#!/usr/bin/python3.7 ################################################################### # Copyright (c) 2016 by TAOS Technologies, Inc. # All rights reserved. # # This file is proprietary and confidential to TAOS Technologies. # No part of this file may be reproduced, stored, transmitted, # disclosed or used in any form or by any means other than as # expressly provided by the written permission from Jianhui Tao # ################################################################### # -*- coding: utf-8 -*- from __future__ import annotations # For type hinting before definition, ref: https://stackoverflow.com/questions/33533148/how-do-i-specify-that-the-return-type-of-a-method-is-the-same-as-the-class-itsel import sys import os import traceback # Require Python 3 if sys.version_info[0] < 3: raise Exception("Must be using Python 3") import getopt import argparse import copy import threading import random import time import logging import datetime import textwrap from typing import List from typing import Dict from typing import Set from util.log import * from util.dnodes import * from util.cases import * from util.sql import * import crash_gen import taos # Global variables, tried to keep a small number. # Command-line/Environment Configurations, will set a bit later # ConfigNameSpace = argparse.Namespace gConfig = argparse.Namespace() # Dummy value, will be replaced later logger = None def runThread(wt: WorkerThread): wt.run() class CrashGenError(Exception): def __init__(self, msg=None, errno=None): self.msg = msg self.errno = errno def __str__(self): return self.msg class WorkerThread: def __init__(self, pool: ThreadPool, tid, tc: ThreadCoordinator, # te: TaskExecutor, ): # note: main thread context! # self._curStep = -1 self._pool = pool self._tid = tid self._tc = tc # type: ThreadCoordinator # self.threadIdent = threading.get_ident() self._thread = threading.Thread(target=runThread, args=(self,)) self._stepGate = threading.Event() # Let us have a DB connection of our own if ( gConfig.per_thread_db_connection ): # type: ignore self._dbConn = DbConn() def logDebug(self, msg): logger.debug(" TRD[{}] {}".format(self._tid, msg)) def logInfo(self, msg): logger.info(" TRD[{}] {}".format(self._tid, msg)) def getTaskExecutor(self): return self._tc.getTaskExecutor() def start(self): self._thread.start() # AFTER the thread is recorded def run(self): # initialization after thread starts, in the thread context # self.isSleeping = False logger.info("Starting to run thread: {}".format(self._tid)) if ( gConfig.per_thread_db_connection ): # type: ignore self._dbConn.open() self._doTaskLoop() # clean up if ( gConfig.per_thread_db_connection ): # type: ignore self._dbConn.close() def _doTaskLoop(self) : # while self._curStep < self._pool.maxSteps: # tc = ThreadCoordinator(None) while True: tc = self._tc # Thread Coordinator, the overall master tc.crossStepBarrier() # shared barrier first, INCLUDING the last one logger.debug("[TRD] Worker thread [{}] exited barrier...".format(self._tid)) self.crossStepGate() # then per-thread gate, after being tapped logger.debug("[TRD] Worker thread [{}] exited step gate...".format(self._tid)) if not self._tc.isRunning(): logger.debug("[TRD] Thread Coordinator not running any more, worker thread now stopping...") break logger.debug("[TRD] Worker thread [{}] about to fetch task".format(self._tid)) task = tc.fetchTask() logger.debug("[TRD] Worker thread [{}] about to execute task: {}".format(self._tid, task.__class__.__name__)) task.execute(self) tc.saveExecutedTask(task) logger.debug("[TRD] Worker thread [{}] finished executing task".format(self._tid)) def verifyThreadSelf(self): # ensure we are called by this own thread if ( threading.get_ident() != self._thread.ident ): raise RuntimeError("Unexpectly called from other threads") def verifyThreadMain(self): # ensure we are called by the main thread if ( threading.get_ident() != threading.main_thread().ident ): raise RuntimeError("Unexpectly called from other threads") def verifyThreadAlive(self): if ( not self._thread.is_alive() ): raise RuntimeError("Unexpected dead thread") # A gate is different from a barrier in that a thread needs to be "tapped" def crossStepGate(self): self.verifyThreadAlive() self.verifyThreadSelf() # only allowed by ourselves # Wait again at the "gate", waiting to be "tapped" logger.debug("[TRD] Worker thread {} about to cross the step gate".format(self._tid)) self._stepGate.wait() self._stepGate.clear() # self._curStep += 1 # off to a new step... def tapStepGate(self): # give it a tap, release the thread waiting there self.verifyThreadAlive() self.verifyThreadMain() # only allowed for main thread logger.debug("[TRD] Tapping worker thread {}".format(self._tid)) self._stepGate.set() # wake up! time.sleep(0) # let the released thread run a bit def execSql(self, sql): # TODO: expose DbConn directly if ( gConfig.per_thread_db_connection ): return self._dbConn.execute(sql) else: return self._tc.getDbManager().getDbConn().execute(sql) def getDbConn(self): if ( gConfig.per_thread_db_connection ): return self._dbConn else: return self._tc.getDbManager().getDbConn() # def querySql(self, sql): # not "execute", since we are out side the DB context # if ( gConfig.per_thread_db_connection ): # return self._dbConn.query(sql) # else: # return self._tc.getDbState().getDbConn().query(sql) class ThreadCoordinator: def __init__(self, pool, dbManager): self._curStep = -1 # first step is 0 self._pool = pool # self._wd = wd self._te = None # prepare for every new step self._dbManager = dbManager self._executedTasks: List[Task] = [] # in a given step self._lock = threading.RLock() # sync access for a few things self._stepBarrier = threading.Barrier(self._pool.numThreads + 1) # one barrier for all threads self._execStats = ExecutionStats() def getTaskExecutor(self): return self._te def getDbManager(self) -> DbManager : return self._dbManager def crossStepBarrier(self): self._stepBarrier.wait() def run(self): self._pool.createAndStartThreads(self) # Coordinate all threads step by step self._curStep = -1 # not started yet maxSteps = gConfig.max_steps # type: ignore self._execStats.startExec() # start the stop watch failed = False while(self._curStep < maxSteps-1 and not failed): # maxStep==10, last curStep should be 9 if not gConfig.debug: print(".", end="", flush=True) # print this only if we are not in debug mode logger.debug("[TRD] Main thread going to sleep") # Now ready to enter a step self.crossStepBarrier() # let other threads go past the pool barrier, but wait at the thread gate self._stepBarrier.reset() # Other worker threads should now be at the "gate" # At this point, all threads should be pass the overall "barrier" and before the per-thread "gate" try: self._dbManager.getStateMachine().transition(self._executedTasks) # at end of step, transiton the DB state except taos.error.ProgrammingError as err: if ( err.msg == 'network unavailable' ): # broken DB connection logger.info("DB connection broken, execution failed") traceback.print_stack() failed = True self._te = None # Not running any more self._execStats.registerFailure("Broken DB Connection") # continue # don't do that, need to tap all threads at end, and maybe signal them to stop else: raise finally: pass self.resetExecutedTasks() # clear the tasks after we are done # Get ready for next step logger.debug("<-- Step {} finished".format(self._curStep)) self._curStep += 1 # we are about to get into next step. TODO: race condition here! logger.debug("\r\n--> Step {} starts with main thread waking up".format(self._curStep)) # Now not all threads had time to go to sleep # A new TE for the new step if not failed: # only if not failed self._te = TaskExecutor(self._curStep) logger.debug("[TRD] Main thread waking up at step {}, tapping worker threads".format(self._curStep)) # Now not all threads had time to go to sleep self.tapAllThreads() logger.debug("Main thread ready to finish up...") if not failed: # only in regular situations self.crossStepBarrier() # Cross it one last time, after all threads finish self._stepBarrier.reset() logger.debug("Main thread in exclusive zone...") self._te = None # No more executor, time to end logger.debug("Main thread tapping all threads one last time...") self.tapAllThreads() # Let the threads run one last time logger.debug("Main thread joining all threads") self._pool.joinAll() # Get all threads to finish logger.info("All worker thread finished") self._execStats.endExec() def logStats(self): self._execStats.logStats() def tapAllThreads(self): # in a deterministic manner wakeSeq = [] for i in range(self._pool.numThreads): # generate a random sequence if Dice.throw(2) == 1 : wakeSeq.append(i) else: wakeSeq.insert(0, i) logger.debug("[TRD] Main thread waking up worker thread: {}".format(str(wakeSeq))) # TODO: set dice seed to a deterministic value for i in wakeSeq: self._pool.threadList[i].tapStepGate() # TODO: maybe a bit too deep?! time.sleep(0) # yield def isRunning(self): return self._te != None def fetchTask(self) -> Task : if ( not self.isRunning() ): # no task raise RuntimeError("Cannot fetch task when not running") # return self._wd.pickTask() # Alternatively, let's ask the DbState for the appropriate task # dbState = self.getDbState() # tasks = dbState.getTasksAtState() # TODO: create every time? # nTasks = len(tasks) # i = Dice.throw(nTasks) # logger.debug(" (dice:{}/{}) ".format(i, nTasks)) # # return copy.copy(tasks[i]) # Needs a fresh copy, to save execution results, etc. # return tasks[i].clone() # TODO: still necessary? taskType = self.getDbManager().getStateMachine().pickTaskType() # pick a task type for current state return taskType(self.getDbManager(), self._execStats) # create a task from it def resetExecutedTasks(self): self._executedTasks = [] # should be under single thread def saveExecutedTask(self, task): with self._lock: self._executedTasks.append(task) # We define a class to run a number of threads in locking steps. class ThreadPool: def __init__(self, numThreads, maxSteps): self.numThreads = numThreads self.maxSteps = maxSteps # Internal class variables self.curStep = 0 self.threadList = [] # starting to run all the threads, in locking steps def createAndStartThreads(self, tc: ThreadCoordinator): for tid in range(0, self.numThreads): # Create the threads workerThread = WorkerThread(self, tid, tc) self.threadList.append(workerThread) workerThread.start() # start, but should block immediately before step 0 def joinAll(self): for workerThread in self.threadList: logger.debug("Joining thread...") workerThread._thread.join() # A queue of continguous POSITIVE integers, used by DbManager to generate continuous numbers # for new table names class LinearQueue(): def __init__(self): self.firstIndex = 1 # 1st ever element self.lastIndex = 0 self._lock = threading.RLock() # our functions may call each other self.inUse = set() # the indexes that are in use right now def toText(self): return "[{}..