Modify cli.JobExecutor to use SubmitManyJobs
[ganeti-local] / lib / bootstrap.py
index 6cd7424..7cd0d92 100644 (file)
@@ -25,8 +25,9 @@
 
 import os
 import os.path
-import sha
 import re
+import logging
+import tempfile
 
 from ganeti import rpc
 from ganeti import ssh
@@ -34,19 +35,17 @@ from ganeti import utils
 from ganeti import errors
 from ganeti import config
 from ganeti import constants
+from ganeti import objects
 from ganeti import ssconf
+from ganeti import hypervisor
 
 
-def _InitSSHSetup(node):
+def _InitSSHSetup():
   """Setup the SSH configuration for the cluster.
 
-
   This generates a dsa keypair for root, adds the pub key to the
   permitted hosts and adds the hostkey to its own known hosts.
 
-  Args:
-    node: the name of this host as a fqdn
-
   """
   priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS)
 
@@ -69,31 +68,52 @@ def _InitSSHSetup(node):
     f.close()
 
 
-def _InitGanetiServerSetup(ss):
+def _GenerateSelfSignedSslCert(file_name, validity=(365 * 5)):
+  """Generates a self-signed SSL certificate.
+
+  @type file_name: str
+  @param file_name: Path to output file
+  @type validity: int
+  @param validity: Validity for certificate in days
+
+  """
+  (fd, tmp_file_name) = tempfile.mkstemp(dir=os.path.dirname(file_name))
+  try:
+    try:
+      # Set permissions before writing key
+      os.chmod(tmp_file_name, 0600)
+
+      result = utils.RunCmd(["openssl", "req", "-new", "-newkey", "rsa:1024",
+                             "-days", str(validity), "-nodes", "-x509",
+                             "-keyout", tmp_file_name, "-out", tmp_file_name,
+                             "-batch"])
+      if result.failed:
+        raise errors.OpExecError("Could not generate SSL certificate, command"
+                                 " %s had exitcode %s and error message %s" %
+                                 (result.cmd, result.exit_code, result.output))
+
+      # Make read-only
+      os.chmod(tmp_file_name, 0400)
+
+      os.rename(tmp_file_name, file_name)
+    finally:
+      utils.RemoveFile(tmp_file_name)
+  finally:
+    os.close(fd)
+
+
+def _InitGanetiServerSetup():
   """Setup the necessary configuration for the initial node daemon.
 
   This creates the nodepass file containing the shared password for
   the cluster and also generates the SSL certificate.
 
-  Args:
-    ss: A WritableSimpleStore
-
   """
-  # Create pseudo random password
-  randpass = sha.new(os.urandom(64)).hexdigest()
-  # and write it into sstore
-  ss.SetKey(ss.SS_NODED_PASS, randpass)
-
-  result = utils.RunCmd(["openssl", "req", "-new", "-newkey", "rsa:1024",
-                         "-days", str(365*5), "-nodes", "-x509",
-                         "-keyout", constants.SSL_CERT_FILE,
-                         "-out", constants.SSL_CERT_FILE, "-batch"])
-  if result.failed:
-    raise errors.OpExecError("could not generate server ssl cert, command"
-                             " %s had exitcode %s and error message %s" %
-                             (result.cmd, result.exit_code, result.output))
+  _GenerateSelfSignedSslCert(constants.SSL_CERT_FILE)
 
-  os.chmod(constants.SSL_CERT_FILE, 0400)
+  # Don't overwrite existing file
+  if not os.path.exists(constants.RAPI_CERT_FILE):
+    _GenerateSelfSignedSslCert(constants.RAPI_CERT_FILE)
 
   result = utils.RunCmd([constants.NODE_INITD_SCRIPT, "restart"])
 
@@ -103,21 +123,27 @@ def _InitGanetiServerSetup(ss):
                              (result.cmd, result.exit_code, result.output))
 
 
-def InitCluster(cluster_name, hypervisor_type, mac_prefix, def_bridge,
-                master_netdev, file_storage_dir,
-                secondary_ip=None,
-                vg_name=None):
+def InitCluster(cluster_name, mac_prefix, def_bridge,
+                master_netdev, file_storage_dir, candidate_pool_size,
+                secondary_ip=None, vg_name=None, beparams=None, hvparams=None,
+                enabled_hypervisors=None, default_hypervisor=None):
   """Initialise the cluster.
 
