Add ParseCpuMask() utility function

[ganeti-local] / lib / luxi.py

diff --git a/lib/luxi.py b/lib/luxi.py
index 541e059..9859e9b 100644 (file)
--- a/lib/luxi.py
+++ b/lib/luxi.py
@@ -21,11 +21,11 @@
 
 """Module for the unix socket protocol
 
 
 """Module for the unix socket protocol
 

-This module implements the local unix socket protocl. You only need
+This module implements the local unix socket protocol. You only need

 this module and the opcodes module in the client program in order to
 communicate with the master.
 
 this module and the opcodes module in the client program in order to
 communicate with the master.
 

-The module is also be used by the master daemon.
+The module is also used by the master daemon.

 
 """
 
 
 """
 

@@ -33,17 +33,21 @@ import socket
 import collections
 import time
 import errno
 import collections
 import time
 import errno

+import logging

 
 from ganeti import serializer
 from ganeti import constants
 
 from ganeti import serializer
 from ganeti import constants

+from ganeti import errors
+from ganeti import utils

 
 
 
 

-KEY_METHOD = 'method'
-KEY_ARGS = 'args'
+KEY_METHOD = "method"
+KEY_ARGS = "args"

 KEY_SUCCESS = "success"
 KEY_RESULT = "result"
 
 REQ_SUBMIT_JOB = "SubmitJob"
 KEY_SUCCESS = "success"
 KEY_RESULT = "result"
 
 REQ_SUBMIT_JOB = "SubmitJob"

+REQ_SUBMIT_MANY_JOBS = "SubmitManyJobs"

 REQ_WAIT_FOR_JOB_CHANGE = "WaitForJobChange"
 REQ_CANCEL_JOB = "CancelJob"
 REQ_ARCHIVE_JOB = "ArchiveJob"
 REQ_WAIT_FOR_JOB_CHANGE = "WaitForJobChange"
 REQ_CANCEL_JOB = "CancelJob"
 REQ_ARCHIVE_JOB = "ArchiveJob"

@@ -53,13 +57,20 @@ REQ_QUERY_INSTANCES = "QueryInstances"
 REQ_QUERY_NODES = "QueryNodes"
 REQ_QUERY_EXPORTS = "QueryExports"
 REQ_QUERY_CONFIG_VALUES = "QueryConfigValues"
 REQ_QUERY_NODES = "QueryNodes"
 REQ_QUERY_EXPORTS = "QueryExports"
 REQ_QUERY_CONFIG_VALUES = "QueryConfigValues"

+REQ_QUERY_CLUSTER_INFO = "QueryClusterInfo"
+REQ_QUERY_TAGS = "QueryTags"
+REQ_QUEUE_SET_DRAIN_FLAG = "SetDrainFlag"
+REQ_SET_WATCHER_PAUSE = "SetWatcherPause"

 
 DEF_CTMO = 10
 DEF_RWTO = 60
 
 
 DEF_CTMO = 10
 DEF_RWTO = 60
 

+# WaitForJobChange timeout
+WFJC_TIMEOUT = (DEF_RWTO - 1) / 2

 
 

-class ProtocolError(Exception):
-  """Denotes an error in the server communication"""
+
+class ProtocolError(errors.GenericError):
+  """Denotes an error in the LUXI protocol"""

 
 
 class ConnectionClosedError(ProtocolError):
 
 
 class ConnectionClosedError(ProtocolError):

@@ -70,14 +81,6 @@ class TimeoutError(ProtocolError):
   """Operation timeout error"""
 
 
   """Operation timeout error"""
 
 

-class EncodingError(ProtocolError):
-  """Encoding failure on the sending side"""
-
-
-class DecodingError(ProtocolError):
-  """Decoding failure on the receiving side"""
-
-

 class RequestError(ProtocolError):
   """Error on request
 
 class RequestError(ProtocolError):
   """Error on request
 

@@ -112,15 +115,12 @@ class Transport:
 
   """
 
 
   """
 

-  def __init__(self, address, timeouts=None, eom=None):
+  def __init__(self, address, timeouts=None):

     """Constructor for the Client class.
 
     Arguments:
       - address: a valid address the the used transport class
       - timeout: a list of timeouts, to be used on connect and read/write
     """Constructor for the Client class.
 
     Arguments:
       - address: a valid address the the used transport class
       - timeout: a list of timeouts, to be used on connect and read/write

-      - eom: an identifier to be used as end-of-message which the
-        upper-layer will guarantee that this identifier will not appear
-        in any message

 
     There are two timeouts used since we might want to wait for a long
     time for a response, but the connect timeout should be lower.
 
     There are two timeouts used since we might want to wait for a long
     time for a response, but the connect timeout should be lower.

