import sys
import SocketServer
-import threading
import time
import collections
import Queue
from ganeti import errors
from ganeti import ssconf
from ganeti import logger
+from ganeti import workerpool
+from ganeti import rpc
+CLIENT_REQUEST_WORKERS = 16
+
EXIT_NOTMASTER = constants.EXIT_NOTMASTER
EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR
+class ClientRequestWorker(workerpool.BaseWorker):
+ def RunTask(self, server, request, client_address):
+ """Process the request.
+
+ This is copied from the code in ThreadingMixIn.
+
+ """
+ try:
+ server.finish_request(request, client_address)
+ server.close_request(request)
+ except:
+ server.handle_error(request, client_address)
+ server.close_request(request)
+
+
class IOServer(SocketServer.UnixStreamServer):
"""IO thread class.
cleanup at shutdown.
"""
- QUEUE_PROCESSOR_SIZE = 5
-
- def __init__(self, address, rqhandler, context):
+ def __init__(self, address, rqhandler):
"""IOServer constructor
Args:
address: the address to bind this IOServer to
rqhandler: RequestHandler type object
- context: Context Object common to all worker threads
"""
SocketServer.UnixStreamServer.__init__(self, address, rqhandler)
- self.do_quit = False
- self.queue = jqueue.QueueManager()
- self.context = context
- self.processors = []
# We'll only start threads once we've forked.
- self.jobqueue = None
-
- signal.signal(signal.SIGINT, self.handle_quit_signals)
- signal.signal(signal.SIGTERM, self.handle_quit_signals)
+ self.context = None
+ self.request_workers = None
def setup_queue(self):
- self.jobqueue = jqueue.JobQueue(self.context)
-
- def setup_processors(self):
- """Spawn the processors threads.
-
- This initializes the queue and the thread processors. It is done
- separately from the constructor because we want the clone()
- syscalls to happen after the daemonize part.
-
- """
- for i in range(self.QUEUE_PROCESSOR_SIZE):
- self.processors.append(threading.Thread(target=PoolWorker,
- args=(i, self.queue.new_queue,
- self.context)))
- for t in self.processors:
- t.start()
-
- def process_request_thread(self, request, client_address):
- """Process the request.
-
- This is copied from the code in ThreadingMixIn.
-
- """
- try:
- self.finish_request(request, client_address)
- self.close_request(request)
- except:
- self.handle_error(request, client_address)
- self.close_request(request)
+ self.context = GanetiContext()
+ self.request_workers = workerpool.WorkerPool(CLIENT_REQUEST_WORKERS,
+ ClientRequestWorker)
def process_request(self, request, client_address):
- """Start a new thread to process the request.
-
- This is copied from the coode in ThreadingMixIn.
+ """Add task to workerpool to process request.
"""
- t = threading.Thread(target=self.process_request_thread,
- args=(request, client_address))
- t.start()
-
- def handle_quit_signals(self, signum, frame):
- print "received %s in %s" % (signum, frame)
- self.do_quit = True
+ self.request_workers.AddTask(self, request, client_address)
def serve_forever(self):
"""Handle one request at a time until told to quit."""
- while not self.do_quit:
- self.handle_request()
- print "served request, quit=%s" % (self.do_quit)
+ sighandler = utils.SignalHandler([signal.SIGINT, signal.SIGTERM])
+ try:
+ while not sighandler.called:
+ self.handle_request()
+ finally:
+ sighandler.Reset()
def server_cleanup(self):
"""Cleanup the server.
"""
try:
self.server_close()
- utils.RemoveFile(constants.MASTER_SOCKET)
- for i in range(self.QUEUE_PROCESSOR_SIZE):
- self.queue.new_queue.put(None)
- for idx, t in enumerate(self.processors):
- logging.debug("waiting for processor thread %s...", idx)
- t.join()
- logging.debug("threads done")
finally:
- if self.jobqueue:
- self.jobqueue.Shutdown()
+ if self.request_workers:
+ self.request_workers.TerminateWorkers()
+ if self.context:
+ self.context.jobqueue.Shutdown()
class ClientRqHandler(SocketServer.BaseRequestHandler):
while True:
msg = self.read_message()
if msg is None:
- print "client closed connection"
+ logging.info("client closed connection")
break
+
request = simplejson.loads(msg)
+ logging.debug("request: %s", request)
if not isinstance(request, dict):
- print "wrong request received: %s" % msg
+ logging.error("wrong request received: %s", msg)
break
- method = request.get('request', None)
- data = request.get('data', None)
- if method is None or data is None:
- print "no method or data in request"
+
+ method = request.get(luxi.KEY_METHOD, None)
+ args = request.get(luxi.KEY_ARGS, None)
+ if method is None or args is None:
+ logging.error("no method or args in request")
break
- print "request:", method, data
- result = self._ops.handle_request(method, data)
- print "result:", result
- self.send_message(simplejson.dumps({'success': True, 'result': result}))
+
+ success = False
+ try:
+ result = self._ops.handle_request(method, args)
+ success = True
+ except:
+ logging.error("Unexpected exception", exc_info=True)
+ err = sys.exc_info()
+ result = "Caught exception: %s" % str(err[1])
+
+ response = {
+ luxi.KEY_SUCCESS: success,
+ luxi.KEY_RESULT: result,
+ }
+ logging.debug("response: %s", response)
+ self.send_message(simplejson.dumps(response))
def read_message(self):
while not self._msgs:
"""Class holding high-level client operations."""
