#
#
-# 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 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):
"""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
return timeout
-class OpExecCbBase: # pylint: disable-msg=W0232
+class OpExecCbBase: # pylint: disable=W0232
"""Base class for OpCode execution callbacks.
"""
"""
+ 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)
+
+
+def _RpcResultsToHooksResults(rpc_results):
+ """Function to convert RPC results to the format expected by HooksMaster.
+
+ @type rpc_results: dict(node: L{rpc.RpcResult})
+ @param rpc_results: RPC results
+ @rtype: dict(node: (fail_msg, offline, hooks_results))
+ @return: RPC results unpacked according to the format expected by
+ L({mcpu.HooksMaster}
+
+ """
+ return dict((node, (rpc_res.fail_msg, rpc_res.offline, rpc_res.payload))
+ for (node, rpc_res) in rpc_results.items())
+
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,
- opcodes.OpQuery: cmdlib.LUQuery,
- opcodes.OpQueryFields: cmdlib.LUQueryFields,
- # 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.OpMigrateNode: cmdlib.LUMigrateNode,
- opcodes.OpNodeEvacuationStrategy: cmdlib.LUNodeEvacuationStrategy,
- # 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,
- # node group lu
- opcodes.OpQueryGroups: cmdlib.LUQueryGroups,
- # os lu
- opcodes.OpDiagnoseOS: cmdlib.LUDiagnoseOS,
- # exports lu
- opcodes.OpQueryExports: cmdlib.LUQueryExports,
- opcodes.OpPrepareExport: cmdlib.LUPrepareExport,
- 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,
- opcodes.OpTestJobqueue: cmdlib.LUTestJobqueue,
- }
+ DISPATCH_TABLE = _ComputeDispatchTable()
def __init__(self, context, ec_id):
"""Constructor for Processor
self.context = context
self._ec_id = ec_id
self._cbs = None
- self.rpc = rpc.RpcRunner(context.cfg)
+ self.rpc = context.rpc
self.hmclass = HooksMaster
def _AcquireLocks(self, level, names, shared, timeout, priority):
return acquired
+ def _ProcessResult(self, result):
+ """Examines opcode result.
+
+ If necessary, additional processing on the result is done.
+
+ """
+ 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)
+
+ hm = self.BuildHooksManager(lu)
h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE)
lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results,
self.Log, None)
return lu.dry_run_result
try:
- result = lu.Exec(self.Log)
+ 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)
return result
+ def BuildHooksManager(self, lu):
+ return self.hmclass.BuildFromLu(lu.rpc.call_hooks_runner, lu)
+
def _LockAndExecLU(self, lu, level, calc_timeout, priority):
"""Execute a Logical Unit, with the needed locks.
# Acquiring locks
needed_locks = lu.needed_locks[level]
- acquired = self._AcquireLocks(level, needed_locks, share,
- calc_timeout(), priority)
+ self._AcquireLocks(level, needed_locks, share,
+ calc_timeout(), priority)
else:
# Adding locks
add_locks = lu.add_locks[level]
" 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:
"""
if not isinstance(op, opcodes.OpCode):
raise errors.ProgrammerError("Non-opcode instance passed"
- " to ExecOpcode")
+ " to ExecOpcode (%s)" % type(op))
lu_class = self.DISPATCH_TABLE.get(op.__class__, None)
if lu_class is None:
assert lu.needed_locks is not None, "needed_locks not set by LU"
try:
- return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout,
- priority)
+ result = self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout,
+ priority)
finally:
if self._ec_id:
self.context.cfg.DropECReservations(self._ec_id)
finally:
self._cbs = None
+ resultcheck_fn = op.OP_RESULT
+ if not (resultcheck_fn is None or resultcheck_fn(result)):
+ logging.error("Expected opcode result matching %s, got %s",
+ resultcheck_fn, result)
+ raise errors.OpResultError("Opcode result does not match %s" %
+ resultcheck_fn)
+
+ return result
+
def Log(self, *args):
"""Forward call to feedback callback function.
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
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, opcode, hooks_path, nodes, hooks_execution_fn,
+ hooks_results_adapt_fn, build_env_fn, log_fn, htype=None, cluster_name=None,
+ master_name=None):
+ """Base class for hooks masters.
+
+ This class invokes the execution of hooks according to the behaviour
+ specified by its parameters.
+
+ @type opcode: string
+ @param opcode: opcode of the operation to which the hooks are tied
+ @type hooks_path: string
+ @param hooks_path: prefix of the hooks directories
+ @type nodes: 2-tuple of lists
+ @param nodes: 2-tuple of lists containing nodes on which pre-hooks must be
+ run and nodes on which post-hooks must be run
+ @type hooks_execution_fn: function that accepts the following parameters:
+ (node_list, hooks_path, phase, environment)
+ @param hooks_execution_fn: function that will execute the hooks; can be
+ None, indicating that no conversion is necessary.
