X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/a42872ff4ca20a0eaab9fcf9ba6d5999640094d6..c8fcde472922e4ee664d904e0bf1a583f1d5040d:/daemons/ganeti-masterd diff --git a/daemons/ganeti-masterd b/daemons/ganeti-masterd index 1c884bb..2305fc6 100755 --- a/daemons/ganeti-masterd +++ b/daemons/ganeti-masterd @@ -27,6 +27,8 @@ inheritance from parent classes requires it. """ +import os +import errno import sys import SocketServer import time @@ -34,7 +36,6 @@ import collections import Queue import random import signal -import simplejson import logging from cStringIO import StringIO @@ -50,9 +51,10 @@ from ganeti import luxi from ganeti import utils from ganeti import errors from ganeti import ssconf -from ganeti import logger from ganeti import workerpool from ganeti import rpc +from ganeti import bootstrap +from ganeti import serializer CLIENT_REQUEST_WORKERS = 16 @@ -87,9 +89,8 @@ class IOServer(SocketServer.UnixStreamServer): def __init__(self, address, rqhandler): """IOServer constructor - Args: - address: the address to bind this IOServer to - rqhandler: RequestHandler type object + @param address: the address to bind this IOServer to + @param rqhandler: RequestHandler type object """ SocketServer.UnixStreamServer.__init__(self, address, rqhandler) @@ -148,10 +149,10 @@ class ClientRqHandler(SocketServer.BaseRequestHandler): while True: msg = self.read_message() if msg is None: - logging.info("client closed connection") + logging.debug("client closed connection") break - request = simplejson.loads(msg) + request = serializer.LoadJson(msg) logging.debug("request: %s", request) if not isinstance(request, dict): logging.error("wrong request received: %s", msg) @@ -167,6 +168,9 @@ class ClientRqHandler(SocketServer.BaseRequestHandler): try: result = self._ops.handle_request(method, args) success = True + except errors.GenericError, err: + success = False + result = (err.__class__.__name__, err.args) except: logging.error("Unexpected exception", exc_info=True) err = sys.exc_info() @@ -177,7 +181,7 @@ class ClientRqHandler(SocketServer.BaseRequestHandler): luxi.KEY_RESULT: result, } logging.debug("response: %s", response) - self.send_message(simplejson.dumps(response)) + self.send_message(serializer.DumpJson(response)) def read_message(self): while not self._msgs: @@ -205,48 +209,94 @@ class ClientOps: # TODO: Parameter validation if method == luxi.REQ_SUBMIT_JOB: + logging.info("Received new job") ops = [opcodes.OpCode.LoadOpCode(state) for state in args] return queue.SubmitJob(ops) + if method == luxi.REQ_SUBMIT_MANY_JOBS: + logging.info("Received multiple jobs") + jobs = [] + for ops in args: + jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops]) + return queue.SubmitManyJobs(jobs) + elif method == luxi.REQ_CANCEL_JOB: job_id = args + logging.info("Received job cancel request for %s", job_id) return queue.CancelJob(job_id) elif method == luxi.REQ_ARCHIVE_JOB: job_id = args + logging.info("Received job archive request for %s", job_id) return queue.ArchiveJob(job_id) + elif method == luxi.REQ_AUTOARCHIVE_JOBS: + (age, timeout) = args + logging.info("Received job autoarchive request for age %s, timeout %s", + age, timeout) + return queue.AutoArchiveJobs(age, timeout) + elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE: (job_id, fields, prev_job_info, prev_log_serial, timeout) = args + logging.info("Received job poll request for %s", job_id) return queue.WaitForJobChanges(job_id, fields, prev_job_info, prev_log_serial, timeout) elif method == luxi.REQ_QUERY_JOBS: (job_ids, fields) = args + if isinstance(job_ids, (tuple, list)) and job_ids: + msg = ", ".join(job_ids) + else: + msg = str(job_ids) + logging.info("Received job query request for %s", msg) return queue.QueryJobs(job_ids, fields) elif method == luxi.REQ_QUERY_INSTANCES: - (names, fields) = args - op = opcodes.OpQueryInstances(names=names, output_fields=fields) + (names, fields, use_locking) = args + logging.info("Received instance query request for %s", names) + if use_locking: + raise errors.OpPrereqError("Sync queries are not allowed") + op = opcodes.OpQueryInstances(names=names, output_fields=fields, + use_locking=use_locking) return self._Query(op) elif