Make _GenerateDRBD8Branch accept different VG names
[ganeti-local] / lib / mcpu.py
index f9d5ab7..4ee6de5 100644 (file)
@@ -1,7 +1,7 @@
 #
 #
 
-# Copyright (C) 2006, 2007 Google Inc.
+# Copyright (C) 2006, 2007, 2011 Google Inc.
 #
 # This program is free software; you can redistribute it and/or modify
 # it under the terms of the GNU General Public License as published by
@@ -29,6 +29,8 @@ are two kinds of classes defined:
 """
 
 import logging
+import random
+import time
 
 from ganeti import opcodes
 from ganeti import constants
@@ -39,7 +41,87 @@ from ganeti import locking
 from ganeti import utils
 
 
-class OpExecCbBase:
+_OP_PREFIX = "Op"
+_LU_PREFIX = "LU"
+
+
+class LockAcquireTimeout(Exception):
+  """Exception to report timeouts on acquiring locks.
+
+  """
+
+
+def _CalculateLockAttemptTimeouts():
+  """Calculate timeouts for lock attempts.
+
+  """
+  result = [constants.LOCK_ATTEMPTS_MINWAIT]
+  running_sum = result[0]
+
+  # Wait for a total of at least LOCK_ATTEMPTS_TIMEOUT before doing a
+  # blocking acquire
+  while running_sum < constants.LOCK_ATTEMPTS_TIMEOUT:
+    timeout = (result[-1] * 1.05) ** 1.25
+
+    # Cap max timeout. This gives other jobs a chance to run even if
+    # we're still trying to get our locks, before finally moving to a
+    # blocking acquire.
+    timeout = min(timeout, constants.LOCK_ATTEMPTS_MAXWAIT)
+    # And also cap the lower boundary for safety
+    timeout = max(timeout, constants.LOCK_ATTEMPTS_MINWAIT)
+
+    result.append(timeout)
+    running_sum += timeout
+
+  return result
+
+
+class LockAttemptTimeoutStrategy(object):
+  """Class with lock acquire timeout strategy.
+
+  """
+  __slots__ = [
+    "_timeouts",
+    "_random_fn",
+    "_time_fn",
+    ]
+
+  _TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts()
+
+  def __init__(self, _time_fn=time.time, _random_fn=random.random):
+    """Initializes this class.
+
+    @param _time_fn: Time function for unittests
+    @param _random_fn: Random number generator for unittests
+
+    """
+    object.__init__(self)
+
+    self._timeouts = iter(self._TIMEOUT_PER_ATTEMPT)
+    self._time_fn = _time_fn
+    self._random_fn = _random_fn
+
+  def NextAttempt(self):
+    """Returns the timeout for the next attempt.
+
+    """
+    try:
+      timeout = self._timeouts.next()
+    except StopIteration:
+      # No more timeouts, do blocking acquire
+      timeout = None
+
+    if timeout is not None:
+      # Add a small variation (-/+ 5%) to timeout. This helps in situations
+      # where two or more jobs are fighting for the same lock(s).
+      variation_range = timeout * 0.1
+      timeout += ((self._random_fn() * variation_range) -
+                  (variation_range * 0.5))
+
+    return timeout
+
+
+class OpExecCbBase: # pylint: disable-msg=W0232
   """Base class for OpCode execution callbacks.
 
   """
@@ -56,125 +138,75 @@ class OpExecCbBase:
 
     """
 
-  def ReportLocks(self, msg):
-    """Report lock operations.
+  def CheckCancel(self):
+    """Check whether job has been cancelled.
 
     """
 