+  @type candidate_pool_size: int
+  @param candidate_pool_size: master candidate pool size
+
   """
+  # TODO: complete the docstring
   if config.ConfigWriter.IsCluster():
     raise errors.OpPrereqError("Cluster is already initialised")
 
-  if hypervisor_type == constants.HT_XEN_HVM31:
-    if not os.path.exists(constants.VNC_PASSWORD_FILE):
-      raise errors.OpPrereqError("Please prepare the cluster VNC"
-                                 "password file %s" %
-                                 constants.VNC_PASSWORD_FILE)
+  if not enabled_hypervisors:
+    raise errors.OpPrereqError("Enabled hypervisors list must contain at"
+                               " least one member")
+  invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
+  if invalid_hvs:
+    raise errors.OpPrereqError("Enabled hypervisors contains invalid"
+                               " entries: %s" % invalid_hvs)
 
   hostname = utils.HostInfo()
 
@@ -126,8 +152,7 @@ def InitCluster(cluster_name, hypervisor_type, mac_prefix, def_bridge,
                                " range (%s). Please fix DNS or %s." %
                                (hostname.ip, constants.ETC_HOSTS))
 
-  if not utils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT,
-                       source=constants.LOCALHOST_IP_ADDRESS):
+  if not utils.OwnIpAddress(hostname.ip):
     raise errors.OpPrereqError("Inconsistency: this host's name resolves"
                                " to %s,\nbut this ip address does not"
                                " belong to this host."
@@ -143,11 +168,12 @@ def InitCluster(cluster_name, hypervisor_type, mac_prefix, def_bridge,
     if not utils.IsValidIP(secondary_ip):
       raise errors.OpPrereqError("Invalid secondary ip given")
     if (secondary_ip != hostname.ip and
-        (not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
-                           source=constants.LOCALHOST_IP_ADDRESS))):
+        not utils.OwnIpAddress(secondary_ip)):
       raise errors.OpPrereqError("You gave %s as secondary IP,"
                                  " but it does not belong to this host." %
                                  secondary_ip)
+  else:
+    secondary_ip = hostname.ip
 
   if vg_name is not None:
     # Check if volume group is valid
@@ -178,10 +204,6 @@ def InitCluster(cluster_name, hypervisor_type, mac_prefix, def_bridge,
   if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix):
     raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix)
 
-  if hypervisor_type not in constants.HYPER_TYPES:
-    raise errors.OpPrereqError("Invalid hypervisor type given '%s'" %
-                               hypervisor_type)
-
   result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev])
   if result.failed:
     raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" %
@@ -193,22 +215,18 @@ def InitCluster(cluster_name, hypervisor_type, mac_prefix, def_bridge,
     raise errors.OpPrereqError("Init.d script '%s' missing or not"
                                " executable." % constants.NODE_INITD_SCRIPT)
 
-  # set up the simple store
-  ss = ssconf.WritableSimpleStore()
-  ss.SetKey(ss.SS_HYPERVISOR, hypervisor_type)
-  ss.SetKey(ss.SS_MASTER_NODE, hostname.name)
-  ss.SetKey(ss.SS_MASTER_IP, clustername.ip)
-  ss.SetKey(ss.SS_MASTER_NETDEV, master_netdev)
-  ss.SetKey(ss.SS_CLUSTER_NAME, clustername.name)
-  ss.SetKey(ss.SS_FILE_STORAGE_DIR, file_storage_dir)
-  ss.SetKey(ss.SS_CONFIG_VERSION, constants.CONFIG_VERSION)
+  dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)]
+  utils.EnsureDirs(dirs)
 