@@ -143,22 +143,16 @@ class Transport:
     self._buffer = ""
     self._msgs = collections.deque()
 
     self._buffer = ""
     self._msgs = collections.deque()
 

-    if eom is None:
-      self.eom = '\3'
-    else:
-      self.eom = eom
-

     try:
       self.socket = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
     try:
       self.socket = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)

-      self.socket.settimeout(self._ctimeout)
+
+      # Try to connect

       try:
       try:

-        self.socket.connect(address)
-      except socket.timeout, err:
-        raise TimeoutError("Connect timed out: %s" % str(err))
-      except socket.error, err:
-        if err.args[0] in (errno.ENOENT, errno.ECONNREFUSED):
-          raise NoMasterError((address,))
-        raise
+        utils.Retry(self._Connect, 1.0, self._ctimeout,
+                    args=(self.socket, address, self._ctimeout))
+      except utils.RetryTimeout:
+        raise TimeoutError("Connect timed out")
+

       self.socket.settimeout(self._rwtimeout)
     except (socket.error, NoMasterError):
       if self.socket is not None:
       self.socket.settimeout(self._rwtimeout)
     except (socket.error, NoMasterError):
       if self.socket is not None:

@@ -166,6 +160,21 @@ class Transport:
       self.socket = None
       raise
 
       self.socket = None
       raise
 

+  @staticmethod
+  def _Connect(sock, address, timeout):
+    sock.settimeout(timeout)
+    try:
+      sock.connect(address)
+    except socket.timeout, err:
+      raise TimeoutError("Connect timed out: %s" % str(err))
+    except socket.error, err:
+      if err.args[0] in (errno.ENOENT, errno.ECONNREFUSED):
+        raise NoMasterError(address)
+      if err.args[0] == errno.EAGAIN:
+        # Server's socket backlog is full at the moment
+        raise utils.RetryAgain()
+      raise
+

   def _CheckSocket(self):
     """Make sure we are connected.
 
   def _CheckSocket(self):
     """Make sure we are connected.
 

@@ -179,16 +188,18 @@ class Transport:
     This just sends a message and doesn't wait for the response.
 
     """
     This just sends a message and doesn't wait for the response.
 
     """

-    if self.eom in msg:
-      raise EncodingError("Message terminator found in payload")
+    if constants.LUXI_EOM in msg:
+      raise ProtocolError("Message terminator found in payload")
+

     self._CheckSocket()
     try:
     self._CheckSocket()
     try:

-      self.socket.sendall(msg + self.eom)
+      # TODO: sendall is not guaranteed to send everything
+      self.socket.sendall(msg + constants.LUXI_EOM)

     except socket.timeout, err:
       raise TimeoutError("Sending timeout: %s" % str(err))
 
   def Recv(self):
     except socket.timeout, err:
       raise TimeoutError("Sending timeout: %s" % str(err))
 
   def Recv(self):

-    """Try to receive a messae from the socket.
+    """Try to receive a message from the socket.

 
     In case we already have messages queued, we just return from the
     queue. Otherwise, we try to read data with a _rwtimeout network
 
     In case we already have messages queued, we just return from the
     queue. Otherwise, we try to read data with a _rwtimeout network

@@ -201,13 +212,19 @@ class Transport:
     while not self._msgs:
       if time.time() > etime:
         raise TimeoutError("Extended receive timeout")
     while not self._msgs:
       if time.time() > etime:
         raise TimeoutError("Extended receive timeout")

-      try:
-        data = self.socket.recv(4096)
-      except socket.timeout, err:
-        raise TimeoutError("Receive timeout: %s" % str(err))
+      while True:
+        try:
+          data = self.socket.recv(4096)
+        except socket.error, err:
+          if err.args and err.args[0] == errno.EAGAIN:
+            continue
+          raise
+        except socket.timeout, err:
+          raise TimeoutError("Receive timeout: %s" % str(err))
+        break

       if not data:
         raise ConnectionClosedError("Connection closed while reading")
       if not data:
         raise ConnectionClosedError("Connection closed while reading")

-      new_msgs = (self._buffer + data).split(self.eom)
+      new_msgs = (self._buffer + data).split(constants.LUXI_EOM)

       self._buffer = new_msgs.pop()
       self._msgs.extend(new_msgs)
     return self._msgs.popleft()
       self._buffer = new_msgs.pop()
       self._msgs.extend(new_msgs)
     return self._msgs.popleft()

@@ -228,6 +245,99 @@ class Transport:
       self.socket = None
 
 
       self.socket = None
 
 

