombineResults(results, requests, procedure)
-def call_blockdev_create(node, bdev, size, on_primary, info):
- """Request creation of a given block device.
+class _RpcClientBase:
+ def __init__(self, resolver, encoder_fn, lock_monitor_cb=None,
+ _req_process_fn=None):
+ """Initializes this class.
- This is a single-node call.
-
- """
- params = [bdev.ToDict(), size, on_primary, info]
- c = Client("blockdev_create", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ """
+ proc = _RpcProcessor(resolver,
+ netutils.GetDaemonPort(constants.NODED),
+ lock_monitor_cb=lock_monitor_cb)
+ self._proc = compat.partial(proc, _req_process_fn=_req_process_fn)
+ self._encoder = compat.partial(self._EncodeArg, encoder_fn)
+ @staticmethod
+ def _EncodeArg(encoder_fn, (argkind, value)):
+ """Encode argument.
-def call_blockdev_remove(node, bdev):
- """Request removal of a given block device.
+ """
+ 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.
- """
- c = Client("blockdev_remove", [bdev.ToDict()])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ """
+ (procedure, _, resolver_opts, timeout, argdefs,
+ prep_fn, postproc_fn, _) = cdef
+ if callable(timeout):
+ read_timeout = timeout(args)
+ else:
+ read_timeout = timeout
-def call_blockdev_assemble(node, disk, on_primary):
- """Request assembling of a given block device.
+ if callable(resolver_opts):
+ req_resolver_opts = resolver_opts(args)
+ else:
+ req_resolver_opts = resolver_opts
- This is a single-node call.
+ if len(args) != len(argdefs):
+ raise errors.ProgrammerError("Number of passed arguments doesn't match")
- """
- params = [disk.ToDict(), on_primary]
- c = Client("blockdev_assemble", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ enc_args = map(self._encoder, zip(map(compat.snd, argdefs), args))
+ if prep_fn is None:
+ # for a no-op prep_fn, we serialise the body once, and then we
+ # reuse it in the dictionary values
+ body = serializer.DumpJson(enc_args)
+ pnbody = dict((n, body) for n in node_list)
+ else:
+ # for a custom prep_fn, we pass the encoded arguments and the
+ # node name to the prep_fn, and we serialise its return value
+ assert callable(prep_fn)
+ pnbody = dict((n, serializer.DumpJson(prep_fn(n, enc_args)))
+ for n in node_list)
+
+ result = self._proc(node_list, procedure, pnbody, read_timeout,
+ req_resolver_opts)
+
+ if postproc_fn:
+ return dict(map(lambda (key, value): (key, postproc_fn(value)),
+ result.items()))
+ else:
+ return result
-def call_blockdev_shutdown(node, disk):
- """Request shutdown of a given block device.
+def _ObjectToDict(value):
+ """Converts an object to a dictionary.
- This is a single-node call.
+ @note: See L{objects}.
"""
- c = Client("blockdev_shutdown", [disk.ToDict()])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ return value.ToDict()
-def call_blockdev_addchild(node, bdev, ndev):
- """Request adding a new child to a (mirroring) device.
-
- This is a single-node call.
+def _ObjectListToDict(value):
+ """Converts a list of L{objects} to dictionaries.
"""
- params = [bdev.ToDict(), ndev.ToDict()]
- c = Client("blockdev_addchild", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
-
+ return map(_ObjectToDict, value)
-def call_blockdev_removechild(node, bdev, ndev):
- """Request removing a new child from a (mirroring) device.
- This is a single-node call.
+def _EncodeNodeToDiskDict(value):
+ """Encodes a dictionary with node name as key and disk objects as values.
"""
- params = [bdev.ToDict(), ndev.ToDict()]
- c = Client("blockdev_removechild", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ return dict((name, _ObjectListToDict(disks))
+ for name, disks in value.items())
-def call_blockdev_getmirrorstatus(node, disks):
- """Request status of a (mirroring) device.
-
- This is a single-node call.
+def _PrepareFileUpload(getents_fn, filename):
+ """Loads a file and prepares it for an upload to nodes.
"""
- params = [dsk.ToDict() for dsk in disks]
- c = Client("blockdev_getmirrorstatus", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
-
+ statcb = utils.FileStatHelper()
+ data = _Compress(utils.ReadFile(filename, preread=statcb))
+ st = statcb.st
-def call_blockdev_find(node, disk):
- """Request identification of a given block device.
+ if getents_fn is None:
+ getents_fn = runtime.GetEnts
- This is a single-node call.
-
- """
- c = Client("blockdev_find", [disk.ToDict()])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ getents = getents_fn()
+ return [filename, data, st.st_mode, getents.LookupUid(st.st_uid),
+ getents.LookupGid(st.st_gid), st.st_atime, st.st_mtime]
-def call_upload_file(node_list, file_name):
- """Upload a file.
- The node will refuse the operation in case the file is not on the
- approved file list.