{}], in use: {}".format(self.firstIndex, self.lastIndex, self.inUse) # Push (add new element, largest) to the tail, and mark it in use def push(self): with self._lock: # if ( self.isEmpty() ): # self.lastIndex = self.firstIndex # return self.firstIndex # Otherwise we have something self.lastIndex += 1 self.allocate(self.lastIndex) # self.inUse.add(self.lastIndex) # mark it in use immediately return self.lastIndex def pop(self): with self._lock: if ( self.isEmpty() ): # raise RuntimeError("Cannot pop an empty queue") return False # TODO: None? index = self.firstIndex if ( index in self.inUse ): return False self.firstIndex += 1 return index def isEmpty(self): return self.firstIndex > self.lastIndex def popIfNotEmpty(self): with self._lock: if (self.isEmpty()): return 0 return self.pop() def allocate(self, i): with self._lock: # logger.debug("LQ allocating item {}".format(i)) if ( i in self.inUse ): raise RuntimeError("Cannot re-use same index in queue: {}".format(i)) self.inUse.add(i) def release(self, i): with self._lock: # logger.debug("LQ releasing item {}".format(i)) self.inUse.remove(i) # KeyError possible, TODO: why? def size(self): return self.lastIndex + 1 - self.firstIndex def pickAndAllocate(self): if ( self.isEmpty() ): return None with self._lock: cnt = 0 # counting the interations while True: cnt += 1 if ( cnt > self.size()*10 ): # 10x iteration already # raise RuntimeError("Failed to allocate LinearQueue element") return None ret = Dice.throwRange(self.firstIndex, self.lastIndex+1) if ( not ret in self.inUse ): self.allocate(ret) return ret class DbConn: def __init__(self): self._conn = None self._cursor = None self.isOpen = False def open(self): # Open connection if ( self.isOpen ): raise RuntimeError("Cannot re-open an existing DB connection") cfgPath = "../../build/test/cfg" self._conn = taos.connect(host="127.0.0.1", config=cfgPath) # TODO: make configurable self._cursor = self._conn.cursor() # Get the connection/cursor ready self._cursor.execute('reset query cache') # self._cursor.execute('use db') # note we do this in _findCurrenState # Open connection self._tdSql = TDSql() self._tdSql.init(self._cursor) self.isOpen = True def resetDb(self): # reset the whole database, etc. if ( not self.isOpen ): raise RuntimeError("Cannot reset database until connection is open") # self._tdSql.prepare() # Recreate database, etc. self._cursor.execute('drop database if exists db') logger.debug("Resetting DB, dropped database") # self._cursor.execute('create database db') # self._cursor.execute('use db') # tdSql.execute('show databases') def close(self): if ( not self.isOpen ): raise RuntimeError("Cannot clean up database until connection is open") self._tdSql.close() self.isOpen = False def execute(self, sql): if ( not self.isOpen ): raise RuntimeError("Cannot execute database commands until connection is open") logger.debug("[SQL] Executing SQL: {}".format(sql)) nRows = self._tdSql.execute(sql) logger.debug("[SQL] Execution Result, nRows = {}, SQL = {}".format(nRows, sql)) return nRows def query(self, sql) : # return rows affected if ( not self.isOpen ): raise RuntimeError("Cannot query database until connection is open") logger.debug("[SQL] Executing SQL: {}".format(sql)) nRows = self._tdSql.query(sql) logger.debug("[SQL] Execution Result, nRows = {}, SQL = {}".format(nRows, sql)) return nRows # results are in: return self._tdSql.queryResult def getQueryResult(self): return self._tdSql.queryResult def _queryAny(self, sql) : # actual query result as an int if ( not self.isOpen ): raise RuntimeError("Cannot query database until connection is open") tSql = self._tdSql nRows = tSql.query(sql) if nRows != 1 : raise RuntimeError("Unexpected result for query: {}, rows = {}".format(sql, nRows)) if tSql.queryRows != 1 or tSql.queryCols != 1: raise RuntimeError("Unexpected result set for query: {}".format(sql)) return tSql.queryResult[0][0] def queryScalar(self, sql) -> int : return self._queryAny(sql) def queryString(self, sql) -> str : return self._queryAny(sql) class AnyState: STATE_INVALID = -1 STATE_EMPTY = 0 # nothing there, no even a DB STATE_DB_ONLY = 1 # we have a DB, but nothing else STATE_TABLE_ONLY = 2 # we have a table, but totally empty STATE_HAS_DATA = 3 # we have some data in the table _stateNames = ["Invalid", "Empty", "DB_Only", "Table_Only", "Has_Data"] STATE_VAL_IDX = 0 CAN_CREATE_DB = 1 CAN_DROP_DB = 2 CAN_CREATE_FIXED_SUPER_TABLE = 3 CAN_DROP_FIXED_SUPER_TABLE = 4 CAN_ADD_DATA = 5 CAN_READ_DATA = 6 def __init__(self): self._info = self.getInfo() def __str__(self): return self._stateNames[self._info[self.STATE_VAL_IDX] + 1] # -1 hack to accomodate the STATE_INVALID case def getInfo(self): raise RuntimeError("Must be overriden by child classes") def equals(self, other): if isinstance(other, int): return self.getValIndex() == other elif isinstance(other, AnyState): return self.getValIndex() == other.getValIndex() else: raise RuntimeError("Unexpected comparison, type = {}".format(type(other))) def verifyTasksToState(self, tasks, newState): raise RuntimeError("Must be overriden by child classes") def getValIndex(self): return self._info[self.STATE_VAL_IDX] def getValue(self): return self._info[self.STATE_VAL_IDX] def canCreateDb(self): return self._info[self.CAN_CREATE_DB] def canDropDb(self): return self._info[self.CAN_DROP_DB] def canCreateFixedSuperTable(self): return self._info[self.CAN_CREATE_FIXED_SUPER_TABLE] def canDropFixedSuperTable(self): return self._info[self.CAN_DROP_FIXED_SUPER_TABLE] def canAddData(self): return self._info[self.CAN_ADD_DATA] def canReadData(self): return self._info[self.CAN_READ_DATA] def assertAtMostOneSuccess(self, tasks, cls): sCnt = 0 for task in tasks : if not isinstance(task, cls): continue if task.isSuccess(): # task.logDebug("Task success found") sCnt += 1 if ( sCnt >= 2 ): raise RuntimeError("Unexpected more than 1 success with task: {}".format(cls)) def assertIfExistThenSuccess(self, tasks, cls): sCnt = 0 exists = False for task in tasks : if not isinstance(task, cls): continue exists = True # we have a valid instance if task.isSuccess(): sCnt += 1 if ( exists and sCnt <= 0 ): raise RuntimeError("Unexpected zero success for task: {}".format(cls)) def assertNoTask(self, tasks, cls): for task in tasks : if isinstance(task, cls): raise CrashGenError("This task: {}, is not expected to be present, given the success/failure of others".format(cls.