ClusterShell.Propagation

1 #!/usr/bin/env python 2 # 3 # Copyright (C) 2010-2016 CEA/DAM 4 # Copyright (C) 2010-2011 Henri Doreau <henri.doreau@cea.fr> 5 # Copyright (C) 2015-2016 Stephane Thiell <sthiell@stanford.edu> 6 # 7 # This file is part of ClusterShell. 8 # 9 # ClusterShell is free software; you can redistribute it and/or 10 # modify it under the terms of the GNU Lesser General Public 11 # License as published by the Free Software Foundation; either 12 # version 2.1 of the License, or (at your option) any later version. 13 # 14 # ClusterShell is distributed in the hope that it will be useful, 15 # but WITHOUT ANY WARRANTY; without even the implied warranty of 16 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 17 # Lesser General Public License for more details. 18 # 19 # You should have received a copy of the GNU Lesser General Public 20 # License along with ClusterShell; if not, write to the Free Software 21 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 22 23 """ 24 ClusterShell Propagation module. Use the topology tree to send commands 25 through gateways and gather results. 26 """ 27 28 from collections import deque 29 import logging 30 31 from ClusterShell.Defaults import DEFAULTS 32 from ClusterShell.NodeSet import NodeSet 33 from ClusterShell.Communication import Channel 34 from ClusterShell.Communication import ControlMessage, StdOutMessage 35 from ClusterShell.Communication import StdErrMessage, RetcodeMessage 36 from ClusterShell.Communication import StartMessage, EndMessage 37 from ClusterShell.Communication import RoutedMessageBase, ErrorMessage 38 from ClusterShell.Communication import ConfigurationMessage, TimeoutMessage 39 from ClusterShell.Topology import TopologyError 40 41

42 -class RouteResolvingError(Exception):

43 """error raised on invalid conditions during routing operations"""

44

45 -class PropagationTreeRouter(object):

46 """performs routes resolving operations within a propagation tree. 47 This object provides a next_hop method, that will look for the best 48 directly connected node to use to forward a message to a remote 49 node. 50 51 Upon instanciation, the router will parse the topology tree to 52 generate its routing table. 53 """

54 - def __init__(self, root, topology, fanout=0):

55 self.root = root 56 self.topology = topology 57 self.fanout = fanout 58 self.nodes_fanin = {} 59 self.table = None 60 61 self.table_generate(root, topology) 62 self._unreachable_hosts = NodeSet()

63

64 - def table_generate(self, root, topology):

65 """The router relies on a routing table. The keys are the 66 destination nodes and the values are the next hop gateways to 67 use to reach these nodes. 68 """ 69 self.table = {} 70 try: 71 root_group = topology.find_nodegroup(root) 72 except TopologyError: 73 msgfmt = "Invalid root or gateway node: %s" 74 raise RouteResolvingError(msgfmt % root) 75 76 for group in root_group.children(): 77 self.table[group.nodeset] = NodeSet() 78 stack = [group] 79 while len(stack) > 0: 80 curr = stack.pop() 81 self.table[group.nodeset].add(curr.children_ns()) 82 stack += curr.children() 83 84 # reverse table (it was crafted backward) 85 self.table = dict((v, k) for k, v in self.table.iteritems())

86

87 - def dispatch(self, dst):

88 """dispatch nodes from a target nodeset to the directly 89 connected gateways. 90 91 The method acts as an iterator, returning a gateway and the 92 associated hosts. It should provide a rather good load balancing 93 between the gateways. 94 """ 95 ### Disabled to handle all remaining nodes as directly connected nodes 96 ## Check for directly connected targets 97 #res = [tmp & dst for tmp in self.table.values()] 98 #nexthop = NodeSet() 99 #[nexthop.add(x) for x in res] 100 #if len(nexthop) > 0: 101 # yield nexthop, nexthop 102 103 # Check for remote targets, that require a gateway to be reached 104 for network in self.table.iterkeys(): 105 dst_inter = network & dst 106 dst.difference_update(dst_inter) 107 for host in dst_inter.nsiter(): 108 yield self.next_hop(host), host 109 110 # remaining nodes are considered as directly connected nodes 111 if dst: 112 yield dst, dst

113

114 - def next_hop(self, dst):

115 """perform the next hop resolution. If several hops are 116 available, then, the one with the least number of current jobs 117 will be returned 118 """ 119 if dst in self._unreachable_hosts: 120 raise RouteResolvingError( 121 'Invalid destination: %s, host is unreachable' % dst) 122 123 # can't resolve if source == destination 124 if self.root == dst: 125 raise RouteResolvingError( 126 'Invalid resolution request: %s -> %s' % (self.root, dst)) 127 128 ## ------------------ 129 # the routing table is organized this way: 130 # 131 # NETWORK | NEXT HOP 132 # ------------+----------- 133 # node[0-9] | gateway0 134 # node[10-19] | gateway[1-2] 135 # ... 136 # --------- 137 for network, nexthops in self.table.iteritems(): 138 # destination contained in current network 139 if dst in network: 140 res = self._best_next_hop(nexthops) 141 if res is None: 142 raise RouteResolvingError('No route available to %s' % \ 143 str(dst)) 144 self.nodes_fanin[res] += len(dst) 145 return res 146 # destination contained in current next hops (ie. directly 147 # connected) 148 if dst in nexthops: 149 return dst 150 151 raise RouteResolvingError( 152 'No route from %s to host %s' % (self.root, dst))