-
- This is a multi-node call.
+def _PrepareFinalizeExportDisks(snap_disks):
+ """Encodes disks for finalizing export.
"""
- fh = file(file_name)
- try:
- data = fh.read()
- finally:
- fh.close()
- st = os.stat(file_name)
- params = [file_name, data, st.st_mode, st.st_uid, st.st_gid,
- st.st_atime, st.st_mtime]
- c = Client("upload_file", params)
- c.connect_list(node_list)
- c.run()
- return c.getresult()
+ flat_disks = []
+ for disk in snap_disks:
+ if isinstance(disk, bool):
+ flat_disks.append(disk)
+ else:
+ flat_disks.append(disk.ToDict())
-def call_os_diagnose(node_list):
- """Request a diagnose of OS definitions.
+ return flat_disks
- This is a multi-node call.
- """
- c = Client("os_diagnose", [])
- c.connect_list(node_list)
- c.run()
- result = c.getresult()
- new_result = {}
- for node_name in result:
- nr = []
- if result[node_name]:
- for data in result[node_name]:
- if data:
- if isinstance(data, dict):
- nr.append(objects.OS.FromDict(data))
- elif isinstance(data, tuple) and len(data) == 3:
- nr.append(errors.InvalidOS(data[0], data[1], data[2]))
- else:
- raise errors.ProgrammerError("Invalid data from"
- " xcserver.os_diagnose")
- new_result[node_name] = nr
- return new_result
-
-
-def call_os_get(node_list, name):
- """Returns an OS definition.
-
- This is a multi-node call.
+def _EncodeImportExportIO((ieio, ieioargs)):
+ """Encodes import/export I/O information.
"""
- c = Client("os_get", [name])
- c.connect_list(node_list)
- c.run()
- result = c.getresult()
- new_result = {}
- for node_name in result:
- data = result[node_name]
- if isinstance(data, dict):
- new_result[node_name] = objects.OS.FromDict(data)
- elif isinstance(data, tuple) and len(data) == 3:
- new_result[node_name] = errors.InvalidOS(data[0], data[1], data[2])
- else:
- new_result[node_name] = data
- return new_result
+ if ieio == constants.IEIO_RAW_DISK:
+ assert len(ieioargs) == 1
+ return (ieio, (ieioargs[0].ToDict(), ))
+ if ieio == constants.IEIO_SCRIPT:
+ assert len(ieioargs) == 2
+ return (ieio, (ieioargs[0].ToDict(), ieioargs[1]))
-def call_hooks_runner(node_list, hpath, phase, env):
- """Call the hooks runner.
+ return (ieio, ieioargs)
- Args:
- - op: the OpCode instance
- - env: a dictionary with the environment
- This is a multi-node call.
+def _EncodeBlockdevRename(value):
+ """Encodes information for renaming block devices.
"""
- params = [hpath, phase, env]
- c = Client("hooks_runner", params)
- c.connect_list(node_list)
- c.run()
- result = c.getresult()
- return result
-
-
-def call_blockdev_snapshot(node, cf_bdev):
- """Request a snapshot of the given block device.
+ return [(d.ToDict(), uid) for d, uid in value]
- This is a single-node call.
-
- """
- c = Client("blockdev_snapshot", [cf_bdev.ToDict()])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+#: 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_COMPRESS: _Compress,
+ rpc_defs.ED_FINALIZE_EXPORT_DISKS: _PrepareFinalizeExportDisks,
+ rpc_defs.ED_IMPEXP_IO: _EncodeImportExportIO,
+ rpc_defs.ED_BLOCKDEV_RENAME: _EncodeBlockdevRename,
+ }
-def call_snapshot_export(node, snap_bdev, dest_node, instance):
- """Request the export of a given snapshot.
- This is a single-node call.
+class RpcRunner(_RpcClientBase,
+ _generated_rpc.RpcClientDefault,
+ _generated_rpc.RpcClientBootstrap,
+ _generated_rpc.RpcClientConfig):
+ """RPC runner class.
"""
- params = [snap_bdev.ToDict(), dest_node, instance.ToDict()]
- c = Client("snapshot_export", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ def __init__(self, cfg, lock_monitor_cb, _req_process_fn=None, _getents=None):
+ """Initialized the RPC runner.
+ @type cfg: L{config.ConfigWriter}
+ @param cfg: Configuration
+ @type lock_monitor_cb: callable
+ @param lock_monitor_cb: Lock monitor callback
-def call_finalize_export(node, instance, snap_disks):
- """Request the completion of an export operation.
