"""
-# 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
from ganeti import netutils
from ganeti import ssconf
from ganeti import runtime
+from ganeti import compat
# 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)
_TIMEOUTS = {
}
+#: Special value to describe an offline host
+_OFFLINE = object()
+
def Init():
"""Initializes the module-global HTTP client manager.
curl.setopt(pycurl.CONNECTTIMEOUT, _RPC_CONNECT_TIMEOUT)
-# Aliasing this module avoids the following warning by epydoc: "Warning: No
-# information available for ganeti.rpc._RpcThreadLocal's base threading.local"
-_threading = threading
-
-
-class _RpcThreadLocal(_threading.local):
- def GetHttpClientPool(self):
- """Returns a per-thread HTTP client pool.
-
- @rtype: L{http.client.HttpClientPool}
-
- """
- try:
- pool = self.hcp
- except AttributeError:
- pool = http.client.HttpClientPool(_ConfigRpcCurl)
- self.hcp = pool
-
- return pool
-
-
-# Remove module alias (see above)
-del _threading
-
-
-_thread_local = _RpcThreadLocal()
-
-
def _RpcTimeout(secs):
"""Timeout decorator.
return wrapper
+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):
"""RPC Result class.
args = (msg, ecode)
else:
args = (msg, )
- raise ec(*args) # pylint: disable-msg=W0142
+ raise ec(*args) # pylint: disable=W0142
-def _AddressLookup(node_list,
- ssc=ssconf.SimpleStore,
- nslookup_fn=netutils.Hostname.GetIP):
+def _SsconfResolver(node_list,
+ ssc=ssconf.SimpleStore,
+ nslookup_fn=netutils.Hostname.GetIP):
"""Return addresses for given node names.
@type node_list: list
@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 addresses and/or None's
- @returns: List of corresponding addresses, if found
+ @rtype: list of tuple; (string, string)
+ @return: List of tuples containing node name and IP address
"""
ss = ssc()
iplist = ss.GetNodePrimaryIPList()
family = ss.GetPrimaryIPFamily()
- addresses = []
ipmap = dict(entry.split() for entry in iplist)
- for node in node_list:
- address = ipmap.get(node)
- if address is None:
- address = nslookup_fn(node, family=family)
- addresses.append(address)
-
- return addresses
+ result = []
+ for node in node_list:
+ ip = ipmap.get(node)
+ if ip is None:
+ ip = nslookup_fn(node, family=family)
+ result.append((node, ip))
-class Client:
- """RPC Client class.
+ return result
- 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).
- One current bug is that generic failure is still signaled by
- 'False' result, which is not good. This overloading of values can
- cause bugs.
+class _StaticResolver:
+ def __init__(self, addresses):
+ """Initializes this class.
- """
- def __init__(self, procedure, body, port, address_lookup_fn=_AddressLookup):
- assert procedure in _TIMEOUTS, ("New RPC call not declared in the"
- " timeouts table")
- self.procedure = procedure
- self.body = body
- self.port = port
- self._request = {}
- self._address_lookup_fn = address_lookup_fn
-
- def ConnectList(self, node_list, address_list=None, read_timeout=None):
- """Add a list of nodes to the target nodes.
+ """
+ self._addresses = addresses
- @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
- @type read_timeout: int
- @param read_timeout: overwrites default timeout for operation
+ def __call__(self, hosts):
+ """Returns static addresses for hosts.
"""
- if address_list is None:
- # Always use IP address instead of node name
- address_list = self._address_lookup_fn(node_list)
+ assert len(hosts) == len(self._addresses)
+ return zip(hosts, self._addresses)
- assert len(node_list) == len(address_list), \
- "Name and address lists must have the same length"
- for node, address in zip(node_list, address_list):
- self.ConnectNode(node, address, read_timeout=read_timeout)
+def _CheckConfigNode(name, node):
+ """Checks if a node is online.
- def ConnectNode(self, name, address=None, read_timeout=None):
- """Add a node to the target list.
+ @type name: string
+ @param name: Node name
+ @type node: L{objects.Node} or None
+ @param node: Node object
- @type name: str
- @param name: the node name
- @type address: str
- @param address: the node address, if known
- @type read_timeout: int
- @param read_timeout: overwrites default timeout for operation
+ """
+ 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)
- """
- if address is None:
- # Always use IP address instead of node name
- address = self._address_lookup_fn([name])[0]
- assert(address is not None)
+def _NodeConfigResolver(single_node_fn, all_nodes_fn, hosts):
+ """Calculate node addresses using configuration.