 
+def _LUNameForOpName(opname):
+  """Computes the LU name for a given OpCode name.
+
+  """
+  assert opname.startswith(_OP_PREFIX), \
+      "Invalid OpCode name, doesn't start with %s: %s" % (_OP_PREFIX, opname)
+
+  return _LU_PREFIX + opname[len(_OP_PREFIX):]
+
+
+def _ComputeDispatchTable():
+  """Computes the opcode-to-lu dispatch table.
+
+  """
+  return dict((op, getattr(cmdlib, _LUNameForOpName(op.__name__)))
+              for op in opcodes.OP_MAPPING.values()
+              if op.WITH_LU)
+
+
 class Processor(object):
   """Object which runs OpCodes"""
-  DISPATCH_TABLE = {
-    # Cluster
-    opcodes.OpPostInitCluster: cmdlib.LUPostInitCluster,
-    opcodes.OpDestroyCluster: cmdlib.LUDestroyCluster,
-    opcodes.OpQueryClusterInfo: cmdlib.LUQueryClusterInfo,
-    opcodes.OpVerifyCluster: cmdlib.LUVerifyCluster,
-    opcodes.OpQueryConfigValues: cmdlib.LUQueryConfigValues,
-    opcodes.OpRenameCluster: cmdlib.LURenameCluster,
-    opcodes.OpVerifyDisks: cmdlib.LUVerifyDisks,
-    opcodes.OpSetClusterParams: cmdlib.LUSetClusterParams,
-    opcodes.OpRedistributeConfig: cmdlib.LURedistributeConfig,
-    opcodes.OpRepairDiskSizes: cmdlib.LURepairDiskSizes,
-    # node lu
-    opcodes.OpAddNode: cmdlib.LUAddNode,
-    opcodes.OpQueryNodes: cmdlib.LUQueryNodes,
-    opcodes.OpQueryNodeVolumes: cmdlib.LUQueryNodeVolumes,
-    opcodes.OpQueryNodeStorage: cmdlib.LUQueryNodeStorage,
-    opcodes.OpModifyNodeStorage: cmdlib.LUModifyNodeStorage,
-    opcodes.OpRepairNodeStorage: cmdlib.LURepairNodeStorage,
-    opcodes.OpRemoveNode: cmdlib.LURemoveNode,
-    opcodes.OpSetNodeParams: cmdlib.LUSetNodeParams,
-    opcodes.OpPowercycleNode: cmdlib.LUPowercycleNode,
-    opcodes.OpEvacuateNode: cmdlib.LUEvacuateNode,
-    opcodes.OpMigrateNode: cmdlib.LUMigrateNode,
-    # instance lu
-    opcodes.OpCreateInstance: cmdlib.LUCreateInstance,
-    opcodes.OpReinstallInstance: cmdlib.LUReinstallInstance,
-    opcodes.OpRemoveInstance: cmdlib.LURemoveInstance,
-    opcodes.OpRenameInstance: cmdlib.LURenameInstance,
-    opcodes.OpActivateInstanceDisks: cmdlib.LUActivateInstanceDisks,
-    opcodes.OpShutdownInstance: cmdlib.LUShutdownInstance,
-    opcodes.OpStartupInstance: cmdlib.LUStartupInstance,
-    opcodes.OpRebootInstance: cmdlib.LURebootInstance,
-    opcodes.OpDeactivateInstanceDisks: cmdlib.LUDeactivateInstanceDisks,
-    opcodes.OpReplaceDisks: cmdlib.LUReplaceDisks,
-    opcodes.OpRecreateInstanceDisks: cmdlib.LURecreateInstanceDisks,
-    opcodes.OpFailoverInstance: cmdlib.LUFailoverInstance,
-    opcodes.OpMigrateInstance: cmdlib.LUMigrateInstance,
-    opcodes.OpMoveInstance: cmdlib.LUMoveInstance,
-    opcodes.OpConnectConsole: cmdlib.LUConnectConsole,
-    opcodes.OpQueryInstances: cmdlib.LUQueryInstances,
-    opcodes.OpQueryInstanceData: cmdlib.LUQueryInstanceData,
-    opcodes.OpSetInstanceParams: cmdlib.LUSetInstanceParams,
-    opcodes.OpGrowDisk: cmdlib.LUGrowDisk,
-    # os lu
-    opcodes.OpDiagnoseOS: cmdlib.LUDiagnoseOS,
-    # exports lu
-    opcodes.OpQueryExports: cmdlib.LUQueryExports,
-    opcodes.OpExportInstance: cmdlib.LUExportInstance,
-    opcodes.OpRemoveExport: cmdlib.LURemoveExport,
-    # tags lu
-    opcodes.OpGetTags: cmdlib.LUGetTags,
-    opcodes.OpSearchTags: cmdlib.LUSearchTags,
-    opcodes.OpAddTags: cmdlib.LUAddTags,
-    opcodes.OpDelTags: cmdlib.LUDelTags,
-    # test lu
-    opcodes.OpTestDelay: cmdlib.LUTestDelay,
-    opcodes.OpTestAllocator: cmdlib.LUTestAllocator,
-    }
-
-  def __init__(self, context):
+  DISPATCH_TABLE = _ComputeDispatchTable()
+
+  def __init__(self, context, ec_id):
     """Constructor for Processor
 
+    @type context: GanetiContext
+    @param context: global Ganeti context
+    @type ec_id: string
+    @param ec_id: execution context identifier
+
     """
     self.context = context
+    self._ec_id = ec_id
     self._cbs = None
-    self.exclusive_BGL = False
     self.rpc = rpc.RpcRunner(context.cfg)
     self.hmclass = HooksMaster
 
-  def _ReportLocks(self, level, names, shared, acquired):
-    """Reports lock operations.
+  def _AcquireLocks(self, level, names, shared, timeout, priority):
+    """Acquires locks via the Ganeti lock manager.
 