__name__)) def assertNoSuccess(self, tasks, cls): for task in tasks : if isinstance(task, cls): if task.isSuccess(): raise RuntimeError("Unexpected successful task: {}".format(cls)) def hasSuccess(self, tasks, cls): for task in tasks : if not isinstance(task, cls): continue if task.isSuccess(): return True return False def hasTask(self, tasks, cls): for task in tasks : if isinstance(task, cls): return True return False class StateInvalid(AnyState): def getInfo(self): return [ self.STATE_INVALID, False, False, # can create/drop Db False, False, # can create/drop fixed table False, False, # can insert/read data with fixed table ] # def verifyTasksToState(self, tasks, newState): class StateEmpty(AnyState): def getInfo(self): return [ self.STATE_EMPTY, True, False, # can create/drop Db False, False, # can create/drop fixed table False, False, # can insert/read data with fixed table ] def verifyTasksToState(self, tasks, newState): if ( self.hasSuccess(tasks, TaskCreateDb) ): # at EMPTY, if there's succes in creating DB if ( not self.hasTask(tasks, TaskDropDb) ) : # and no drop_db tasks self.assertAtMostOneSuccess(tasks, TaskCreateDb) # we must have at most one. TODO: compare numbers class StateDbOnly(AnyState): def getInfo(self): return [ self.STATE_DB_ONLY, False, True, True, False, False, False, ] def verifyTasksToState(self, tasks, newState): if ( not self.hasTask(tasks, TaskCreateDb) ): self.assertAtMostOneSuccess(tasks, TaskDropDb) # only if we don't create any more self.assertIfExistThenSuccess(tasks, TaskDropDb) # self.assertAtMostOneSuccess(tasks, CreateFixedTableTask) # not true in massively parrallel cases # Nothing to be said about adding data task # if ( self.hasSuccess(tasks, DropDbTask) ): # dropped the DB # self.assertHasTask(tasks, DropDbTask) # implied by hasSuccess # self.assertAtMostOneSuccess(tasks, DropDbTask) # self._state = self.STATE_EMPTY if ( self.hasSuccess(tasks, TaskCreateSuperTable) ): # did not drop db, create table success # self.assertHasTask(tasks, CreateFixedTableTask) # tried to create table if ( not self.hasTask(tasks, TaskDropSuperTable) ): self.assertAtMostOneSuccess(tasks, TaskCreateSuperTable) # at most 1 attempt is successful, if we don't drop anything # self.assertNoTask(tasks, DropDbTask) # should have have tried # if ( not self.hasSuccess(tasks, AddFixedDataTask) ): # just created table, no data yet # # can't say there's add-data attempts, since they may all fail # self._state = self.STATE_TABLE_ONLY # else: # self._state = self.STATE_HAS_DATA # What about AddFixedData? # elif ( self.hasSuccess(tasks, AddFixedDataTask) ): # self._state = self.STATE_HAS_DATA # else: # no success in dropping db tasks, no success in create fixed table? read data should also fail # # raise RuntimeError("Unexpected no-success scenario") # We might just landed all failure tasks, # self._state = self.STATE_DB_ONLY # no change class StateSuperTableOnly(AnyState): def getInfo(self): return [ self.STATE_TABLE_ONLY, False, True, False, True, True, True, ] def verifyTasksToState(self, tasks, newState): if ( self.hasSuccess(tasks, TaskDropSuperTable) ): # we are able to drop the table self.assertAtMostOneSuccess(tasks, TaskDropSuperTable) # self._state = self.STATE_DB_ONLY # elif ( self.hasSuccess(tasks, AddFixedDataTask) ): # no success dropping the table, but added data # self.assertNoTask(tasks, DropFixedTableTask) # not true in massively parrallel cases # self._state = self.STATE_HAS_DATA # elif ( self.hasSuccess(tasks, ReadFixedDataTask) ): # no success in prev cases, but was able to read data # self.assertNoTask(tasks, DropFixedTableTask) # self.assertNoTask(tasks, AddFixedDataTask) # self._state = self.STATE_TABLE_ONLY # no change # else: # did not drop table, did not insert data, did not read successfully, that is impossible # raise RuntimeError("Unexpected no-success scenarios") # TODO: need to revamp!! class StateHasData(AnyState): def getInfo(self): return [ self.STATE_HAS_DATA, False, True, False, True, True, True, ] def verifyTasksToState(self, tasks, newState): if ( newState.equals(AnyState.STATE_EMPTY) ): self.hasSuccess(tasks, TaskDropDb) if ( not self.hasTask(tasks, TaskCreateDb) ) : self.assertAtMostOneSuccess(tasks, TaskDropDb) # TODO: dicy elif ( newState.equals(AnyState.STATE_DB_ONLY) ): # in DB only if ( not self.hasTask(tasks, TaskCreateDb)): # without a create_db task self.assertNoTask(tasks, TaskDropDb) # we must have drop_db task self.hasSuccess(tasks, TaskDropSuperTable) # self.assertAtMostOneSuccess(tasks, DropFixedSuperTableTask) # TODO: dicy elif ( newState.equals(AnyState.STATE_TABLE_ONLY) ): # data deleted self.assertNoTask(tasks, TaskDropDb) self.assertNoTask(tasks, TaskDropSuperTable) self.assertNoTask(tasks, TaskAddData) # self.hasSuccess(tasks, DeleteDataTasks) else: # should be STATE_HAS_DATA if (not self.hasTask(tasks, TaskCreateDb) ): # only if we didn't create one self.assertNoTask(tasks, TaskDropDb) # we shouldn't have dropped it if (not self.hasTask(tasks, TaskCreateSuperTable)) : # if we didn't create the table self.assertNoTask(tasks, TaskDropSuperTable) # we should not have a task that drops it # self.assertIfExistThenSuccess(tasks, ReadFixedDataTask) class StateMechine : def __init__(self, dbConn): self._dbConn = dbConn self._curState = self._findCurrentState() # starting state self._stateWeights = [1,3,5,15] # transitition target probabilities, indexed with value of STATE_EMPTY, STATE_DB_ONLY, etc. def getCurrentState(self): return self._curState # May be slow, use cautionsly... def getTaskTypes(self): # those that can run (directly/indirectly) from the current state allTaskClasses = StateTransitionTask.