#
#
-# 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
from ganeti import rpc
from ganeti import cmdlib
from ganeti import locking
+from ganeti import utils
+from ganeti import compat
-class _LockAcquireTimeout(Exception):
- """Internal exception to report timeouts on acquiring locks.
+_OP_PREFIX = "Op"
+_LU_PREFIX = "LU"
+
+
+class LockAcquireTimeout(Exception):
+ """Exception to report timeouts on acquiring locks.
"""
"""Calculate timeouts for lock attempts.
"""
- result = [1.0]
+ result = [constants.LOCK_ATTEMPTS_MINWAIT]
+ running_sum = result[0]
- # Wait for a total of at least 150s before doing a blocking acquire
- while sum(result) < 150.0:
+ # 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 timeout at 10 seconds. 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.
- if timeout > 10.0:
- timeout = 10.0
-
- elif timeout < 0.1:
- # Lower boundary for safety
- timeout = 0.1
+ # 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 LockAttemptTimeoutStrategy(object):
"""Class with lock acquire timeout strategy.
"""
__slots__ = [
- "_attempt",
+ "_timeouts",
"_random_fn",
- "_start_time",
"_time_fn",
- "_running_timeout",
]
_TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts()
- def __init__(self, attempt=0, _time_fn=time.time, _random_fn=random.random):
+ def __init__(self, _time_fn=time.time, _random_fn=random.random):
"""Initializes this class.
- @type attempt: int
- @param attempt: Current attempt number
@param _time_fn: Time function for unittests
@param _random_fn: Random number generator for unittests
"""
object.__init__(self)
- if attempt < 0:
- raise ValueError("Attempt must be zero or positive")
-
- self._attempt = attempt
+ self._timeouts = iter(self._TIMEOUT_PER_ATTEMPT)
self._time_fn = _time_fn
self._random_fn = _random_fn
- try:
- timeout = self._TIMEOUT_PER_ATTEMPT[attempt]
- except IndexError:
- # No more timeouts, do blocking acquire
- timeout = None
-
- self._running_timeout = locking.RunningTimeout(timeout, False,
- _time_fn=_time_fn)
-
def NextAttempt(self):
- """Returns the strategy for the next attempt.
+ """Returns the timeout for the next attempt.
"""
- return _LockAttemptTimeoutStrategy(attempt=self._attempt + 1,
- _time_fn=self._time_fn,
- _random_fn=self._random_fn)
-
- def CalcRemainingTimeout(self):
- """Returns the remaining timeout.
-
- """
- timeout = self._running_timeout.Remaining()
+ 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
"""
- def ReportLocks(self, msg):
- """Report lock operations.
+ def CheckCancel(self):
+ """Check whether job has been cancelled.
"""
+ def SubmitManyJobs(self, jobs):
+ """Submits jobs for processing.
+
+ See L{jqueue.JobQueue.SubmitManyJobs}.
+
+ """
+ raise NotImplementedError
+
+
+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,
- }
+ DISPATCH_TABLE = _ComputeDispatchTable()
def __init__(self, context, ec_id):
"""Constructor for Processor
self.rpc = rpc.RpcRunner(context.cfg)
self.hmclass = HooksMaster
- def _ReportLocks(self, level, names, shared, timeout, acquired, result):
- """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
@param shared: Whether the locks should be acquired in shared mode
@type timeout: None or float
@param timeout: Timeout for acquiring the locks
- @type acquired: bool
- @param acquired: Whether the locks have already been acquired
- @type result: None or set
- @param result: Result from L{locking.GanetiLockManager.acquire}
+ @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
+ amount of time
"""
- parts = []
-
- # Build message
- if acquired:
- if result is None:
- parts.append("timeout")
- else:
- parts.append("acquired")
- else:
- parts.append("waiting")
- if timeout is None:
- parts.append("blocking")
- else:
- parts.append("timeout=%0.6fs" % timeout)
-
- parts.append(locking.LEVEL_NAMES[level])
-
- if names == locking.ALL_SET:
- parts.append("ALL")
- elif isinstance(names, basestring):
- parts.append(names)
- else:
- parts.append(",".join(sorted(names)))
-
- if shared:
- parts.append("shared")
- else:
- parts.append("exclusive")
-
- msg = "/".join(parts)
+ if self._cbs:
+ self._cbs.CheckCancel()
- logging.debug("LU locks %s", msg)
+ acquired = self.context.glm.acquire(level, names, shared=shared,
+ timeout=timeout, priority=priority)
- if self._cbs:
- self._cbs.ReportLocks(msg)
+ if acquired is None:
+ raise LockAcquireTimeout()
- def _AcquireLocks(self, level, names, shared, timeout):
- """Acquires locks via the Ganeti lock manager.