-  # set up the inter-node password and certificate
-  _InitGanetiServerSetup(ss)
+  utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES)
+  # hvparams is a mapping of hypervisor->hvparams dict
+  for hv_name, hv_params in hvparams.iteritems():
+    utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
+    hv_class = hypervisor.GetHypervisor(hv_name)
+    hv_class.CheckParameterSyntax(hv_params)
 
-  # start the master ip
-  # TODO: Review rpc call from bootstrap
-  rpc.call_node_start_master(hostname.name)
+  # set up the inter-node password and certificate
+  _InitGanetiServerSetup()
 
   # set up ssh config and /etc/hosts
   f = open(constants.SSH_HOST_RSA_PUB, 'r')
@@ -219,11 +237,301 @@ def InitCluster(cluster_name, hypervisor_type, mac_prefix, def_bridge,
   sshkey = sshline.split(" ")[1]
 
   utils.AddHostToEtcHosts(hostname.name)
-  _InitSSHSetup(hostname.name)
+  _InitSSHSetup()
 
   # init of cluster config file
+  cluster_config = objects.Cluster(
+    serial_no=1,
+    rsahostkeypub=sshkey,
+    highest_used_port=(constants.FIRST_DRBD_PORT - 1),
+    mac_prefix=mac_prefix,
+    volume_group_name=vg_name,
+    default_bridge=def_bridge,
+    tcpudp_port_pool=set(),
+    master_node=hostname.name,
+    master_ip=clustername.ip,
+    master_netdev=master_netdev,
+    cluster_name=clustername.name,
+    file_storage_dir=file_storage_dir,
+    enabled_hypervisors=enabled_hypervisors,
+    default_hypervisor=default_hypervisor,
+    beparams={constants.BEGR_DEFAULT: beparams},
+    hvparams=hvparams,
+    candidate_pool_size=candidate_pool_size,
+    )
+  master_node_config = objects.Node(name=hostname.name,
+                                    primary_ip=hostname.ip,
+                                    secondary_ip=secondary_ip,
+                                    serial_no=1,
+                                    master_candidate=True,
+                                    offline=False, drained=False,
+                                    )
+
+  sscfg = InitConfig(constants.CONFIG_VERSION,
+                     cluster_config, master_node_config)
+  ssh.WriteKnownHostsFile(sscfg, constants.SSH_KNOWN_HOSTS_FILE)
+  cfg = config.ConfigWriter()
+  cfg.Update(cfg.GetClusterInfo())
+
+  # start the master ip
+  # TODO: Review rpc call from bootstrap
+  rpc.RpcRunner.call_node_start_master(hostname.name, True)
+
+
+def InitConfig(version, cluster_config, master_node_config,
+               cfg_file=constants.CLUSTER_CONF_FILE):
+  """Create the initial cluster configuration.
+
+  It will contain the current node, which will also be the master
+  node, and no instances.
+
+  @type version: int
+  @param version: configuration version
+  @type cluster_config: L{objects.Cluster}
+  @param cluster_config: cluster configuration
+  @type master_node_config: L{objects.Node}
+  @param master_node_config: master node configuration
+  @type cfg_file: string
+  @param cfg_file: configuration file path
+
+  @rtype: L{ssconf.SimpleConfigWriter}
+  @return: initialized config instance
+
+  """
+  nodes = {
+    master_node_config.name: master_node_config,
+    }
+
+  config_data = objects.ConfigData(version=version,
+                                   cluster=cluster_config,
+                                   nodes=nodes,
+                                   instances={},
+                                   serial_no=1)
+  cfg = ssconf.SimpleConfigWriter.FromDict(config_data.ToDict(), cfg_file)
+  cfg.Save()
+
+  return cfg
+
+
+def FinalizeClusterDestroy(master):
+  """Execute the last steps of cluster destroy
+
+  This function shuts down all the daemons, completing the destroy
+  begun in cmdlib.LUDestroyOpcode.
+
+  """
+  result = rpc.RpcRunner.call_node_stop_master(master, True)
+  if result.failed or not result.data:
+    logging.warning("Could not disable the master role")
+  result = rpc.RpcRunner.call_node_leave_cluster(master)
+  if result.failed or not result.data:
+    logging.warning("Could not shutdown the node daemon and cleanup the node")
+
+
+def SetupNodeDaemon(cluster_name, node, ssh_key_check):
+  """Add a node to the cluster.
+
+  This function must be called before the actual opcode, and will ssh
+  to the remote node, copy the needed files, and start ganeti-noded,
+  allowing the master to do the rest via normal rpc calls.
+
+  @param cluster_name: the cluster name
+  @param node: the name of the new node
+  @param ssh_key_check: whether to do a strict key check
+
+  """
+  sshrunner = ssh.SshRunner(cluster_name)