__subclasses__() # all state transition tasks firstTaskTypes = [] for tc in allTaskClasses: # t = tc(self) # create task object if tc.canBeginFrom(self._curState): firstTaskTypes.append(tc) # now we have all the tasks that can begin directly from the current state, let's figure out the INDIRECT ones taskTypes = firstTaskTypes.copy() # have to have these for task1 in firstTaskTypes: # each task type gathered so far endState = task1.getEndState() # figure the end state if endState == None: # does not change end state continue # no use, do nothing for tc in allTaskClasses: # what task can further begin from there? if tc.canBeginFrom(endState) and (tc not in firstTaskTypes): taskTypes.append(tc) # gather it if len(taskTypes) <= 0: raise RuntimeError("No suitable task types found for state: {}".format(self._curState)) logger.debug("[OPS] Tasks found for state {}: {}".format(self._curState, taskTypes)) return taskTypes def _findCurrentState(self): dbc = self._dbConn ts = time.time() # we use this to debug how fast/slow it is to do the various queries to find the current DB state if dbc.query("show databases") == 0 : # no database?! # logger.debug("Found EMPTY state") logger.debug("[STT] empty database found, between {} and {}".format(ts, time.time())) return StateEmpty() dbc.execute("use db") # did not do this when openning connection if dbc.query("show tables") == 0 : # no tables # logger.debug("Found DB ONLY state") logger.debug("[STT] DB_ONLY found, between {} and {}".format(ts, time.time())) return StateDbOnly() if dbc.query("SELECT * FROM db.{}".format(DbManager.getFixedSuperTableName()) ) == 0 : # no regular tables # logger.debug("Found TABLE_ONLY state") logger.debug("[STT] SUPER_TABLE_ONLY found, between {} and {}".format(ts, time.time())) return StateSuperTableOnly() else: # has actual tables # logger.debug("Found HAS_DATA state") logger.debug("[STT] HAS_DATA found, between {} and {}".format(ts, time.time())) return StateHasData() def transition(self, tasks): if ( len(tasks) == 0 ): # before 1st step, or otherwise empty return # do nothing self._dbConn.execute("show dnodes") # this should show up in the server log, separating steps # Generic Checks, first based on the start state if self._curState.canCreateDb(): self._curState.assertIfExistThenSuccess(tasks, TaskCreateDb) # self.assertAtMostOneSuccess(tasks, CreateDbTask) # not really, in case of multiple creation and drops if self._curState.canDropDb(): self._curState.assertIfExistThenSuccess(tasks, TaskDropDb) # self.assertAtMostOneSuccess(tasks, DropDbTask) # not really in case of drop-create-drop # if self._state.canCreateFixedTable(): # self.assertIfExistThenSuccess(tasks, CreateFixedTableTask) # Not true, DB may be dropped # self.assertAtMostOneSuccess(tasks, CreateFixedTableTask) # not really, in case of create-drop-create # if self._state.canDropFixedTable(): # self.assertIfExistThenSuccess(tasks, DropFixedTableTask) # Not True, the whole DB may be dropped # self.assertAtMostOneSuccess(tasks, DropFixedTableTask) # not really in case of drop-create-drop # if self._state.canAddData(): # self.assertIfExistThenSuccess(tasks, AddFixedDataTask) # not true actually # if self._state.canReadData(): # Nothing for sure newState = self._findCurrentState() logger.debug("[STT] New DB state determined: {}".format(newState)) self._curState.verifyTasksToState(tasks, newState) # can old state move to new state through the tasks? self._curState = newState def pickTaskType(self): taskTypes = self.getTaskTypes() # all the task types we can choose from at curent state weights = [] for tt in taskTypes: endState = tt.getEndState() if endState != None : weights.append(self._stateWeights[endState.getValIndex()]) # TODO: change to a method else: weights.append(10) # read data task, default to 10: TODO: change to a constant i = self._weighted_choice_sub(weights) # logger.debug(" (weighted random:{}/{}) ".format(i, len(taskTypes))) return taskTypes[i] def _weighted_choice_sub(self, weights): # ref: https://eli.thegreenplace.net/2010/01/22/weighted-random-generation-in-python/ rnd = random.random() * sum(weights) # TODO: use our dice to ensure it being determinstic? for i, w in enumerate(weights): rnd -= w if rnd < 0: return i # Manager of the Database Data/Connection class DbManager(): def __init__(self, resetDb = True): self.tableNumQueue = LinearQueue() self._lastTick = self.setupLastTick() # datetime.datetime(2019, 1, 1) # initial date time tick self._lastInt = 0 # next one is initial integer self._lock = threading.RLock() # self.openDbServerConnection() self._dbConn = DbConn() try: self._dbConn.open() # may throw taos.error.ProgrammingError: disconnected except taos.error.ProgrammingError as err: # print("Error type: {}, msg: {}, value: {}".format(type(err), err.msg, err)) if ( err.msg == 'client disconnected' ): # cannot open DB connection print("Cannot establish DB connection, please re-run script without parameter, and follow the instructions.") sys.exit() else: raise except: print("[=] Unexpected exception") raise if resetDb : self._dbConn.resetDb() # drop and recreate DB self._stateMachine = StateMechine(self._dbConn) # Do this after dbConn is in proper shape def getDbConn(self): return self._dbConn def getStateMachine(self): return self._stateMachine # def getState(self): # return self._stateMachine.getCurrentState() # We aim to create a starting time tick, such that, whenever we run our test here once # We should be able to safely create 100,000 records, which will not have any repeated time stamp # when we re-run the test in 3 minutes (180 seconds), basically we should expand time duration # by a factor of 500. # TODO: what if it goes beyond 10 years into the future # TODO: fix the error as result of above: "tsdb timestamp is out of range" def setupLastTick(self): t1 = datetime.datetime(2020, 6, 1) t2 = datetime.datetime.now() elSec = int(t2.timestamp() - t1.timestamp()) # maybe a very large number, takes 69 years to exceed Python int range elSec2 = ( elSec % (8 * 12 * 30 * 24 * 60 * 60 / 500 ) ) * 500 # a number representing seconds within 10 years # print("elSec = {}".format(elSec)) t3 = datetime.datetime(2012, 1, 1) # default "keep" is 10 years t4 = datetime.datetime.fromtimestamp( t3.timestamp() + elSec2) # see explanation above logger.info("Setting up TICKS to start from: {}".format(t4)) return t4 def pickAndAllocateTable(self): # pick