+ return acquired
- @type level: int
- @param level: Lock level
- @type names: list or string
- @param names: Lock names
- @type shared: bool
- @param shared: Whether the locks should be acquired in shared mode
- @type timeout: None or float
- @param timeout: Timeout for acquiring the locks
+ def _ProcessResult(self, result):
+ """
"""
- self._ReportLocks(level, names, shared, timeout, False, None)
+ if isinstance(result, cmdlib.ResultWithJobs):
+ # Submit jobs
+ job_submission = self._cbs.SubmitManyJobs(result.jobs)
- acquired = self.context.glm.acquire(level, names, shared=shared,
- timeout=timeout)
+ # Build dictionary
+ result = result.other
- self._ReportLocks(level, names, shared, timeout, True, acquired)
+ assert constants.JOB_IDS_KEY not in result, \
+ "Key '%s' found in additional return values" % constants.JOB_IDS_KEY
- return acquired
+ result[constants.JOB_IDS_KEY] = job_submission
+
+ return result
def _ExecLU(self, lu):
"""Logical Unit execution sequence.
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
return lu.dry_run_result
try:
- result = lu.Exec(self._Feedback)
+ result = self._ProcessResult(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:
return result
- def _LockAndExecLU(self, lu, level, calc_timeout):
+ 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
needed_locks = lu.needed_locks[level]
acquired = self._AcquireLocks(level, needed_locks, share,
- calc_timeout())
-
- if acquired is None:
- raise _LockAcquireTimeout()
-
+ calc_timeout(), priority)
else:
# Adding locks
add_locks = lu.add_locks[level]
try:
lu.acquired_locks[level] = acquired
- result = self._LockAndExecLU(lu, level + 1, calc_timeout)
+ 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])
self.context.glm.release(level)
else:
- result = self._LockAndExecLU(lu, level + 1, calc_timeout)
+ 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")
-
- timeout_strategy = _LockAttemptTimeoutStrategy()
+ # 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:
+ lu = lu_class(self, op, self.context, self.rpc)
+ lu.ExpandNames()
+ assert lu.needed_locks is not None, "needed_locks not set by LU"
- while True:
try:
- acquire_timeout = timeout_strategy.CalcRemainingTimeout()
-
- # 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.
- if self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
- not lu_class.REQ_BGL, acquire_timeout) is None:
- raise _LockAcquireTimeout()
-
- try:
- lu = lu_class(self, op, self.context, self.rpc)
- lu.ExpandNames()
- assert lu.needed_locks is not None, "needed_locks not set by LU"
-
- try:
- return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE,
- timeout_strategy.CalcRemainingTimeout)
- finally:
- if self._ec_id:
- self.context.cfg.DropECReservations(self._ec_id)
-
- finally:
- self.context.glm.release(locking.LEVEL_CLUSTER)
-
- except _LockAcquireTimeout:
- # Timeout while waiting for lock, try again
- pass
-
- timeout_strategy = timeout_strategy.NextAttempt()
-
+ 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)
finally:
self._cbs = None
- def _Feedback(self, *args):
+ def Log(self, *args):
"""Forward call to feedback callback function.
"""
"""
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.
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.
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:
- errors.ProgrammerError("Tried to use execution context id when not set")
+ raise errors.ProgrammerError("Tried to use execution context id when"
+ " not set")
return self._ec_id
self.callfn = callfn
self.lu = lu
self.op = lu.op
- self.env, node_list_pre, node_list_post = self._BuildEnv()
- self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
- constants.HOOKS_PHASE_POST: node_list_post}
+ self.pre_env = self._BuildEnv(constants.HOOKS_PHASE_PRE)
- def _BuildEnv(self):
+ if self.lu.HPATH is None:
+ nodes = (None, None)
+ else:
+ nodes = map(frozenset, self.lu.BuildHooksNodes())
+
+ (self.pre_nodes, self.post_nodes) = nodes
+
+ def _BuildEnv(self, phase):
"""Compute the environment and the target nodes.
Based on the opcode and the current node list, this builds the
environment for the hooks and the target node list for the run.