def __init__(self, server):
self.server = server
- self._cpu = None
-
- def _getcpu(self):
- if self._cpu is None:
- self._cpu = mcpu.Processor(lambda x: None)
- return self._cpu
-
- def handle_request(self, operation, args):
- print operation, args
- if operation == "submit":
- return self.put(args)
- elif operation == "query":
- return self.query(args)
- else:
- raise ValueError("Invalid operation")
- def put(self, args):
- job = luxi.UnserializeJob(args)
- rid = self.server.queue.put(job)
- return rid
-
- def query(self, args):
- path = args["object"]
- fields = args["fields"]
- names = args["names"]
- if path == "instances":
- opclass = opcodes.OpQueryInstances
- elif path == "jobs":
- # early exit because job query-ing is special (not via opcodes)
- return self.query_jobs(fields, names)
- else:
- raise ValueError("Invalid object %s" % path)
+ def handle_request(self, method, args):
+ queue = self.server.context.jobqueue
- op = opclass(output_fields = fields, names=names)
- cpu = self._getcpu()
- result = cpu.ExecOpCode(op)
- return result
+ # TODO: Parameter validation
- def query_jobs(self, fields, names):
- return self.server.queue.query_jobs(fields, names)
+ if method == luxi.REQ_SUBMIT_JOB:
+ ops = [opcodes.OpCode.LoadOpCode(state) for state in args]
+ return queue.SubmitJob(ops)
+ elif method == luxi.REQ_CANCEL_JOB:
+ job_id = args
+ return queue.CancelJob(job_id)
-def JobRunner(proc, job, context):
- """Job executor.
+ elif method == luxi.REQ_ARCHIVE_JOB:
+ job_id = args
+ return queue.ArchiveJob(job_id)
- This functions processes a single job in the context of given
- processor instance.
+ elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE:
+ (job_id, fields, prev_job_info, prev_log_serial, timeout) = args
+ return queue.WaitForJobChanges(job_id, fields, prev_job_info,
+ prev_log_serial, timeout)
- Args:
- proc: Ganeti Processor to run the job on
- job: The job to run (unserialized format)
- context: Ganeti shared context
+ elif method == luxi.REQ_QUERY_JOBS:
+ (job_ids, fields) = args
+ return queue.QueryJobs(job_ids, fields)
- """
- job.SetStatus(opcodes.Job.STATUS_RUNNING)
- fail = False
- for idx, op in enumerate(job.data.op_list):
- job.data.op_status[idx] = opcodes.Job.STATUS_RUNNING
- try:
- job.data.op_result[idx] = proc.ExecOpCode(op)
- job.data.op_status[idx] = opcodes.Job.STATUS_SUCCESS
- except (errors.OpPrereqError, errors.OpExecError), err:
- fail = True
- job.data.op_result[idx] = str(err)
- job.data.op_status[idx] = opcodes.Job.STATUS_FAIL
- if fail:
- job.SetStatus(opcodes.Job.STATUS_FAIL)
- else:
- job.SetStatus(opcodes.Job.STATUS_SUCCESS)
+ elif method == luxi.REQ_QUERY_INSTANCES:
+ (names, fields) = args
+ op = opcodes.OpQueryInstances(names=names, output_fields=fields)
+ return self._Query(op)
+ elif method == luxi.REQ_QUERY_NODES:
+ (names, fields) = args
+ op = opcodes.OpQueryNodes(names=names, output_fields=fields)
+ return self._Query(op)
-def PoolWorker(worker_id, incoming_queue, context):
- """A worker thread function.
+ elif method == luxi.REQ_QUERY_EXPORTS:
+ nodes = args
+ op = opcodes.OpQueryExports(nodes=nodes)
+ return self._Query(op)
- This is the actual processor of a single thread of Job execution.
+ else:
+ raise ValueError("Invalid operation")
- Args:
- worker_id: the unique id for this worker
- incoming_queue: a queue to get jobs from
- context: the common server context, containing all shared data and
- synchronization structures.
+ def _DummyLog(self, *args):
+ pass
- """
- while True:
- logging.debug("worker %s sleeping", worker_id)
- item = incoming_queue.get(True)
- if item is None:
- break
- logging.debug("worker %s processing job %s", worker_id, item.data.job_id)
- proc = mcpu.Processor(context, feedback=lambda x: None)
- try:
- JobRunner(proc, item, context)
- except errors.GenericError, err:
- msg = "ganeti exception"
- logging.error(msg, exc_info=err)
- item.SetStatus(opcodes.Job.STATUS_FAIL, result=[msg])
- except Exception, err:
- msg = "unhandled exception"
- logging.error(msg, exc_info=err)
- item.SetStatus(opcodes.Job.STATUS_FAIL, result=[msg])
- except:
- msg = "unhandled unknown exception"
- logging.error(msg, exc_info=True)
- item.SetStatus(opcodes.Job.STATUS_FAIL, result=[msg])
- logging.debug("worker %s finish job %s", worker_id, item.data.job_id)
- logging.debug("worker %s exiting", worker_id)
+ def _Query(self, op):
+ """Runs the specified opcode and returns the result.
