X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/70a6a92676eefc671ab8acd3bad03fc85f2f0906..52f3310349e34843fdcb00ae993766eeb6c05cc5:/lib/mcpu.py diff --git a/lib/mcpu.py b/lib/mcpu.py index a724eb9..5728c3b 100644 --- a/lib/mcpu.py +++ b/lib/mcpu.py @@ -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 @@ -39,6 +39,11 @@ 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): @@ -51,23 +56,23 @@ def _CalculateLockAttemptTimeouts(): """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 @@ -139,74 +144,37 @@ class OpExecCbBase: # pylint: disable-msg=W0232 """ + 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, - 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 @@ -249,6 +217,24 @@ class Processor(object): 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. @@ -269,7 +255,7 @@ class Processor(object): 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) @@ -442,8 +428,12 @@ class Processor(object): 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 @@ -463,50 +453,90 @@ class HooksMaster(object): 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) + + 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): + 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): @@ -522,17 +552,24 @@ class HooksMaster(object): @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: @@ -540,15 +577,19 @@ class HooksMaster(object): 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: @@ -558,8 +599,10 @@ class HooksMaster(object): 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): @@ -572,4 +615,4 @@ class HooksMaster(object): 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)