any table, and "use" it return self.tableNumQueue.pickAndAllocate() def addTable(self): with self._lock: tIndex = self.tableNumQueue.push() return tIndex @classmethod def getFixedSuperTableName(cls): return "fs_table" def releaseTable(self, i): # return the table back, so others can use it self.tableNumQueue.release(i) def getNextTick(self): with self._lock: # prevent duplicate tick self._lastTick += datetime.timedelta(0, 1) # add one second to it return self._lastTick def getNextInt(self): with self._lock: self._lastInt += 1 return self._lastInt def getNextBinary(self): return "Beijing_Shanghai_Los_Angeles_New_York_San_Francisco_Chicago_Beijing_Shanghai_Los_Angeles_New_York_San_Francisco_Chicago_{}".format(self.getNextInt()) def getNextFloat(self): return 0.9 + self.getNextInt() def getTableNameToDelete(self): tblNum = self.tableNumQueue.pop() # TODO: race condition! if ( not tblNum ): # maybe false return False return "table_{}".format(tblNum) def cleanUp(self): self._dbConn.close() class TaskExecutor(): def __init__(self, curStep): self._curStep = curStep def getCurStep(self): return self._curStep def execute(self, task: Task, wt: WorkerThread): # execute a task on a thread task.execute(wt) # def logInfo(self, msg): # logger.info(" T[{}.x]: ".format(self._curStep) + msg) # def logDebug(self, msg): # logger.debug(" T[{}.x]: ".format(self._curStep) + msg) class Task(): taskSn = 100 @classmethod def allocTaskNum(cls): Task.taskSn += 1 # IMPORTANT: cannot use cls.taskSn, since each sub class will have a copy # logger.debug("Allocating taskSN: {}".format(Task.taskSn)) return Task.taskSn def __init__(self, dbManager: DbManager, execStats: ExecutionStats): self._dbManager = dbManager self._workerThread = None self._err = None self._curStep = None self._numRows = None # Number of rows affected # Assign an incremental task serial number self._taskNum = self.allocTaskNum() # logger.debug("Creating new task {}...".format(self._taskNum)) self._execStats = execStats def isSuccess(self): return self._err == None def clone(self): # TODO: why do we need this again? newTask = self.__class__(self._dbManager, self._execStats) return newTask def logDebug(self, msg): self._workerThread.logDebug("Step[{}.{}] {}".format(self._curStep, self._taskNum, msg)) def logInfo(self, msg): self._workerThread.logInfo("Step[{}.{}] {}".format(self._curStep, self._taskNum, msg)) def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): raise RuntimeError("To be implemeted by child classes, class name: {}".format(self.__class__.__name__)) def execute(self, wt: WorkerThread): wt.verifyThreadSelf() self._workerThread = wt # type: ignore te = wt.getTaskExecutor() self._curStep = te.getCurStep() self.logDebug("[-] executing task {}...".format(self.__class__.__name__)) self._err = None self._execStats.beginTaskType(self.__class__.__name__) # mark beginning try: self._executeInternal(te, wt) # TODO: no return value? except taos.error.ProgrammingError as err: self.logDebug("[=] Taos library exception: errno={:X}, msg: {}".format(err.errno, err)) self._err = err except: self.logDebug("[=] Unexpected exception") raise self._execStats.endTaskType(self.__class__.__name__, self.isSuccess()) self.logDebug("[X] task execution completed, {}, status: {}".format(self.__class__.__name__, "Success" if self.isSuccess() else "Failure")) self._execStats.incExecCount(self.__class__.__name__, self.isSuccess()) # TODO: merge with above. def execSql(self, sql): return self._dbManager.execute(sql) class ExecutionStats: def __init__(self): self._execTimes: Dict[str, [int, int]] = {} # total/success times for a task self._tasksInProgress = 0 self._lock = threading.Lock() self._firstTaskStartTime = None self._execStartTime = None self._elapsedTime = 0.0 # total elapsed time self._accRunTime = 0.0 # accumulated run time self._failed = False self._failureReason = None def startExec(self): self._execStartTime = time.time() def endExec(self): self._elapsedTime = time.time() - self._execStartTime def incExecCount(self, klassName, isSuccess): # TODO: add a lock here if klassName not in self._execTimes: self._execTimes[klassName] = [0, 0] t = self._execTimes[klassName] # tuple for the data t[0] += 1 # index 0 has the "total" execution times if isSuccess: t[1] += 1 # index 1 has the "success" execution times def beginTaskType(self, klassName): with self._lock: if self._tasksInProgress == 0 : # starting a new round self._firstTaskStartTime = time.time() # I am now the first task self._tasksInProgress += 1 def endTaskType(self, klassName, isSuccess): with self._lock: self._tasksInProgress -= 1 if self._tasksInProgress == 0 : # all tasks have stopped self._accRunTime += (time.time() - self._firstTaskStartTime) self._firstTaskStartTime = None def registerFailure(self, reason): self._failed = True self._failureReason = reason def logStats(self): logger.info("----------------------------------------------------------------------") logger.info("| Crash_Gen test {}, with the following stats:". format("FAILED (reason: {})".format(self._failureReason) if self._failed else "SUCCEEDED")) logger.info("| Task Execution Times (success/total):") execTimesAny = 0 for k, n in self._execTimes.items(): execTimesAny += n[0] logger.info("| {0:<24}: {1}/{2}".format(k,n[1],n[0])) logger.info("| Total Tasks Executed (success or not): {} ".format(execTimesAny)) logger.info("| Total Tasks In Progress at End: {}".format(self._tasksInProgress)) logger.info("| Total Task Busy Time (elapsed time when any task is in progress): {:.3f} seconds".format(self._accRunTime)) logger.info("| Average Per-Task Execution Time: {:.3f} seconds".format(self._accRunTime/execTimesAny)) logger.info("| Total Elapsed Time (from wall clock): {:.3f} seconds".format(self._elapsedTime)) logger.info("----------------------------------------------------------------------") class StateTransitionTask(Task): @classmethod