- if read_timeout is None:
- read_timeout = _TIMEOUTS[self.procedure]
+ """
+ # 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]
- self._request[name] = \
- http.client.HttpClientRequest(str(address), self.port,
- http.HTTP_PUT, str("/%s" % self.procedure),
- headers=_RPC_CLIENT_HEADERS,
- post_data=str(self.body),
- read_timeout=read_timeout)
- def GetResults(self, http_pool=None):
- """Call nodes and return results.
+class _RpcProcessor:
+ def __init__(self, resolver, port, lock_monitor_cb=None):
+ """Initializes this class.
- @rtype: list
- @return: List of RPC results
+ @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
"""
- if not http_pool:
- http_pool = _thread_local.GetHttpClientPool()
+ self._resolver = resolver
+ self._port = port
+ self._lock_monitor_cb = lock_monitor_cb
- http_pool.ProcessRequests(self._request.values())
+ @staticmethod
+ def _PrepareRequests(hosts, port, procedure, body, read_timeout):
+ """Prepares requests by sorting offline hosts into separate list.
+ """
results = {}
+ requests = {}
- for name, req in self._request.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
+ 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_PUT, str("/%s" % procedure),
+ headers=_RPC_CLIENT_HEADERS,
+ post_data=body,
+ read_timeout=read_timeout,
+ nicename="%s/%s" % (name, procedure),
+ curl_config_fn=_ConfigRpcCurl)
+
+ return (results, requests)
- # TODO: Better error reporting
- if req.error:
- msg = req.error
+ @staticmethod
+ def _CombineResults(results, requests, procedure):
+ """Combines pre-computed results for offline hosts with actual call results.
+
+ """
+ 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:
- msg = req.resp_body
+ # TODO: Better error reporting
+ if req.error:
+ msg = req.error
+ else:
+ msg = req.resp_body
- 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)
+ 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=None,
+ _req_process_fn=http.client.ProcessRequests):
+ """Makes an RPC request to a number of nodes.
+
+ @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
+
+ """
+ assert procedure in _TIMEOUTS, "RPC call not declared in the timeouts table"
+
+ if read_timeout is None:
+ read_timeout = _TIMEOUTS[procedure]
+
+ (results, requests) = \
+ self._PrepareRequests(self._resolver(hosts), self._port, procedure,
+ str(body), read_timeout)
+
+ _req_process_fn(requests.values(), lock_monitor_cb=self._lock_monitor_cb)
+
+ assert not frozenset(results).intersection(requests)
+
+ return self._CombineResults(results, requests, procedure)
+
def _EncodeImportExportIO(ieio, ieioargs):
"""Encodes import/export I/O information.
class RpcRunner(object):
- """RPC runner class"""
+ """RPC runner class.
- def __init__(self, cfg):
- """Initialized the rpc runner.
+ """
+ def __init__(self, context):
+ """Initialized the RPC runner.
- @type cfg: C{config.ConfigWriter}
- @param cfg: the configuration object that will be used to get data
- about the cluster
+ @type context: C{masterd.GanetiContext}
+ @param context: Ganeti context
"""
- self._cfg = cfg
- self.port = netutils.GetDaemonPort(constants.NODED)
+ self._cfg = context.cfg
+ self._proc = _RpcProcessor(compat.partial(_NodeConfigResolver,
+ self._cfg.GetNodeInfo,
+ self._cfg.GetAllNodesInfo),
+ netutils.GetDaemonPort(constants.NODED),
+ lock_monitor_cb=context.glm.AddToLockMonitor)
def _InstDict(self, instance, hvp=None, bep=None, osp=None):
"""Convert the given instance to a dict.