     @type level: int
     @param level: Lock level
     @type names: list or string
     @param names: Lock names
     @type shared: bool
-    @param shared: Whether the lock should be acquired in shared mode
-    @type acquired: bool
-    @param acquired: Whether the lock has already been acquired
+    @param shared: Whether the locks should be acquired in shared mode
+    @type timeout: None or float
+    @param timeout: Timeout for acquiring the locks
+    @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
+        amount of time
 
     """
-    parts = []
-
-    # Build message
-    if acquired:
-      parts.append("acquired")
-    else:
-      parts.append("waiting")
-
-    parts.append(locking.LEVEL_NAMES[level])
+    if self._cbs:
+      self._cbs.CheckCancel()
 
-    if names == locking.ALL_SET:
-      parts.append("ALL")
-    elif isinstance(names, basestring):
-      parts.append(names)
-    else:
-      parts.append(",".join(names))
+    acquired = self.context.glm.acquire(level, names, shared=shared,
+                                        timeout=timeout, priority=priority)
 
-    if shared:
-      parts.append("shared")
-    else:
-      parts.append("exclusive")
+    if acquired is None:
+      raise LockAcquireTimeout()
 
-    msg = "/".join(parts)
-
-    logging.debug("LU locks %s", msg)
-
-    if self._cbs:
-      self._cbs.ReportLocks(msg)
+    return acquired
 
   def _ExecLU(self, lu):
     """Logical Unit execution sequence.
@@ -185,7 +217,7 @@ class Processor(object):
     hm = HooksMaster(self.rpc.call_hooks_runner, lu)
     h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE)
     lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results,
-                     self._Feedback, None)
+                     self.Log, None)
 
     if getattr(lu.op, "dry_run", False):
       # in this mode, no post-hooks are run, and the config is not
@@ -196,10 +228,10 @@ class Processor(object):
       return lu.dry_run_result
 
     try:
-      result = lu.Exec(self._Feedback)
+      result = lu.Exec(self.Log)
       h_results = hm.RunPhase(constants.HOOKS_PHASE_POST)
       result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results,
-                                self._Feedback, result)
+                                self.Log, result)
     finally:
       # FIXME: This needs locks if not lu_class.REQ_BGL
       if write_count != self.context.cfg.write_count:
@@ -207,7 +239,7 @@ class Processor(object):
 
     return result
 
-  def _LockAndExecLU(self, lu, level):
+  def _LockAndExecLU(self, lu, level, calc_timeout, priority):
     """Execute a Logical Unit, with the needed locks.
 
     This is a recursive function that starts locking the given level, and
@@ -222,93 +254,111 @@ class Processor(object):
         self._cbs.NotifyStart()
 
       result = self._ExecLU(lu)
+
     elif adding_locks and acquiring_locks:
       # We could both acquire and add locks at the same level, but for now we
       # don't need this, so we'll avoid the complicated code needed.
-      raise NotImplementedError(
-        "Can't declare locks to acquire when adding others")
+      raise NotImplementedError("Can't declare locks to acquire when adding"
+                                " others")
+
     elif adding_locks or acquiring_locks:
       lu.DeclareLocks(level)
       share = lu.share_locks[level]
-      if acquiring_locks:
-        needed_locks = lu.needed_locks[level]
-
-        self._ReportLocks(level, needed_locks, share, False)
-        lu.acquired_locks[level] = self.context.glm.acquire(level,
-                                                            needed_locks,
-                                                            shared=share)
-        self._ReportLocks(level, needed_locks, share, True)
-
-      else: # adding_locks
-        add_locks = lu.add_locks[level]
-        lu.remove_locks[level] = add_locks
-        try:
-          self.context.glm.add(level, add_locks, acquired=1, shared=share)
-        except errors.LockError:
-          raise errors.OpPrereqError(
-            "Couldn't add locks (%s), probably because of a race condition"
-            " with another job, who added them first" % add_locks)
+
       try:
+        assert adding_locks ^ acquiring_locks, \
+          "Locks must be either added or acquired"
+
+        if acquiring_locks:
+          # Acquiring locks
+          needed_locks = lu.needed_locks[level]
+
+          acquired = self._AcquireLocks(level, needed_locks, share,
+                                        calc_timeout(), priority)
+        else:
+          # Adding locks
+          add_locks = lu.add_locks[level]
+          lu.remove_locks[level] = add_locks
+
+          try:
+            self.context.glm.add(level, add_locks, acquired=1, shared=share)
+          except errors.LockError:
+            raise errors.OpPrereqError(
+              "Couldn't add locks (%s), probably because of a race condition"
+              " with another job, who added them first" % add_locks,
+              errors.ECODE_FAULT)
+
+          acquired = add_locks
+
         try:
-          if adding_locks:
-            lu.acquired_locks[level] = add_locks
-          result = self._LockAndExecLU(lu, level + 1)
+          lu.acquired_locks[level] = acquired
+
+          result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority)
         finally:
           if level in lu.remove_locks:
             self.context.glm.remove(level, lu.remove_locks[level])
       finally:
         if self.context.glm.is_owned(level):
           self.context.glm.release(level)
+
     else:
-      result = self._LockAndExecLU(lu, level + 1)
+      result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority)
 