+ """
+ self._cfg = cfg
+
+ encoders = _ENCODERS.copy()
+
+ encoders.update({
+ # Encoders requiring configuration object
+ rpc_defs.ED_INST_DICT: self._InstDict,
+ rpc_defs.ED_INST_DICT_HVP_BEP: self._InstDictHvpBep,
+ rpc_defs.ED_INST_DICT_OSP: self._InstDictOsp,
+
+ # Encoders with special requirements
+ rpc_defs.ED_FILE_DETAILS: compat.partial(_PrepareFileUpload, _getents),
+ })
+
+ # Resolver using configuration
+ resolver = compat.partial(_NodeConfigResolver, cfg.GetNodeInfo,
+ cfg.GetAllNodesInfo)
+
+ # Pylint doesn't recognize multiple inheritance properly, see
+ # <http://www.logilab.org/ticket/36586> and
+ # <http://www.logilab.org/ticket/35642>
+ # pylint: disable=W0233
+ _RpcClientBase.__init__(self, resolver, encoders.get,
+ lock_monitor_cb=lock_monitor_cb,
+ _req_process_fn=_req_process_fn)
+ _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.
+
+ 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
+ @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
- This writes the export config file, etc.
+ """
+ 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
+
+ def _InstDictHvpBep(self, (instance, hvp, bep)):
+ """Wrapper for L{_InstDict}.
- This is a single-node call.
+ """
+ return self._InstDict(instance, hvp=hvp, bep=bep)
- """
- flat_disks = []
- for disk in snap_disks:
- flat_disks.append(disk.ToDict())
- params = [instance.ToDict(), flat_disks]
- c = Client("finalize_export", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ def _InstDictOsp(self, (instance, osparams)):
+ """Wrapper for L{_InstDict}.
+ """
+ return self._InstDict(instance, osp=osparams)
-def call_export_info(node, path):
- """Queries the export information in a given path.
- This is a single-node call.
+class JobQueueRunner(_RpcClientBase, _generated_rpc.RpcClientJobQueue):
+ """RPC wrappers for job queue.
"""
- c = Client("export_info", [path])
- c.connect(node)
- c.run()
- result = c.getresult().get(node, False)
- if not result:
- return result
- return objects.SerializableConfigParser.Loads(result)
+ def __init__(self, context, address_list):
+ """Initializes this class.
+ """
+ if address_list is None:
+ resolver = _SsconfResolver
+ else:
+ # Caller provided an address list
+ resolver = _StaticResolver(address_list)
-def call_instance_os_import(node, inst, osdev, swapdev, src_node, src_image):
- """Request the import of a backup into an instance.
-
- This is a single-node call.
-
- """
- params = [inst.ToDict(), osdev, swapdev, src_node, src_image]
- c = Client("instance_os_import", params)
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
-
+ _RpcClientBase.__init__(self, resolver, _ENCODERS.get,
+ lock_monitor_cb=context.glm.AddToLockMonitor)
+ _generated_rpc.RpcClientJobQueue.__init__(self)
-def call_export_list(node_list):
- """Gets the stored exports list.
- This is a multi-node call.
+class BootstrapRunner(_RpcClientBase, _generated_rpc.RpcClientBootstrap):
+ """RPC wrappers for bootstrapping.
"""
- c = Client("export_list", [])
- c.connect_list(node_list)
- c.run()
- result = c.getresult()
- return result
+ def __init__(self):
+ """Initializes this class.
+ """
+ _RpcClientBase.__init__(self, _SsconfResolver, _ENCODERS.get)
+ _generated_rpc.RpcClientBootstrap.__init__(self)
-def call_export_remove(node, export):
- """Requests removal of a given export.
- This is a single-node call.
+class ConfigRunner(_RpcClientBase, _generated_rpc.RpcClientConfig):
+ """RPC wrappers for L{config}.
"""
- c = Client("export_remove", [export])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
+ def __init__(self, context, address_list, _req_process_fn=None,
+ _getents=None):
+ """Initializes this class.
+ """
+ if context:
+ lock_monitor_cb = context.glm.AddToLockMonitor
+ else:
+ lock_monitor_cb = None
-def call_node_leave_cluster(node):
- """Requests a node to clean the cluster information it has.
-
- This will remove the configuration information from the ganeti data
- dir.
-
- This is a single-node call.
-
- """
- c = Client("node_leave_cluster", [])
- c.connect(node)
- c.run()
- return c.getresult().get(node, False)
-
+ if address_list is None:
+ resolver = _SsconfResolver
+ else:
+ # Caller provided an address list
+ resolver = _StaticResolver(address_list)
-def call_node_volumes(node_list):
- """Gets all volumes on node(s).
+ encoders = _ENCODERS.copy()
- This is a multi-node call.
+ encoders.update({
+ rpc_defs.ED_FILE_DETAILS: compat.partial(_PrepareFileUpload, _getents),
+ })
- """
- c = Client("node_volumes", [])
- c.connect_list(node_list)
- c.run()
- return c.getresult()
+ _RpcClientBase.__init__(self, resolver, encoders.get,
+ lock_monitor_cb=lock_monitor_cb,
+ _req_process_fn=_req_process_fn)
+ _generated_rpc.RpcClientConfig.__init__(self)