+def ParseRequest(msg):
+  """Parses a LUXI request message.
+
+  """
+  try:
+    request = serializer.LoadJson(msg)
+  except ValueError, err:
+    raise ProtocolError("Invalid LUXI request (parsing error): %s" % err)
+
+  logging.debug("LUXI request: %s", request)
+
+  if not isinstance(request, dict):
+    logging.error("LUXI request not a dict: %r", msg)
+    raise ProtocolError("Invalid LUXI request (not a dict)")
+
+  method = request.get(KEY_METHOD, None) # pylint: disable-msg=E1103
+  args = request.get(KEY_ARGS, None) # pylint: disable-msg=E1103
+
+  if method is None or args is None:
+    logging.error("LUXI request missing method or arguments: %r", msg)
+    raise ProtocolError(("Invalid LUXI request (no method or arguments"
+                         " in request): %r") % msg)
+
+  return (method, args)
+
+
+def ParseResponse(msg):
+  """Parses a LUXI response message.
+
+  """
+  # Parse the result
+  try:
+    data = serializer.LoadJson(msg)
+  except Exception, err:
+    raise ProtocolError("Error while deserializing response: %s" % str(err))
+
+  # Validate response
+  if not (isinstance(data, dict) and
+          KEY_SUCCESS in data and
+          KEY_RESULT in data):
+    raise ProtocolError("Invalid response from server: %r" % data)
+
+  return (data[KEY_SUCCESS], data[KEY_RESULT])
+
+
+def FormatResponse(success, result):
+  """Formats a LUXI response message.
+
+  """
+  response = {
+    KEY_SUCCESS: success,
+    KEY_RESULT: result,
+    }
+
+  logging.debug("LUXI response: %s", response)
+
+  return serializer.DumpJson(response)
+
+
+def FormatRequest(method, args):
+  """Formats a LUXI request message.
+
+  """
+  # Build request
+  request = {
+    KEY_METHOD: method,
+    KEY_ARGS: args,
+    }
+
+  # Serialize the request
+  return serializer.DumpJson(request, indent=False)
+
+
+def CallLuxiMethod(transport_cb, method, args):
+  """Send a LUXI request via a transport and return the response.
+
+  """
+  assert callable(transport_cb)
+
+  request_msg = FormatRequest(method, args)
+
+  # Send request and wait for response
+  response_msg = transport_cb(request_msg)
+
+  (success, result) = ParseResponse(response_msg)
+
+  if success:
+    return result
+
+  errors.MaybeRaise(result)
+  raise RequestError(result)
+
+

 class Client(object):
   """High-level client implementation.
 
 class Client(object):
   """High-level client implementation.
 

@@ -250,41 +360,64 @@ class Client(object):
     """
     if address is None:
       address = constants.MASTER_SOCKET
     """
     if address is None:
       address = constants.MASTER_SOCKET

-    self.transport = transport(address, timeouts=timeouts)
+    self.address = address
+    self.timeouts = timeouts
+    self.transport_class = transport
+    self.transport = None
+    self._InitTransport()

 
 

-  def CallMethod(self, method, args):
-    """Send a generic request and return the response.
+  def _InitTransport(self):
+    """(Re)initialize the transport if needed.

 
     """
 
     """

-    # Build request
-    request = {
-      KEY_METHOD: method,
-      KEY_ARGS: args,
-      }
+    if self.transport is None:
+      self.transport = self.transport_class(self.address,
+                                            timeouts=self.timeouts)
+
+  def _CloseTransport(self):
+    """Close the transport, ignoring errors.

 
 

+    """
+    if self.transport is None:
+      return
+    try:
+      old_transp = self.transport
+      self.transport = None
+      old_transp.Close()
+    except Exception: # pylint: disable-msg=W0703
+      pass
+
+  def _SendMethodCall(self, data):

     # Send request and wait for response
     # Send request and wait for response

-    result = self.transport.Call(serializer.DumpJson(request, indent=False))

     try:
     try:

-      data = serializer.LoadJson(result)
-    except Exception, err:
-      raise ProtocolError("Error while deserializing response: %s" % str(err))
+      self._InitTransport()
+      return self.transport.Call(data)
+    except Exception:
+      self._CloseTransport()
+      raise

 
 

-    # Validate response
-    if (not isinstance(data, dict) or
-        KEY_SUCCESS not in data or
-        KEY_RESULT not in data):
-      raise DecodingError("Invalid response from server: %s" % str(data))
+  def CallMethod(self, method, args):
+    """Send a generic request and return the response.