nic['nicparams'])
return idict
- def _ConnectList(self, client, node_list, call, read_timeout=None):
- """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
- @type read_timeout: int
- @param read_timeout: overwrites the default read timeout for the
- given operation
-
- """
- 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
- else:
- val = None
- addr_list.append(val)
- name_list.append(node)
- if name_list:
- client.ConnectList(name_list, address_list=addr_list,
- read_timeout=read_timeout)
- return skip_dict
-
- def _ConnectNode(self, client, node, call, read_timeout=None):
- """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
- @type read_timeout: int
- @param read_timeout: overwrites the default read timeout for the
- given operation
-
- """
- 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, read_timeout=read_timeout)
-
def _MultiNodeCall(self, node_list, procedure, args, read_timeout=None):
"""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,
- read_timeout=read_timeout)
- skip_dict.update(c.GetResults())
- return skip_dict
+ return self._proc(node_list, procedure, body, read_timeout=read_timeout)
- @classmethod
- def _StaticMultiNodeCall(cls, node_list, procedure, args,
+ @staticmethod
+ def _StaticMultiNodeCall(node_list, procedure, args,
address_list=None, read_timeout=None):
"""Helper for making a multi-node static call
"""
body = serializer.DumpJson(args, indent=False)
- c = Client(procedure, body, netutils.GetDaemonPort(constants.NODED))
- c.ConnectList(node_list, address_list=address_list,
- read_timeout=read_timeout)
- return c.GetResults()
+
+ if address_list is None:
+ resolver = _SsconfResolver
+ else:
+ # Caller provided an address list
+ resolver = _StaticResolver(address_list)
+
+ proc = _RpcProcessor(resolver,
+ netutils.GetDaemonPort(constants.NODED))
+ return proc(node_list, procedure, body, read_timeout=read_timeout)
def _SingleNodeCall(self, node, procedure, args, read_timeout=None):
"""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, read_timeout=read_timeout)
- if result is None:
- # we did connect, node is not offline
- result = c.GetResults()[node]
- return result
+ return self._proc([node], procedure, body, read_timeout=read_timeout)[node]
@classmethod
def _StaticSingleNodeCall(cls, node, procedure, args, read_timeout=None):
"""
body = serializer.DumpJson(args, indent=False)
- c = Client(procedure, body, netutils.GetDaemonPort(constants.NODED))
- c.ConnectNode(node, read_timeout=read_timeout)
- return c.GetResults()[node]
+ proc = _RpcProcessor(_SsconfResolver,
+ netutils.GetDaemonPort(constants.NODED))
+ return proc([node], procedure, body, read_timeout=read_timeout)[node]
- @staticmethod
- def _Compress(data):
- """Compresses a string for transport over RPC.
+ #
+ # Begin RPC calls
+ #
- Small amounts of data are not compressed.
+ @_RpcTimeout(_TMO_URGENT)
+ def call_bdev_sizes(self, node_list, devices):
+ """Gets the sizes of requested block devices present on a node
- @type data: str
- @param data: Data
- @rtype: tuple
- @return: Encoded data to send
+ This is a multi-node call.
"""
- # 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
- #
+ return self._MultiNodeCall(node_list, "bdev_sizes", [devices])
@_RpcTimeout(_TMO_URGENT)
def call_lv_list(self, node_list, vg_name):
return self._SingleNodeCall(node, "bridges_exist", [bridges_list])
@_RpcTimeout(_TMO_NORMAL)
- def call_instance_start(self, node, instance, hvp, bep):
+ def call_instance_start(self, node, instance, hvp, bep, startup_paused):
"""Starts an instance.
This is a single-node call.
"""
idict = self._InstDict(instance, hvp=hvp, bep=bep)
- return self._SingleNodeCall(node, "instance_start", [idict])
+ return self._SingleNodeCall(node, "instance_start", [idict, startup_paused])
@_RpcTimeout(_TMO_NORMAL)
def call_instance_shutdown(self, node, instance, timeout):
[self._InstDict(instance), info, target])
@_RpcTimeout(_TMO_NORMAL)
- def call_finalize_migration(self, node, instance, info, success):
+ def call_instance_finalize_migration_dst(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
@param success: whether the migration was a success or a failure
"""
- return self._SingleNodeCall(node, "finalize_migration",
+ return self._SingleNodeCall(node, "instance_finalize_migration_dst",
[self._InstDict(instance), info, success])
@_RpcTimeout(_TMO_SLOW)
return self._SingleNodeCall(node, "instance_migrate",
[self._InstDict(instance), target, live])
+ @_RpcTimeout(_TMO_SLOW)
+ def call_instance_finalize_migration_src(self, node, instance, success, live):
+ """Finalize the instance migration on the source node.
+
+ This is a single-node call.
+
+ @type instance: L{objects.Instance}
+ @param instance: the instance that was migrated
+ @type success: bool
+ @param success: whether the migration succeeded or not
+ @type live: bool
+ @param live: whether the user requested a live migration or not
+
+ """
+ return self._SingleNodeCall(node, "instance_finalize_migration_src",
+ [self._InstDict(instance), success, live])
+
+ @_RpcTimeout(_TMO_SLOW)
+ def call_instance_get_migration_status(self, node, instance):
+ """Report migration status.
+
+ This is a single-node call that must be executed on the source node.
+
+ @type instance: L{objects.Instance}
+ @param instance: the instance that is being migrated
+ @rtype: L{objects.MigrationStatus}
+ @return: the status of the current migration (one of
+ L{constants.HV_MIGRATION_VALID_STATUSES}), plus any additional
+ progress info that can be retrieved from the hypervisor
+
+ """
+ result = self._SingleNodeCall(node, "instance_get_migration_status",
+ [self._InstDict(instance)])
+ if not result.fail_msg and result.payload is not None:
+ result.payload = objects.MigrationStatus.FromDict(result.payload)
+ return result
+
@_RpcTimeout(_TMO_NORMAL)
def call_instance_reboot(self, node, inst, reboot_type, shutdown_timeout):
"""Reboots an instance.
return self._MultiNodeCall(node_list, "instance_list", [hypervisor_list])
@_RpcTimeout(_TMO_FAST)
- 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])
-
- @_RpcTimeout(_TMO_FAST)
def call_node_has_ip_address(self, node, address):
"""Checks if a node has the given IP address.