+
+  noded_cert = utils.ReadFile(constants.SSL_CERT_FILE)
+  rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE)
+
+  # in the base64 pem encoding, neither '!' nor '.' are valid chars,
+  # so we use this to detect an invalid certificate; as long as the
+  # cert doesn't contain this, the here-document will be correctly
+  # parsed by the shell sequence below
+  if (re.search('^!EOF\.', noded_cert, re.MULTILINE) or
+      re.search('^!EOF\.', rapi_cert, re.MULTILINE)):
+    raise errors.OpExecError("invalid PEM encoding in the SSL certificate")
+
+  if not noded_cert.endswith("\n"):
+    noded_cert += "\n"
+  if not rapi_cert.endswith("\n"):
+    rapi_cert += "\n"
+
+  # set up inter-node password and certificate and restarts the node daemon
+  # and then connect with ssh to set password and start ganeti-noded
+  # note that all the below variables are sanitized at this point,
+  # either by being constants or by the checks above
+  mycommand = ("umask 077 && "
+               "cat > '%s' << '!EOF.' && \n"
+               "%s!EOF.\n"
+               "cat > '%s' << '!EOF.' && \n"
+               "%s!EOF.\n"
+               "chmod 0400 %s %s && "
+               "%s restart" %
+               (constants.SSL_CERT_FILE, noded_cert,
+                constants.RAPI_CERT_FILE, rapi_cert,
+                constants.SSL_CERT_FILE, constants.RAPI_CERT_FILE,
+                constants.NODE_INITD_SCRIPT))
+
+  result = sshrunner.Run(node, 'root', mycommand, batch=False,
+                         ask_key=ssh_key_check,
+                         use_cluster_key=False,
+                         strict_host_check=ssh_key_check)
+  if result.failed:
+    raise errors.OpExecError("Remote command on node %s, error: %s,"
+                             " output: %s" %
+                             (node, result.fail_reason, result.output))
+
+
+def MasterFailover(no_voting=False):
+  """Failover the master node.
+
+  This checks that we are not already the master, and will cause the
+  current master to cease being master, and the non-master to become
+  new master.
+
+  @type no_voting: boolean
+  @param no_voting: force the operation without remote nodes agreement
+                      (dangerous)
+
+  """
+  sstore = ssconf.SimpleStore()
+
+  old_master, new_master = ssconf.GetMasterAndMyself(sstore)
+  node_list = sstore.GetNodeList()
+  mc_list = sstore.GetMasterCandidates()
+
+  if old_master == new_master:
+    raise errors.OpPrereqError("This commands must be run on the node"
+                               " where you want the new master to be."
+                               " %s is already the master" %
+                               old_master)
+
+  if new_master not in mc_list:
+    mc_no_master = [name for name in mc_list if name != old_master]
+    raise errors.OpPrereqError("This node is not among the nodes marked"
+                               " as master candidates. Only these nodes"
+                               " can become masters. Current list of"
+                               " master candidates is:\n"
+                               "%s" % ('\n'.join(mc_no_master)))
+
+  if not no_voting:
+    vote_list = GatherMasterVotes(node_list)
+
+    if vote_list:
+      voted_master = vote_list[0][0]
+      if voted_master is None:
+        raise errors.OpPrereqError("Cluster is inconsistent, most nodes did"
+                                   " not respond.")
+      elif voted_master != old_master:
+        raise errors.OpPrereqError("I have a wrong configuration, I believe"
+                                   " the master is %s but the other nodes"
+                                   " voted %s. Please resync the configuration"
+                                   " of this node." %
+                                   (old_master, voted_master))
+  # end checks
+
+  rcode = 0
+
+  logging.info("Setting master to %s, old master: %s", new_master, old_master)
+
+  result = rpc.RpcRunner.call_node_stop_master(old_master, True)
+  if result.failed or not result.data:
+    logging.error("Could not disable the master role on the old master"
+                 " %s, please disable manually", old_master)
+
+  # Here we have a phase where no master should be running
+
+  # instantiate a real config writer, as we now know we have the
+  # configuration data
   cfg = config.ConfigWriter()
-  cfg.InitConfig(hostname.name, hostname.ip, secondary_ip, sshkey,
-                 mac_prefix, vg_name, def_bridge)
 
