-
Michael Hanselmann authored
The design details can be seen in the design document (doc/design-lu-generated-jobs.rst). Signed-off-by:
Michael Hanselmann <hansmi@google.com> Reviewed-by:
René Nussbaumer <rn@google.com>
6a373640
mcpu.py 18.07 KiB
#
#
# 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
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
# 02110-1301, USA.
"""Module implementing the logic behind the cluster operations
This module implements the logic for doing operations in the cluster. There
are two kinds of classes defined:
- logical units, which know how to deal with their specific opcode only
- the processor, which dispatches the opcodes to their logical units
"""
import logging
import random
import time
from ganeti import opcodes
from ganeti import constants
from ganeti import errors
from ganeti import rpc
from ganeti import cmdlib
from ganeti import locking
from ganeti import utils
from ganeti import compat
_OP_PREFIX = "Op"
_LU_PREFIX = "LU"
class LockAcquireTimeout(Exception):
"""Exception to report timeouts on acquiring locks.
"""
def _CalculateLockAttemptTimeouts():
"""Calculate timeouts for lock attempts.
"""
result = [1.0]
# Wait for a total of at least 150s before doing a blocking acquire
while sum(result) < 150.0:
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
result.append(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.
"""
def NotifyStart(self):
"""Called when we are about to execute the LU.
This function is called when we're about to start the lu's Exec() method,
that is, after we have acquired all locks.
"""
def Feedback(self, *args):
"""Sends feedback from the LU code to the end-user.
"""
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 = _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.rpc = rpc.RpcRunner(context.cfg)
self.hmclass = HooksMaster
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 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
"""
if self._cbs:
self._cbs.CheckCancel()
acquired = self.context.glm.acquire(level, names, shared=shared,
timeout=timeout, priority=priority)
if acquired is None:
raise LockAcquireTimeout()
return acquired
def _ProcessResult(self, result):
"""
"""
if isinstance(result, cmdlib.ResultWithJobs):
# Submit jobs
job_submission = self._cbs.SubmitManyJobs(result.jobs)
# Build dictionary
result = result.other
assert constants.JOB_IDS_KEY not in result, \
"Key '%s' found in additional return values" % constants.JOB_IDS_KEY
result[constants.JOB_IDS_KEY] = job_submission
return result
def _ExecLU(self, lu):
"""Logical Unit execution sequence.
"""
write_count = self.context.cfg.write_count
lu.CheckPrereq()
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.Log, None)
if getattr(lu.op, "dry_run", False):
# in this mode, no post-hooks are run, and the config is not
# written (as it might have been modified by another LU, and we
# shouldn't do writeout on behalf of other threads
self.LogInfo("dry-run mode requested, not actually executing"
" the operation")
return lu.dry_run_result
try:
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.Log, result)
finally:
# FIXME: This needs locks if not lu_class.REQ_BGL
if write_count != self.context.cfg.write_count:
hm.RunConfigUpdate()
return result
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
proceeds up, till there are no more locks to acquire. Then it executes the
given LU and its opcodes.
"""
adding_locks = level in lu.add_locks
acquiring_locks = level in lu.needed_locks
if level not in locking.LEVELS:
if self._cbs:
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")
elif adding_locks or acquiring_locks:
lu.DeclareLocks(level)
share = lu.share_locks[level]
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:
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, calc_timeout, priority)
return result
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:
# 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"
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)
finally:
self._cbs = None
def Log(self, *args):
"""Forward call to feedback callback function.
"""
if self._cbs:
self._cbs.Feedback(*args)
def LogStep(self, current, total, message):
"""Log a change in LU execution progress.
"""
logging.debug("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.
The optional keyword argument is 'hint' and can be used to show a
hint to the user (presumably related to the warning). If the
message is empty, it will not be printed at all, allowing one to
show only a hint.
"""
assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \
"Invalid keyword arguments for LogWarning (%s)" % str(kwargs)
if args:
message = message % tuple(args)
if message:
logging.warning(message)
self.Log(" - WARNING: %s" % message)
if "hint" in kwargs:
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.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):
"""Hooks master.
This class distributes the run commands to the nodes based on the
specific LU class.
In order to remove the direct dependency on the rpc module, the
constructor needs a function which actually does the remote
call. This will usually be rpc.call_hooks_runner, but any function
which behaves the same works.
"""
def __init__(self, callfn, lu):
self.callfn = callfn
self.lu = lu
self.op = lu.op
self.pre_env = self._BuildEnv(constants.HOOKS_PHASE_PRE)
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.
"""
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 = self.lu.BuildHooksEnv()
if lu_env:
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:
raise AssertionError("Unknown phase '%s'" % phase)
return env
def _RunWrapper(self, node_list, hpath, phase, phase_env):
"""Simple wrapper over self.callfn.
This method fixes the environment before doing the rpc call.
"""
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):
"""Run all the scripts for a phase.
This is the main function of the HookMaster.
@param phase: one of L{constants.HOOKS_PHASE_POST} or
L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
@param nodes: overrides the predefined list of nodes for the given phase
@return: the processed results of the hooks multi-node rpc call
@raise errors.HooksFailure: on communication failure to the nodes
@raise errors.HooksAbort: on failure of one of the hooks
"""
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
results = self._RunWrapper(nodes, self.lu.HPATH, phase, env)
if not results:
msg = "Communication Failure"
if phase == constants.HOOKS_PHASE_PRE:
raise errors.HooksFailure(msg)
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:
errs.append((node_name, script, output))
else:
if not output:
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):
"""Run the special configuration update hook
This is a special hook that runs only on the master after each
top-level LI if the configuration has been updated.
"""
phase = constants.HOOKS_PHASE_POST
hpath = constants.HOOKS_NAME_CFGUPDATE
nodes = [self.lu.cfg.GetMasterNode()]
self._RunWrapper(nodes, hpath, phase, self.pre_env)