@classmethod
@_RpcTimeout(_TMO_FAST)
- def call_node_start_master(cls, node, start_daemons, no_voting):
- """Tells a node to activate itself as a master.
+ def call_node_start_master_daemons(cls, node, no_voting):
+ """Starts master daemons on a node.
+
+ This is a single-node call.
+
+ """
+ return cls._StaticSingleNodeCall(node, "node_start_master_daemons",
+ [no_voting])
+
+ @classmethod
+ @_RpcTimeout(_TMO_FAST)
+ def call_node_activate_master_ip(cls, node):
+ """Activates master IP on a node.
+
+ This is a single-node call.
+
+ """
+ return cls._StaticSingleNodeCall(node, "node_activate_master_ip", [])
+
+ @classmethod
+ @_RpcTimeout(_TMO_FAST)
+ def call_node_stop_master(cls, node):
+ """Deactivates master IP and stops master daemons on a node.
+
+ This is a single-node call.
+
+ """
+ return cls._StaticSingleNodeCall(node, "node_stop_master", [])
+
+ @classmethod
+ @_RpcTimeout(_TMO_FAST)
+ def call_node_deactivate_master_ip(cls, node):
+ """Deactivates master IP on a node.
This is a single-node call.
"""
- return cls._StaticSingleNodeCall(node, "node_start_master",
- [start_daemons, no_voting])
+ return cls._StaticSingleNodeCall(node, "node_deactivate_master_ip", [])
@classmethod
@_RpcTimeout(_TMO_FAST)
- def call_node_stop_master(cls, node, stop_daemons):
- """Tells a node to demote itself from master status.
+ def call_node_change_master_netmask(cls, node, netmask):
+ """Change master IP netmask.
This is a single-node call.
"""
- return cls._StaticSingleNodeCall(node, "node_stop_master", [stop_daemons])
+ return cls._StaticSingleNodeCall(node, "node_change_master_netmask",
+ [netmask])
@classmethod
@_RpcTimeout(_TMO_URGENT)
"""
file_contents = utils.ReadFile(file_name)
- data = cls._Compress(file_contents)
+ data = _Compress(file_contents)
st = os.stat(file_name)
getents = runtime.GetEnts()
params = [file_name, data, st.st_mode, getents.LookupUid(st.st_uid),
return self._SingleNodeCall(node, "iallocator_runner", [name, idata])
@_RpcTimeout(_TMO_NORMAL)
- def call_blockdev_grow(self, node, cf_bdev, amount):
+ def call_blockdev_grow(self, node, cf_bdev, amount, dryrun):
"""Request a snapshot of the given block device.
This is a single-node call.
"""
return self._SingleNodeCall(node, "blockdev_grow",
- [cf_bdev.ToDict(), amount])
+ [cf_bdev.ToDict(), amount, dryrun])
@_RpcTimeout(_TMO_1DAY)
def call_blockdev_export(self, node, cf_bdev,
"""
return cls._StaticMultiNodeCall(node_list, "jobqueue_update",
- [file_name, cls._Compress(content)],
+ [file_name, _Compress(content)],
address_list=address_list)
@classmethod
return self._SingleNodeCall(node, "x509_cert_remove", [name])
@_RpcTimeout(_TMO_NORMAL)
- def call_import_start(self, node, opts, instance, dest, dest_args):
+ def call_import_start(self, node, opts, instance, component,
+ dest, dest_args):
"""Starts a listener for an import.
This is a single-node call.
@param node: Node name
@type instance: C{objects.Instance}
@param instance: Instance object
+ @type component: string
+ @param component: which part of the instance is being imported
"""
return self._SingleNodeCall(node, "import_start",
[opts.ToDict(),
- self._InstDict(instance), dest,
+ self._InstDict(instance), component, dest,
_EncodeImportExportIO(dest, dest_args)])
@_RpcTimeout(_TMO_NORMAL)
def call_export_start(self, node, opts, host, port,
- instance, source, source_args):
+ instance, component, source, source_args):
"""Starts an export daemon.
This is a single-node call.
@param node: Node name
@type instance: C{objects.Instance}
@param instance: Instance object
+ @type component: string
+ @param component: which part of the instance is being imported
"""
return self._SingleNodeCall(node, "export_start",
[opts.ToDict(), host, port,
- self._InstDict(instance), source,
+ self._InstDict(instance),
+ component, source,
_EncodeImportExportIO(source, source_args)])
@_RpcTimeout(_TMO_FAST)
projects / ganeti-local / blobdiff