def getInfo(cls): # each sub class should supply their own information raise RuntimeError("Overriding method expected") _endState = None @classmethod def getEndState(cls): # TODO: optimize by calling it fewer times raise RuntimeError("Overriding method expected") # @classmethod # def getBeginStates(cls): # return cls.getInfo()[0] # @classmethod # def getEndState(cls): # returning the class name # return cls.getInfo()[0] @classmethod def canBeginFrom(cls, state: AnyState): # return state.getValue() in cls.getBeginStates() raise RuntimeError("must be overriden") def execute(self, wt: WorkerThread): super().execute(wt) class TaskCreateDb(StateTransitionTask): @classmethod def getEndState(cls): return StateDbOnly() @classmethod def canBeginFrom(cls, state: AnyState): return state.canCreateDb() def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): wt.execSql("create database db") class TaskDropDb(StateTransitionTask): @classmethod def getEndState(cls): return StateEmpty() @classmethod def canBeginFrom(cls, state: AnyState): return state.canDropDb() def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): wt.execSql("drop database db") logger.debug("[OPS] database dropped at {}".format(time.time())) class TaskCreateSuperTable(StateTransitionTask): @classmethod def getEndState(cls): return StateSuperTableOnly() @classmethod def canBeginFrom(cls, state: AnyState): return state.canCreateFixedSuperTable() def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): tblName = self._dbManager.getFixedSuperTableName() wt.execSql("create table db.{} (ts timestamp, speed int) tags (b binary(200), f float) ".format(tblName)) # No need to create the regular tables, INSERT will do that automatically class TaskReadData(StateTransitionTask): @classmethod def getEndState(cls): return None # meaning doesn't affect state @classmethod def canBeginFrom(cls, state: AnyState): return state.canReadData() def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): sTbName = self._dbManager.getFixedSuperTableName() dbc = wt.getDbConn() dbc.query("select TBNAME from db.{}".format(sTbName)) # TODO: analyze result set later if random.randrange(5) == 0 : # 1 in 5 chance, simulate a broken connection. TODO: break connection in all situations dbc.close() dbc.open() else: rTables = dbc.getQueryResult() # print("rTables[0] = {}, type = {}".format(rTables[0], type(rTables[0]))) for rTbName in rTables : # regular tables dbc.query("select * from db.{}".format(rTbName[0])) # TODO: check success failure # tdSql.query(" cars where tbname in ('carzero', 'carone')") class TaskDropSuperTable(StateTransitionTask): @classmethod def getEndState(cls): return StateDbOnly() @classmethod def canBeginFrom(cls, state: AnyState): return state.canDropFixedSuperTable() def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): tblName = self._dbManager.getFixedSuperTableName() wt.execSql("drop table db.{}".format(tblName)) class TaskAlterTags(StateTransitionTask): @classmethod def getEndState(cls): return None # meaning doesn't affect state @classmethod def canBeginFrom(cls, state: AnyState): return state.canDropFixedSuperTable() # if we can drop it, we can alter tags def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): tblName = self._dbManager.getFixedSuperTableName() dice = Dice.throw(4) if dice == 0 : wt.execSql("alter table db.{} add tag extraTag int".format(tblName)) elif dice == 1 : wt.execSql("alter table db.{} drop tag extraTag".format(tblName)) elif dice == 2 : wt.execSql("alter table db.{} drop tag newTag".format(tblName)) else: # dice == 3 wt.execSql("alter table db.{} change tag extraTag newTag".format(tblName)) class TaskAddData(StateTransitionTask): activeTable : Set[int] = set() # Track which table is being actively worked on LARGE_NUMBER_OF_TABLES = 35 SMALL_NUMBER_OF_TABLES = 3 LARGE_NUMBER_OF_RECORDS = 50 SMALL_NUMBER_OF_RECORDS = 3 # We use these two files to record operations to DB, useful for power-off tests fAddLogReady = None fAddLogDone = None @classmethod def prepToRecordOps(cls): if gConfig.record_ops : if ( cls.fAddLogReady == None ): logger.info("Recording in a file operations to be performed...") cls.fAddLogReady = open("add_log_ready.txt", "w") if ( cls.fAddLogDone == None ): logger.info("Recording in a file operations completed...") cls.fAddLogDone = open("add_log_done.txt", "w") @classmethod def getEndState(cls): return StateHasData() @classmethod def canBeginFrom(cls, state: AnyState): return state.canAddData() def _executeInternal(self, te: TaskExecutor, wt: WorkerThread): ds = self._dbManager wt.execSql("use db") # TODO: seems to be an INSERT bug to require this tblSeq = list(range(self.LARGE_NUMBER_OF_TABLES if gConfig.larger_data else self.SMALL_NUMBER_OF_TABLES)) random.shuffle(tblSeq) for i in tblSeq: if ( i in self.activeTable ): # wow already active # logger.info("Concurrent data insertion into table: {}".format(i)) # print("ct({})".format(i), end="", flush=True) # Concurrent insertion into table print("x", end="", flush=True) else: self.activeTable.add(i) # marking it active # No need to shuffle data sequence, unless later we decide to do non-increment insertion for j in range(self.LARGE_NUMBER_OF_RECORDS if gConfig.larger_data else self.SMALL_NUMBER_OF_RECORDS) : # number of records per table nextInt = ds.getNextInt() regTableName = "db.reg_table_{}".format(i) if gConfig.record_ops: self.prepToRecordOps() self.fAddLogReady.write("Ready to write {} to {}\n".format(nextInt, regTableName)) self.fAddLogReady.flush() os.fsync(self.fAddLogReady) sql = "insert into {} using {} tags ('{}', {}) values ('{}', {});".format( regTableName, ds.getFixedSuperTableName(), ds.getNextBinary(), ds.getNextFloat(), ds.getNextTick(), nextInt) wt.execSql(sql) if gConfig.record_ops: self.fAddLogDone.write("Wrote {} to {}\n".format(nextInt, regTableName)) self.fAddLogDone.flush() os.fsync(self.fAddLogDone) self.activeTable.discard(i) # not raising an error, unlike remove # Deterministic random number generator class Dice(): seeded = False # static, uninitialized @classmethod def seed(cls, s): # static if (cls.seeded): raise RuntimeError("Cannot seed the random generator more than once") cls.verifyRNG() random.seed(s) cls.seeded = True # TODO: protect against multi-threading @classmethod def verifyRNG(cls): # Verify that the RNG is determinstic random.seed(0) x1 = random.randrange(0, 1000) x2 = random.randrange(0, 1000) x3 = random.randrange(0, 1000) if ( x1 != 864 or x2!