method == luxi.REQ_QUERY_NODES: - (names, fields) = args - op = opcodes.OpQueryNodes(names=names, output_fields=fields) + (names, fields, use_locking) = args + logging.info("Received node query request for %s", names) + if use_locking: + raise errors.OpPrereqError("Sync queries are not allowed") + op = opcodes.OpQueryNodes(names=names, output_fields=fields, + use_locking=use_locking) return self._Query(op) elif method == luxi.REQ_QUERY_EXPORTS: - nodes = args - op = opcodes.OpQueryExports(nodes=nodes) + nodes, use_locking = args + if use_locking: + raise errors.OpPrereqError("Sync queries are not allowed") + logging.info("Received exports query request") + op = opcodes.OpQueryExports(nodes=nodes, use_locking=use_locking) return self._Query(op) elif method == luxi.REQ_QUERY_CONFIG_VALUES: fields = args + logging.info("Received config values query request for %s", fields) op = opcodes.OpQueryConfigValues(output_fields=fields) return self._Query(op) + elif method == luxi.REQ_QUERY_CLUSTER_INFO: + logging.info("Received cluster info query request") + op = opcodes.OpQueryClusterInfo() + return self._Query(op) + + elif method == luxi.REQ_QUEUE_SET_DRAIN_FLAG: + drain_flag = args + logging.info("Received queue drain flag change request to %s", + drain_flag) + return queue.SetDrainFlag(drain_flag) + else: - raise ValueError("Invalid operation") + logging.info("Received invalid request '%s'", method) + raise ValueError("Invalid operation '%s'" % method) def _DummyLog(self, *args): pass @@ -257,7 +307,7 @@ class ClientOps: """ proc = mcpu.Processor(self.server.context) # TODO: Where should log messages go? - return proc.ExecOpCode(op, self._DummyLog) + return proc.ExecOpCode(op, self._DummyLog, None) class GanetiContext(object): @@ -306,7 +356,7 @@ class GanetiContext(object): self.cfg.AddNode(node) # If preseeding fails it'll not be added - self.jobqueue.AddNode(node.name) + self.jobqueue.AddNode(node) # Add the new node to the Ganeti Lock Manager self.glm.add(locking.LEVEL_NODE, node.name) @@ -316,7 +366,7 @@ class GanetiContext(object): """ # Synchronize the queue again - self.jobqueue.AddNode(node.name) + self.jobqueue.AddNode(node) def RemoveNode(self, name): """Removes a node from the configuration and lock manager. @@ -335,8 +385,7 @@ class GanetiContext(object): def ParseOptions(): """Parse the command line options. - Returns: - (options, args) as from OptionParser.parse_args() + @return: (options, args) as from OptionParser.parse_args() """ parser = OptionParser(description="Ganeti master daemon", @@ -350,6 +399,10 @@ def ParseOptions(): parser.add_option("-d", "--debug", dest="debug", help="Enable some debug messages", default=False, action="store_true") + parser.add_option("--no-voting", dest="no_voting", + help="Do not check that the nodes agree on this node" + " being the master and start the daemon unconditionally", + default=False, action="store_true") options, args = parser.parse_args() return options, args @@ -363,53 +416,52 @@ def CheckAgreement(): future we could collect the current node list from our (possibly obsolete) known nodes. + In order to account for cold-start of all nodes, we retry for up to + a minute until we get a real answer as the top-voted one. If the + nodes are more out-of-sync, for now manual startup of the master + should be attempted. + + Note that for a even number of nodes cluster, we need at least half + of the nodes (beside ourselves) to vote for us. This creates a + problem on two-node clusters, since in this case we require the + other node to be up too to confirm our status. + """ myself = utils.HostInfo().name #temp instantiation of a config writer, used only to get the node list cfg = config.ConfigWriter() node_list = cfg.GetNodeList() del cfg - try: - node_list.remove(myself) - except KeyError: - pass - if not node_list: - # either single node cluster, or a misconfiguration, but I won't - # break any other node, so I can proceed - return True - results = rpc.call_master_info(node_list) - if not isinstance(results, dict): - # this should not happen (unless internal error in rpc) - logging.critical("Can't complete rpc call, aborting master startup") - return False - positive = negative = 