+
+ """
+ proc = mcpu.Processor(self.server.context)
+ # TODO: Where should log messages go?
+ return proc.ExecOpCode(op, self._DummyLog)
class GanetiContext(object):
"""
assert self.__class__._instance is None, "double GanetiContext instance"
- # Create a ConfigWriter...
+ # Create global configuration object
self.cfg = config.ConfigWriter()
- # And a GanetiLockingManager...
+
+ # Locking manager
self.glm = locking.GanetiLockManager(
self.cfg.GetNodeList(),
self.cfg.GetInstanceList())
+ # Job queue
+ self.jobqueue = jqueue.JobQueue(self)
+
# setting this also locks the class against attribute modifications
self.__class__._instance = self
assert self.__class__._instance is None, "Attempt to modify Ganeti Context"
object.__setattr__(self, name, value)
+ def AddNode(self, node):
+ """Adds a node to the configuration and lock manager.
+
+ """
+ # Add it to the configuration
+ self.cfg.AddNode(node)
-def CheckMaster(debug):
- """Checks the node setup.
+ # If preseeding fails it'll not be added
+ self.jobqueue.AddNode(node.name)
- If this is the master, the function will return. Otherwise it will
- exit with an exit code based on the node status.
+ # Add the new node to the Ganeti Lock Manager
+ self.glm.add(locking.LEVEL_NODE, node.name)
- """
- try:
- ss = ssconf.SimpleStore()
- master_name = ss.GetMasterNode()
- except errors.ConfigurationError, err:
- print "Cluster configuration incomplete: '%s'" % str(err)
- sys.exit(EXIT_NODESETUP_ERROR)
+ def ReaddNode(self, node):
+ """Updates a node that's already in the configuration
- try:
- myself = utils.HostInfo()
- except errors.ResolverError, err:
- sys.stderr.write("Cannot resolve my own name (%s)\n" % err.args[0])
- sys.exit(EXIT_NODESETUP_ERROR)
+ """
+ # Synchronize the queue again
+ self.jobqueue.AddNode(node.name)
+
+ def RemoveNode(self, name):
+ """Removes a node from the configuration and lock manager.
+
+ """
+ # Remove node from configuration
+ self.cfg.RemoveNode(name)
+
+ # Notify job queue
+ self.jobqueue.RemoveNode(name)
- if myself.name != master_name:
- if debug:
- sys.stderr.write("Not master, exiting.\n")
- sys.exit(EXIT_NOTMASTER)
+ # Remove the node from the Ganeti Lock Manager
+ self.glm.remove(locking.LEVEL_NODE, name)
def ParseOptions():
return options, args
+def CheckAgreement():
+ """Check the agreement on who is the master.
+
+ The function uses a very simple algorithm: we must get more positive
+ than negative answers. Since in most of the cases we are the master,
+ we'll use our own config file for getting the node list. In the
+ future we could collect the current node list from our (possibly
+ obsolete) known nodes.
+
+ """
+ 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)
+ 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")
+ return False
+ return True
+
+
def main():
"""Main function"""
utils.debug = options.debug
utils.no_fork = True
- CheckMaster(options.debug)
+ ssconf.CheckMaster(options.debug)
- master = IOServer(constants.MASTER_SOCKET, ClientRqHandler, GanetiContext())
+ # we believe we are the master, let's ask the other nodes...
+ if not CheckAgreement():
+ return
+
+ master = IOServer(constants.MASTER_SOCKET, ClientRqHandler)
# become a daemon
if options.fork:
utils.Daemonize(logfile=constants.LOG_MASTERDAEMON,
noclose_fds=[master.fileno()])
- logger.SetupDaemon(constants.LOG_MASTERDAEMON, debug=options.debug)
+ utils.WritePidFile(constants.MASTERD_PID)
- logger.Info("ganeti master daemon startup")
+ logger.SetupLogging(constants.LOG_MASTERDAEMON, debug=options.debug,
+ stderr_logging=not options.fork)
- try:
- utils.Lock('cmd', debug=options.debug)
- except errors.LockError, err:
- print >> sys.stderr, str(err)
- master.server_cleanup()
- return
+ logging.info("ganeti master daemon startup")
+ # activate ip
+ master_node = ssconf.SimpleStore().GetMasterNode()
+ if not rpc.call_node_start_master(master_node, False):
+ logging.error("Can't activate master IP address")
+
+ master.setup_queue()
try:
- master.setup_processors()
- master.setup_queue()
- try:
- master.serve_forever()
- finally:
- master.server_cleanup()
+ master.serve_forever()
finally:
- utils.Unlock('cmd')
- utils.LockCleanup()
+ master.server_cleanup()
+ utils.RemovePidFile(constants.MASTERD_PID)
if __name__ == "__main__":