=394 or x3!=776 ): raise RuntimeError("System RNG is not deterministic") @classmethod def throw(cls, stop): # get 0 to stop-1 return cls.throwRange(0, stop) @classmethod def throwRange(cls, start, stop): # up to stop-1 if ( not cls.seeded ): raise RuntimeError("Cannot throw dice before seeding it") return random.randrange(start, stop) # Anyone needing to carry out work should simply come here # class WorkDispatcher(): # def __init__(self, dbState): # # self.totalNumMethods = 2 # self.tasks = [ # # CreateTableTask(dbState), # Obsolete # # DropTableTask(dbState), # # AddDataTask(dbState), # ] # def throwDice(self): # max = len(self.tasks) - 1 # dRes = random.randint(0, max) # # logger.debug("Threw the dice in range [{},{}], and got: {}".format(0,max,dRes)) # return dRes # def pickTask(self): # dice = self.throwDice() # return self.tasks[dice] # def doWork(self, workerThread): # task = self.pickTask() # task.execute(workerThread) class LoggingFilter(logging.Filter): def filter(self, record: logging.LogRecord): if ( record.levelno >= logging.INFO ) : return True # info or above always log msg = record.msg # print("type = {}, value={}".format(type(msg), msg)) # sys.exit() # Commenting out below to adjust... # if msg.startswith("[TRD]"): # return False return True class MainExec: @classmethod def runClient(cls): # resetDb = False # DEBUG only # dbState = DbState(resetDb) # DBEUG only! dbManager = DbManager() # Regular function Dice.seed(0) # initial seeding of dice thPool = ThreadPool(gConfig.num_threads, gConfig.max_steps) tc = ThreadCoordinator(thPool, dbManager) tc.run() tc.logStats() dbManager.cleanUp() @classmethod def runService(cls): print("Running service...") @classmethod def runTemp(cls): # for debugging purposes # # Hack to exercise reading from disk, imcreasing coverage. TODO: fix # dbc = dbState.getDbConn() # sTbName = dbState.getFixedSuperTableName() # dbc.execute("create database if not exists db") # if not dbState.getState().equals(StateEmpty()): # dbc.execute("use db") # rTables = None # try: # the super table may not exist # sql = "select TBNAME from db.{}".format(sTbName) # logger.info("Finding out tables in super table: {}".format(sql)) # dbc.query(sql) # TODO: analyze result set later # logger.info("Fetching result") # rTables = dbc.getQueryResult() # logger.info("Result: {}".format(rTables)) # except taos.error.ProgrammingError as err: # logger.info("Initial Super table OPS error: {}".format(err)) # # sys.exit() # if ( not rTables == None): # # print("rTables[0] = {}, type = {}".format(rTables[0], type(rTables[0]))) # try: # for rTbName in rTables : # regular tables # ds = dbState # logger.info("Inserting into table: {}".format(rTbName[0])) # sql = "insert into db.{} values ('{}', {});".format( # rTbName[0], # ds.getNextTick(), ds.getNextInt()) # dbc.execute(sql) # for rTbName in rTables : # regular tables # dbc.query("select * from db.{}".format(rTbName[0])) # TODO: check success failure # logger.info("Initial READING operation is successful") # except taos.error.ProgrammingError as err: # logger.info("Initial WRITE/READ error: {}".format(err)) # Sandbox testing code # dbc = dbState.getDbConn() # while True: # rows = dbc.query("show databases") # print("Rows: {}, time={}".format(rows, time.time())) return def main(): # Super cool Python argument library: https://docs.python.org/3/library/argparse.html parser = argparse.ArgumentParser( formatter_class=argparse.RawDescriptionHelpFormatter, description=textwrap.dedent('''\ TDengine Auto Crash Generator (PLEASE NOTICE the Prerequisites Below) --------------------------------------------------------------------- 1. You build TDengine in the top level ./build directory, as described in offical docs 2. You run the server there before this script: ./build/bin/taosd -c test/cfg ''')) parser.add_argument('-d', '--debug', action='store_true', help='Turn on DEBUG mode for more logging (default: false)') parser.add_argument('-e', '--run-tdengine', action='store_true', help='Run TDengine service in foreground (default: false)') parser.add_argument('-l', '--larger-data', action='store_true', help='Write larger amount of data during write operations (default: false)') parser.add_argument('-p', '--per-thread-db-connection', action='store_false', help='Use a single shared db connection (default: false)') parser.add_argument('-r', '--record-ops', action='store_true', help='Use a pair of always-fsynced fils to record operations performing + performed, for power-off tests (default: false)') parser.add_argument('-s', '--max-steps', action='store', default=1000, type=int, help='Maximum number of steps to run (default: 100)') parser.add_argument('-t', '--num-threads', action='store', default=5, type=int, help='Number of threads to run (default: 10)') global gConfig gConfig = parser.parse_args() # if len(sys.argv) == 1: # parser.print_help() # sys.exit() global logger logger = logging.getLogger('CrashGen') logger.addFilter(LoggingFilter()) if ( gConfig.debug ): logger.setLevel(logging.DEBUG) # default seems to be INFO else: logger.setLevel(logging.INFO) ch = logging.StreamHandler() logger.addHandler(ch) if gConfig.run_tdengine : # run server MainExec.runService() else : MainExec.runClient() # logger.info("Crash_Gen execution finished") if __name__ == "__main__": main()