0 - other_masters = {} - for node in results: - if not isinstance(results[node], (tuple, list)) or len(results[node]) < 3: - logging.warning("Can't contact node %s", node) + retries = 6 + while retries > 0: + votes = bootstrap.GatherMasterVotes(node_list) + if not votes: + # empty node list, this is a one node cluster + return True + if votes[0][0] is None: + retries -= 1 + time.sleep(10) continue - master_node = results[node][2] - if master_node == myself: - positive += 1 - else: - negative += 1 - if not master_node in other_masters: - other_masters[master_node] = 0 - other_masters[master_node] += 1 - if positive <= negative: - # bad! - logging.critical("It seems we are not the master (%d votes for," - " %d votes against)", positive, negative) - if len(other_masters) > 1: - logging.critical("The other nodes do not agree on a single master") - elif other_masters: - # TODO: resync my files from the master - logging.critical("It seems the real master is %s", - other_masters.keys()[0]) - else: - logging.critical("Can't contact any node for data, aborting startup") + break + if retries == 0: + logging.critical("Cluster inconsistent, most of the nodes didn't answer" + " after multiple retries. Aborting startup") return False - return True + # here a real node is at the top of the list + all_votes = sum(item[1] for item in votes) + top_node, top_votes = votes[0] + result = False + if top_node != myself: + logging.critical("It seems we are not the master (top-voted node" + " is %s with %d out of %d votes)", top_node, top_votes, + all_votes) + elif top_votes < all_votes - top_votes: + logging.critical("It seems we are not the master (%d votes for," + " %d votes against)", top_votes, all_votes - top_votes) + else: + result = True + + return result def main(): @@ -419,37 +471,68 @@ def main(): utils.debug = options.debug utils.no_fork = True - ssconf.CheckMaster(options.debug) + if options.fork: + utils.CloseFDs() + + rpc.Init() + try: + ssconf.CheckMaster(options.debug) + + # we believe we are the master, let's ask the other nodes... + if options.no_voting: + sys.stdout.write("The 'no voting' option has been selected.\n") + sys.stdout.write("This is dangerous, please confirm by" + " typing uppercase 'yes': ") + sys.stdout.flush() + confirmation = sys.stdin.readline().strip() + if confirmation != "YES": + print "Aborting." + return + else: + if not CheckAgreement(): + return + + dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE), + (constants.SOCKET_DIR, constants.SOCKET_DIR_MODE), + ] + utils.EnsureDirs(dirs) - # we believe we are the master, let's ask the other nodes... - if not CheckAgreement(): - return + # This is safe to do as the pid file guarantees against + # concurrent execution. + utils.RemoveFile(constants.MASTER_SOCKET) - master = IOServer(constants.MASTER_SOCKET, ClientRqHandler) + master = IOServer(constants.MASTER_SOCKET, ClientRqHandler) + finally: + rpc.Shutdown() # become a daemon if options.fork: - utils.Daemonize(logfile=constants.LOG_MASTERDAEMON, - noclose_fds=[master.fileno()]) + utils.Daemonize(logfile=constants.LOG_MASTERDAEMON) utils.WritePidFile(constants.MASTERD_PID) + try: + utils.SetupLogging(constants.LOG_MASTERDAEMON, debug=options.debug, + stderr_logging=not options.fork, multithreaded=True) - logger.SetupLogging(constants.LOG_MASTERDAEMON, debug=options.debug, - stderr_logging=not options.fork) - - logging.info("ganeti master daemon startup") + logging.info("Ganeti master daemon startup") - # activate ip - master_node = ssconf.SimpleConfigReader().GetMasterNode() - if not rpc.call_node_start_master(master_node, False): - logging.error("Can't activate master IP address") + rpc.Init() + try: + # activate ip + master_node = ssconf.SimpleConfigReader().GetMasterNode() + if not rpc.RpcRunner.call_node_start_master(master_node, False): + logging.error("Can't activate master IP address") - master.setup_queue() - try: - master.serve_forever() + master.setup_queue() + try: + master.serve_forever() + finally: + master.server_cleanup() + finally: + rpc.Shutdown() finally: - master.server_cleanup() utils.RemovePidFile(constants.MASTERD_PID) + utils.RemoveFile(constants.MASTER_SOCKET) if __name__ == "__main__":