"""
- env = {
- "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
- "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
- "GANETI_OP_CODE": self.op.OP_ID,
- "GANETI_OBJECT_TYPE": self.lu.HTYPE,
- "GANETI_DATA_DIR": constants.DATA_DIR,
- }
+ if phase == constants.HOOKS_PHASE_PRE:
+ prefix = "GANETI_"
+ elif phase == constants.HOOKS_PHASE_POST:
+ prefix = "GANETI_POST_"
+ else:
+ raise AssertionError("Unknown phase '%s'" % phase)
+
+ env = {}
if self.lu.HPATH is not None:
- lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
+ lu_env = self.lu.BuildHooksEnv()
if lu_env:
- for key in lu_env:
- env["GANETI_" + key] = lu_env[key]
+ assert not compat.any(key.upper().startswith(prefix) for key in lu_env)
+ env.update(("%s%s" % (prefix, key), value)
+ for (key, value) in lu_env.items())
+
+ if phase == constants.HOOKS_PHASE_PRE:
+ assert compat.all((key.startswith("GANETI_") and
+ not key.startswith("GANETI_POST_"))
+ for key in env)
+
+ elif phase == constants.HOOKS_PHASE_POST:
+ assert compat.all(key.startswith("GANETI_POST_") for key in env)
+ assert isinstance(self.pre_env, dict)
+
+ # Merge with pre-phase environment
+ assert not compat.any(key.startswith("GANETI_POST_")
+ for key in self.pre_env)
+ env.update(self.pre_env)
else:
- lu_nodes_pre = lu_nodes_post = []
+ raise AssertionError("Unknown phase '%s'" % phase)
- return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post)
+ return env
- def _RunWrapper(self, node_list, hpath, phase):
+ def _RunWrapper(self, node_list, hpath, phase, phase_env):
"""Simple wrapper over self.callfn.
This method fixes the environment before doing the rpc call.
"""
- env = self.env.copy()
- env["GANETI_HOOKS_PHASE"] = phase
- env["GANETI_HOOKS_PATH"] = hpath
- if self.lu.cfg is not None:
- env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName()
- env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode()
+ cfg = self.lu.cfg
+
+ env = {
+ "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
+ "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
+ "GANETI_OP_CODE": self.op.OP_ID,
+ "GANETI_DATA_DIR": constants.DATA_DIR,
+ "GANETI_HOOKS_PHASE": phase,
+ "GANETI_HOOKS_PATH": hpath,
+ }
+
+ if self.lu.HTYPE:
+ env["GANETI_OBJECT_TYPE"] = self.lu.HTYPE
+
+ if cfg is not None:
+ env["GANETI_CLUSTER"] = cfg.GetClusterName()
+ env["GANETI_MASTER"] = cfg.GetMasterNode()
+ if phase_env:
+ assert not (set(env) & set(phase_env)), "Environment variables conflict"
+ env.update(phase_env)
+
+ # Convert everything to strings
env = dict([(str(key), str(val)) for key, val in env.iteritems()])
+ assert compat.all(key == "PATH" or key.startswith("GANETI_")
+ for key in env)
+
return self.callfn(node_list, hpath, phase, env)
def RunPhase(self, phase, nodes=None):
@raise errors.HooksAbort: on failure of one of the hooks
"""
- if not self.node_list[phase] and not nodes:
+ if phase == constants.HOOKS_PHASE_PRE:
+ if nodes is None:
+ nodes = self.pre_nodes
+ env = self.pre_env
+ elif phase == constants.HOOKS_PHASE_POST:
+ if nodes is None:
+ nodes = self.post_nodes
+ env = self._BuildEnv(phase)
+ else:
+ raise AssertionError("Unknown phase '%s'" % phase)
+
+ if not nodes:
# empty node list, we should not attempt to run this as either
# we're in the cluster init phase and the rpc client part can't
# even attempt to run, or this LU doesn't do hooks at all
return
- hpath = self.lu.HPATH
- if nodes is not None:
- results = self._RunWrapper(nodes, hpath, phase)
- else:
- results = self._RunWrapper(self.node_list[phase], hpath, phase)
- errs = []
+
+ results = self._RunWrapper(nodes, self.lu.HPATH, phase, env)
if not results:
msg = "Communication Failure"
if phase == constants.HOOKS_PHASE_PRE:
else:
self.lu.LogWarning(msg)
return results
+
+ errs = []
for node_name in results:
res = results[node_name]
if res.offline:
continue
+
msg = res.fail_msg
if msg:
self.lu.LogWarning("Communication failure to node %s: %s",
node_name, msg)
continue
+
for script, hkr, output in res.payload:
if hkr == constants.HKR_FAIL:
if phase == constants.HOOKS_PHASE_PRE:
output = "(no output)"
self.lu.LogWarning("On %s script %s failed, output: %s" %
(node_name, script, output))
+
if errs and phase == constants.HOOKS_PHASE_PRE:
raise errors.HooksAbort(errs)
+
return results
def RunConfigUpdate(self):
phase = constants.HOOKS_PHASE_POST
hpath = constants.HOOKS_NAME_CFGUPDATE
nodes = [self.lu.cfg.GetMasterNode()]
- self._RunWrapper(nodes, hpath, phase)
+ self._RunWrapper(nodes, hpath, phase, self.pre_env)