153

154 - def mark_unreachable(self, dst):

155 """mark node dst as unreachable and don't advertise routes 156 through it anymore. The cache will be updated only when 157 necessary to avoid performing expensive traversals. 158 """ 159 # Simply mark dst as unreachable in a dedicated NodeSet. This 160 # list will be consulted by the resolution method 161 self._unreachable_hosts.add(dst)

162

163 - def _best_next_hop(self, candidates):

164 """find out a good next hop gateway""" 165 backup = None 166 backup_connections = 1e400 # infinity 167 168 candidates = candidates.difference(self._unreachable_hosts) 169 170 for host in candidates: 171 # the router tracks established connections in the 172 # nodes_fanin table to avoid overloading a gateway 173 connections = self.nodes_fanin.setdefault(host, 0) 174 # FIXME 175 #if connections < self.fanout: 176 # # currently, the first one is the best 177 # return host 178 if backup_connections > connections: 179 backup = host 180 backup_connections = connections 181 return backup

182 183

184 -class PropagationChannel(Channel):

185 """Admin node propagation logic. Instances are able to handle 186 incoming messages from a directly connected gateway, process them 187 and reply. 188 189 In order to take decisions, the instance acts as a finite states 190 machine, whose current state evolves according to received data. 191 192 -- INTERNALS -- 193 Instance can be in one of the 4 different states: 194 - init (implicit) 195 This is the very first state. The instance enters the init 196 state at start() method, and will then send the configuration 197 to the remote node. Once the configuration is sent away, the 198 state changes to cfg. 199 200 - cfg 201 During this second state, the instance will wait for a valid 202 acknowledgement from the gateway to the previously sent 203 configuration message. If such a message is delivered, the 204 control message (the one that contains the actions to perform) 205 is sent, and the state is set to ctl. 206 207 - ctl 208 Third state, the instance is waiting for a valid ack for from 209 the gateway to the ctl packet. Then, the state switch to gtr 210 (gather). 211 212 - gtr 213 Final state: wait for results from the subtree and store them. 214 """

215 - def __init__(self, task, gateway):

216 """ 217 """ 218 Channel.__init__(self) 219 self.task = task 220 self.gateway = gateway 221 self.workers = {} 222 self._cfg_write_hist = deque() # track write requests 223 self._sendq = deque() 224 self._rc = None 225 self.logger = logging.getLogger(__name__)

226

227 - def send_queued(self, ctl):

228 """helper used to send a message, using msg queue if needed""" 229 if self.setup and not self._sendq: 230 # send now if channel is setup and sendq empty 231 self.send(ctl) 232 else: 233 self.logger.debug("send_queued: %d", len(self._sendq)) 234 self._sendq.appendleft(ctl)

235

236 - def send_dequeue(self):

237 """helper used to send one queued message (if any)""" 238 if self._sendq: 239 ctl = self._sendq.pop() 240 self.logger.debug("dequeuing sendq: %s", ctl) 241 self.send(ctl)

242

243 - def start(self):

244 """start propagation channel""" 245 self._init() 246 self._open() 247 # Immediately send CFG 248 cfg = ConfigurationMessage(self.gateway) 249 cfg.data_encode(self.task.topology) 250 self.send(cfg)

251

252 - def recv(self, msg):

253 """process incoming messages""" 254 self.logger.debug("recv: %s", msg) 255 if msg.type == EndMessage.ident: 256 #??#self.ptree.notify_close() 257 self.logger.debug("got EndMessage; closing") 258 # abort worker (now working) 259 self.worker.abort() 260 elif self.setup: 261 self.recv_ctl(msg) 262 elif self.opened: 263 self.recv_cfg(msg) 264 elif msg.type == StartMessage.ident: 265 self.opened = True 266 self.logger.debug('channel started (version %s on remote gateway)', 267 self._xml_reader.version) 268 else: 269 self.logger.error('unexpected message: %s', str(msg))

270

271 - def shell(self, nodes, command, worker, timeout, stderr, gw_invoke_cmd, 272 remote):

273 """command execution through channel""" 274 self.logger.debug("shell nodes=%s timeout=%s worker=%s remote=%s", 275 nodes, timeout, id(worker), remote) 276 277 self.workers[id(worker)] = worker 278 279 ctl = ControlMessage(id(worker)) 280 ctl.action = 'shell' 281 ctl.target = nodes 282 283 # keep only valid task info pairs 284 info = dict((k, v) for k, v in self.task._info.items() 285 if k not in DEFAULTS._task_info_pkeys_bl) 286 287 ctl_data = { 288 'cmd': command, 289 'invoke_gateway': gw_invoke_cmd, # XXX 290 'taskinfo': info, 291 'stderr': stderr, 292 'timeout': timeout, 293 'remote': remote, 294 } 295 ctl.data_encode(ctl_data) 296 self.send_queued(ctl)