 
 

-    if not data[KEY_SUCCESS]:
-      # TODO: decide on a standard exception
-      raise RequestError(data[KEY_RESULT])
+    """
+    return CallLuxiMethod(self._SendMethodCall, method, args)
+
+  def SetQueueDrainFlag(self, drain_flag):
+    return self.CallMethod(REQ_QUEUE_SET_DRAIN_FLAG, drain_flag)

 
 

-    return data[KEY_RESULT]
+  def SetWatcherPause(self, until):
+    return self.CallMethod(REQ_SET_WATCHER_PAUSE, [until])

 
   def SubmitJob(self, ops):
     ops_state = map(lambda op: op.__getstate__(), ops)
     return self.CallMethod(REQ_SUBMIT_JOB, ops_state)
 
 
   def SubmitJob(self, ops):
     ops_state = map(lambda op: op.__getstate__(), ops)
     return self.CallMethod(REQ_SUBMIT_JOB, ops_state)
 

+  def SubmitManyJobs(self, jobs):
+    jobs_state = []
+    for ops in jobs:
+      jobs_state.append([op.__getstate__() for op in ops])
+    return self.CallMethod(REQ_SUBMIT_MANY_JOBS, jobs_state)
+

   def CancelJob(self, job_id):
     return self.CallMethod(REQ_CANCEL_JOB, job_id)
 
   def CancelJob(self, job_id):
     return self.CallMethod(REQ_CANCEL_JOB, job_id)
 

@@ -292,14 +425,36 @@ class Client(object):
     return self.CallMethod(REQ_ARCHIVE_JOB, job_id)
 
   def AutoArchiveJobs(self, age):
     return self.CallMethod(REQ_ARCHIVE_JOB, job_id)
 
   def AutoArchiveJobs(self, age):

-    return self.CallMethod(REQ_AUTOARCHIVE_JOBS, age)
+    timeout = (DEF_RWTO - 1) / 2
+    return self.CallMethod(REQ_AUTOARCHIVE_JOBS, (age, timeout))
+
+  def WaitForJobChangeOnce(self, job_id, fields,
+                           prev_job_info, prev_log_serial,
+                           timeout=WFJC_TIMEOUT):
+    """Waits for changes on a job.
+
+    @param job_id: Job ID
+    @type fields: list
+    @param fields: List of field names to be observed
+    @type prev_job_info: None or list
+    @param prev_job_info: Previously received job information
+    @type prev_log_serial: None or int/long
+    @param prev_log_serial: Highest log serial number previously received
+    @type timeout: int/float
+    @param timeout: Timeout in seconds (values larger than L{WFJC_TIMEOUT} will
+                    be capped to that value)
+
+    """
+    assert timeout >= 0, "Timeout can not be negative"
+    return self.CallMethod(REQ_WAIT_FOR_JOB_CHANGE,
+                           (job_id, fields, prev_job_info,
+                            prev_log_serial,
+                            min(WFJC_TIMEOUT, timeout)))

 
   def WaitForJobChange(self, job_id, fields, prev_job_info, prev_log_serial):
 
   def WaitForJobChange(self, job_id, fields, prev_job_info, prev_log_serial):

-    timeout = (DEF_RWTO - 1) / 2

     while True:
     while True:

-      result = self.CallMethod(REQ_WAIT_FOR_JOB_CHANGE,
-                               (job_id, fields, prev_job_info,
-                                prev_log_serial, timeout))
+      result = self.WaitForJobChangeOnce(job_id, fields,
+                                         prev_job_info, prev_log_serial)

       if result != constants.JOB_NOTCHANGED:
         break
     return result
       if result != constants.JOB_NOTCHANGED:
         break
     return result

@@ -307,16 +462,20 @@ class Client(object):
   def QueryJobs(self, job_ids, fields):
     return self.CallMethod(REQ_QUERY_JOBS, (job_ids, fields))
 
   def QueryJobs(self, job_ids, fields):
     return self.CallMethod(REQ_QUERY_JOBS, (job_ids, fields))
 

-  def QueryInstances(self, names, fields):
-    return self.CallMethod(REQ_QUERY_INSTANCES, (names, fields))
+  def QueryInstances(self, names, fields, use_locking):
+    return self.CallMethod(REQ_QUERY_INSTANCES, (names, fields, use_locking))
+
+  def QueryNodes(self, names, fields, use_locking):
+    return self.CallMethod(REQ_QUERY_NODES, (names, fields, use_locking))

 
 

-  def QueryNodes(self, names, fields):
-    return self.CallMethod(REQ_QUERY_NODES, (names, fields))
+  def QueryExports(self, nodes, use_locking):
+    return self.CallMethod(REQ_QUERY_EXPORTS, (nodes, use_locking))

 
 

-  def QueryExports(self, nodes):
-    return self.CallMethod(REQ_QUERY_EXPORTS, nodes)
+  def QueryClusterInfo(self):
+    return self.CallMethod(REQ_QUERY_CLUSTER_INFO, ())

 
   def QueryConfigValues(self, fields):
     return self.CallMethod(REQ_QUERY_CONFIG_VALUES, fields)
 
 
   def QueryConfigValues(self, fields):
     return self.CallMethod(REQ_QUERY_CONFIG_VALUES, fields)
 

-# TODO: class Server(object)
+  def QueryTags(self, kind, name):
+    return self.CallMethod(REQ_QUERY_TAGS, (kind, name))