-  ssh.WriteKnownHostsFile(cfg, ss, constants.SSH_KNOWN_HOSTS_FILE)
+  cluster_info = cfg.GetClusterInfo()
+  cluster_info.master_node = new_master
+  # this will also regenerate the ssconf files, since we updated the
+  # cluster info
+  cfg.Update(cluster_info)
+
+  # 2.0.X: Don't start the master if no_voting is true
+  result = rpc.RpcRunner.call_node_start_master(new_master, not no_voting)
+  if result.failed or not result.data:
+    logging.error("Could not start the master role on the new master"
+                  " %s, please check", new_master)
+    rcode = 1
+
+  return rcode
+
+
+def GetMaster():
+  """Returns the current master node.
+
+  This is a separate function in bootstrap since it's needed by
+  gnt-cluster, and instead of importing directly ssconf, it's better
+  to abstract it in bootstrap, where we do use ssconf in other
+  functions too.
+
+  """
+  sstore = ssconf.SimpleStore()
+
+  old_master, _ = ssconf.GetMasterAndMyself(sstore)
+
+  return old_master
+
+
+def GatherMasterVotes(node_list):
+  """Check the agreement on who is the master.
+
+  This function will return a list of (node, number of votes), ordered
+  by the number of votes. Errors will be denoted by the key 'None'.
+
+  Note that the sum of votes is the number of nodes this machine
+  knows, whereas the number of entries in the list could be different
+  (if some nodes vote for another master).
+
+  We remove ourselves from the list since we know that (bugs aside)
+  since we use the same source for configuration information for both
+  backend and boostrap, we'll always vote for ourselves.
+
+  @type node_list: list
+  @param node_list: the list of nodes to query for master info; the current
+      node will be removed if it is in the list
+  @rtype: list
+  @return: list of (node, votes)
+
+  """
+  myself = utils.HostInfo().name
+  try:
+    node_list.remove(myself)
+  except ValueError:
+    pass
+  if not node_list:
+    # no nodes left (eventually after removing myself)
+    return []
+  results = rpc.RpcRunner.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 [(None, len(node_list))]
+  votes = {}
+  for node in results:
+    nres = results[node]
+    data = nres.data
+    if nres.failed or not isinstance(data, (tuple, list)) or len(data) < 3:
+      # here the rpc layer should have already logged errors
+      if None not in votes:
+        votes[None] = 0
+      votes[None] += 1
+      continue
+    master_node = data[2]
+    if master_node not in votes:
+      votes[master_node] = 0
+    votes[master_node] += 1
+
+  vote_list = [v for v in votes.items()]
+  # sort first on number of votes then on name, since we want None
+  # sorted later if we have the half of the nodes not responding, and
+  # half voting all for the same master
+  vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True)
+
+  return vote_list