297

298 - def write(self, nodes, buf, worker):

299 """write buffer through channel to nodes on standard input""" 300 self.logger.debug("write buflen=%d", len(buf)) 301 assert id(worker) in self.workers 302 303 ctl = ControlMessage(id(worker)) 304 ctl.action = 'write' 305 ctl.target = nodes 306 307 ctl_data = { 308 'buf': buf, 309 } 310 ctl.data_encode(ctl_data) 311 self._cfg_write_hist.appendleft((ctl.msgid, nodes, len(buf), worker)) 312 self.send_queued(ctl)

313

314 - def set_write_eof(self, nodes, worker):

315 """send EOF through channel to specified nodes""" 316 self.logger.debug("set_write_eof") 317 assert id(worker) in self.workers 318 319 ctl = ControlMessage(id(worker)) 320 ctl.action = 'eof' 321 ctl.target = nodes 322 self.send_queued(ctl)

323

324 - def recv_cfg(self, msg):

325 """handle incoming messages for state 'propagate configuration'""" 326 self.logger.debug("recv_cfg") 327 if msg.type == 'ACK': 328 self.logger.debug("CTL - connection with gateway fully established") 329 self.setup = True 330 self.send_dequeue() 331 else: 332 self.logger.debug("_state_config error (msg=%s)", msg)

333

334 - def recv_ctl(self, msg):

335 """handle incoming messages for state 'control'""" 336 if msg.type == 'ACK': 337 self.logger.debug("got ack (%s)", msg.type) 338 # check if ack matches write history msgid to generate ev_written 339 if self._cfg_write_hist and msg.ack == self._cfg_write_hist[-1][0]: 340 _, nodes, bytes_count, metaworker = self._cfg_write_hist.pop() 341 for node in nodes: 342 # we are losing track of the gateway here, we could override 343 # on_written in WorkerTree if needed (eg. for stats) 344 metaworker._on_written(node, bytes_count, 'stdin') 345 self.send_dequeue() 346 elif isinstance(msg, RoutedMessageBase): 347 metaworker = self.workers[msg.srcid] 348 if msg.type == StdOutMessage.ident: 349 nodeset = NodeSet(msg.nodes) 350 decoded = msg.data_decode() + '\n' 351 for line in decoded.splitlines(): 352 for node in nodeset: 353 metaworker._on_remote_node_msgline(node, line, 'stdout', 354 self.gateway) 355 elif msg.type == StdErrMessage.ident: 356 nodeset = NodeSet(msg.nodes) 357 decoded = msg.data_decode() + '\n' 358 for line in decoded.splitlines(): 359 for node in nodeset: 360 metaworker._on_remote_node_msgline(node, line, 'stderr', 361 self.gateway) 362 elif msg.type == RetcodeMessage.ident: 363 rc = msg.retcode 364 for node in NodeSet(msg.nodes): 365 metaworker._on_remote_node_rc(node, rc, self.gateway) 366 elif msg.type == TimeoutMessage.ident: 367 self.logger.debug("TimeoutMessage for %s", msg.nodes) 368 for node in NodeSet(msg.nodes): 369 metaworker._on_remote_node_timeout(node, self.gateway) 370 elif msg.type == ErrorMessage.ident: 371 # tree runtime error, could generate a new event later 372 raise TopologyError("%s: %s" % (self.gateway, msg.reason)) 373 else: 374 self.logger.debug("recv_ctl: unhandled msg %s", msg) 375 """ 376 return 377 if self.ptree.upchannel is not None: 378 self.logger.debug("_state_gather ->upchan %s" % msg) 379 self.ptree.upchannel.send(msg) # send to according event handler passed by shell() 380 else: 381 assert False 382 """

383

384 - def ev_hup(self, worker):

385 """Channel command is closing""" 386 self._rc = worker.current_rc

387

388 - def ev_close(self, worker):

389 """Channel is closing""" 390 # do not use worker buffer or rc accessors here as we doesn't use 391 # common stream names 392 gateway = str(worker.nodes) 393 self.logger.debug("ev_close gateway=%s %s", gateway, self) 394 self.logger.debug("ev_close rc=%s", self._rc) # may be None 395 396 if self._rc: # got explicit error code 397 # ev_routing? 398 self.logger.debug("unreachable gateway %s", gateway) 399 worker.task.router.mark_unreachable(gateway) 400 self.logger.debug("worker.task.gateways=%s", worker.task.gateways)

401 # TODO: find best gateway, update WorkerTree counters, relaunch... 402

Source Code for Module ClusterShell.Propagation