+ @type hooks_results_adapt_fn: function
+ @param hooks_results_adapt_fn: function that will adapt the return value of
+ hooks_execution_fn to the format expected by RunPhase
+ @type build_env_fn: function that returns a dictionary having strings as
+ keys
+ @param build_env_fn: function that builds the environment for the hooks
+ @type log_fn: function that accepts a string
+ @param log_fn: logging function
+ @type htype: string or None
+ @param htype: None or one of L{constants.HTYPE_CLUSTER},
+ L{constants.HTYPE_NODE}, L{constants.HTYPE_INSTANCE}
+ @type cluster_name: string
+ @param cluster_name: name of the cluster
+ @type master_name: string
+ @param master_name: name of the master
- """
- def __init__(self, callfn, lu):
- 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}
-
- def _BuildEnv(self):
+ """
+ self.opcode = opcode
+ self.hooks_path = hooks_path
+ self.hooks_execution_fn = hooks_execution_fn
+ self.hooks_results_adapt_fn = hooks_results_adapt_fn
+ self.build_env_fn = build_env_fn
+ self.log_fn = log_fn
+ self.htype = htype
+ self.cluster_name = cluster_name
+ self.master_name = master_name
+
+ self.pre_env = self._BuildEnv(constants.HOOKS_PHASE_PRE)
+ (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.hooks_path is not None:
+ phase_env = self.build_env_fn()
+ if phase_env:
+ assert not compat.any(key.upper().startswith(prefix)
+ for key in phase_env)
+ env.update(("%s%s" % (prefix, key), value)
+ for (key, value) in phase_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 executing the hooks.
+
+ """
env = {
- "PATH": "/sbin:/bin:/usr/sbin:/usr/bin",
+ "PATH": constants.HOOKS_PATH,
"GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
- "GANETI_OP_CODE": self.op.OP_ID,
- "GANETI_OBJECT_TYPE": self.lu.HTYPE,
+ "GANETI_OP_CODE": self.opcode,
"GANETI_DATA_DIR": constants.DATA_DIR,
+ "GANETI_HOOKS_PHASE": phase,
+ "GANETI_HOOKS_PATH": hpath,
}
- if self.lu.HPATH is not None:
- lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
- if lu_env:
- for key in lu_env:
- env["GANETI_" + key] = lu_env[key]
- else:
- lu_nodes_pre = lu_nodes_post = []
-
- return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post)
+ if self.htype:
+ env["GANETI_OBJECT_TYPE"] = self.htype
- def _RunWrapper(self, node_list, hpath, phase):
- """Simple wrapper over self.callfn.
+ if self.cluster_name is not None:
+ env["GANETI_CLUSTER"] = self.cluster_name
- This method fixes the environment before doing the rpc call.
+ if self.master_name is not None:
+ env["GANETI_MASTER"] = self.master_name
- """
- 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()
+ if phase_env:
+ env = utils.algo.JoinDisjointDicts(env, phase_env)
+ # Convert everything to strings
env = dict([(str(key), str(val)) for key, val in env.iteritems()])
- return self.callfn(node_list, hpath, phase, env)
+ assert compat.all(key == "PATH" or key.startswith("GANETI_")
+ for key in env)
+
+ return self.hooks_execution_fn(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.
+ It executes self.hooks_execution_fn, and after running
+ self.hooks_results_adapt_fn on its results it expects them to be in the form
+ {node_name: (fail_msg, [(script, result, output), ...]}).
@param phase: one of L{constants.HOOKS_PHASE_POST} or
L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
@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.hooks_path, phase, env)
if not results:
msg = "Communication Failure"
if phase == constants.HOOKS_PHASE_PRE:
raise errors.HooksFailure(msg)
else:
- self.lu.LogWarning(msg)
+ self.log_fn(msg)
return results
- for node_name in results:
- res = results[node_name]
- if res.offline:
+
+ converted_res = results
+ if self.hooks_results_adapt_fn:
+ converted_res = self.hooks_results_adapt_fn(results)
+
+ errs = []
+ for node_name, (fail_msg, offline, hooks_results) in converted_res.items():
+ if offline:
continue
- msg = res.fail_msg
- if msg:
- self.lu.LogWarning("Communication failure to node %s: %s",
- node_name, msg)
+
+ if fail_msg:
+ self.log_fn("Communication failure to node %s: %s", node_name, fail_msg)
continue
- for script, hkr, output in res.payload:
+
+ for script, hkr, output in hooks_results:
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))
+ self.log_fn("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)
+ nodes = [self.master_name]
+ self._RunWrapper(nodes, hpath, phase, self.pre_env)
+
+ @staticmethod
+ def BuildFromLu(hooks_execution_fn, lu):
+ if lu.HPATH is None:
+ nodes = (None, None)
+ else:
+ nodes = map(frozenset, lu.BuildHooksNodes())
+
+ master_name = cluster_name = None
+ if lu.cfg:
+ master_name = lu.cfg.GetMasterNode()
+ cluster_name = lu.cfg.GetClusterName()
+
+ return HooksMaster(lu.op.OP_ID, lu.HPATH, nodes, hooks_execution_fn,
+ _RpcResultsToHooksResults, lu.BuildHooksEnv,
+ lu.LogWarning, lu.HTYPE, cluster_name, master_name)
projects / ganeti-local / blobdiff