     return result
 
-  def ExecOpCode(self, op, cbs):
+  def ExecOpCode(self, op, cbs, timeout=None, priority=None):
     """Execute an opcode.
 
     @type op: an OpCode instance
     @param op: the opcode to be executed
     @type cbs: L{OpExecCbBase}
     @param cbs: Runtime callbacks
+    @type timeout: float or None
+    @param timeout: Maximum time to acquire all locks, None for no timeout
+    @type priority: number or None
+    @param priority: Priority for acquiring lock(s)
+    @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
+        amount of time
 
     """
     if not isinstance(op, opcodes.OpCode):
       raise errors.ProgrammerError("Non-opcode instance passed"
                                    " to ExecOpcode")
 
+    lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
+    if lu_class is None:
+      raise errors.OpCodeUnknown("Unknown opcode")
+
+    if timeout is None:
+      calc_timeout = lambda: None
+    else:
+      calc_timeout = utils.RunningTimeout(timeout, False).Remaining
+
     self._cbs = cbs
     try:
-      lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
-      if lu_class is None:
-        raise errors.OpCodeUnknown("Unknown opcode")
-
-      # Acquire the Big Ganeti Lock exclusively if this LU requires it, and in a
-      # shared fashion otherwise (to prevent concurrent run with an exclusive
-      # LU.
-      self._ReportLocks(locking.LEVEL_CLUSTER, [locking.BGL],
-                        not lu_class.REQ_BGL, False)
-      try:
-        self.context.glm.acquire(locking.LEVEL_CLUSTER, [locking.BGL],
-                                 shared=not lu_class.REQ_BGL)
-      finally:
-        self._ReportLocks(locking.LEVEL_CLUSTER, [locking.BGL],
-                          not lu_class.REQ_BGL, True)
+      # Acquire the Big Ganeti Lock exclusively if this LU requires it,
+      # and in a shared fashion otherwise (to prevent concurrent run with
+      # an exclusive LU.
+      self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
+                          not lu_class.REQ_BGL, calc_timeout(),
+                          priority)
       try:
-        self.exclusive_BGL = lu_class.REQ_BGL
         lu = lu_class(self, op, self.context, self.rpc)
         lu.ExpandNames()
         assert lu.needed_locks is not None, "needed_locks not set by LU"
-        result = self._LockAndExecLU(lu, locking.LEVEL_INSTANCE)
+
+        try:
+          return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout,
+                                     priority)
+        finally:
+          if self._ec_id:
+            self.context.cfg.DropECReservations(self._ec_id)
       finally:
         self.context.glm.release(locking.LEVEL_CLUSTER)
-        self.exclusive_BGL = False
     finally:
       self._cbs = None
 
-    return result
-
-  def _Feedback(self, *args):
+  def Log(self, *args):
     """Forward call to feedback callback function.
 
     """
@@ -320,7 +370,7 @@ class Processor(object):
 
     """
     logging.debug("Step %d/%d %s", current, total, message)
-    self._Feedback("STEP %d/%d %s" % (current, total, message))
+    self.Log("STEP %d/%d %s" % (current, total, message))
 
   def LogWarning(self, message, *args, **kwargs):
     """Log a warning to the logs and the user.
@@ -337,9 +387,9 @@ class Processor(object):
       message = message % tuple(args)
     if message:
       logging.warning(message)
-      self._Feedback(" - WARNING: %s" % message)
+      self.Log(" - WARNING: %s" % message)
     if "hint" in kwargs:
-      self._Feedback("      Hint: %s" % kwargs["hint"])
+      self.Log("      Hint: %s" % kwargs["hint"])
 
   def LogInfo(self, message, *args):
     """Log an informational message to the logs and the user.
@@ -348,7 +398,16 @@ class Processor(object):
     if args:
       message = message % tuple(args)
     logging.info(message)
-    self._Feedback(" - INFO: %s" % message)
+    self.Log(" - INFO: %s" % message)
+
+  def GetECId(self):
+    """Returns the current execution context ID.
+
+    """
+    if not self._ec_id:
+      raise errors.ProgrammerError("Tried to use execution context id when"
+                                   " not set")
+    return self._ec_id
 
 
 class HooksMaster(object):