Merge branch 'develop' into hotfix/sangshuduo/TD-3143-taosdemo-windows

documentation20/cn/10.cluster/docs.md

@@ -225,7 +225,13 @@ SHOW MNODES;
 
 ## <a class="anchor" id="arbitrator"></a>Arbitrator的使用
 
-如果副本数为偶数，当一个vnode group里一半vnode不工作时，是无法从中选出master的。同理，一半mnode不工作时，是无法选出mnode的master的，因为存在“split brain”问题。为解决这个问题，TDengine引入了Arbitrator的概念。Arbitrator模拟一个vnode或mnode在工作，但只简单的负责网络连接，不处理任何数据插入或访问。只要包含Arbitrator在内，超过半数的vnode或mnode工作，那么该vnode group或mnode组就可以正常的提供数据插入或查询服务。比如对于副本数为2的情形，如果一个节点A离线，但另外一个节点B正常，而且能连接到Arbitrator，那么节点B就能正常工作。
+如果副本数为偶数，当一个 vnode group 里一半 vnode 不工作时，是无法从中选出 master 的。同理，一半 mnode 不工作时，是无法选出 mnode 的 master 的，因为存在“split brain”问题。为解决这个问题，TDengine 引入了 Arbitrator 的概念。Arbitrator 模拟一个 vnode 或 mnode 在工作，但只简单的负责网络连接，不处理任何数据插入或访问。只要包含 Arbitrator 在内，超过半数的 vnode 或 mnode 工作，那么该 vnode group 或 mnode 组就可以正常的提供数据插入或查询服务。比如对于副本数为 2 的情形，如果一个节点 A 离线，但另外一个节点 B 正常，而且能连接到 Arbitrator，那么节点 B 就能正常工作。
 
-TDengine提供一个执行程序，名为 tarbitrator，找任何一台Linux服务器运行它即可。请点击[安装包下载](https://www.taosdata.com/cn/all-downloads/)，在TDengine Arbitrator Linux一节中，选择适合的版本下载并安装。该程序对系统资源几乎没有要求，只需要保证有网络连接即可。该应用的命令行参数`-p`可以指定其对外服务的端口号，缺省是6042。配置每个taosd实例时，可以在配置文件taos.cfg里将参数arbitrator设置为Arbitrator的End Point。如果该参数配置了，当副本数为偶数时，系统将自动连接配置的Arbitrator。如果副本数为奇数，即使配置了Arbitrator，系统也不会去建立连接。
+总之，在目前版本下，TDengine 建议在双副本环境要配置 Arbitrator，以提升系统的可用性。
+
+Arbitrator 的执行程序名为 tarbitrator。该程序对系统资源几乎没有要求，只需要保证有网络连接，找任何一台 Linux 服务器运行它即可。以下简要描述安装配置的步骤：
+1. 请点击 [安装包下载](https://www.taosdata.com/cn/all-downloads/)，在 TDengine Arbitrator Linux 一节中，选择合适的版本下载并安装。
+2. 该应用的命令行参数 `-p` 可以指定其对外服务的端口号，缺省是 6042。
+3. 修改每个 taosd 实例的配置文件，在 taos.cfg 里将参数 arbitrator 设置为 tarbitrator 程序所对应的 End Point。（如果该参数配置了，当副本数为偶数时，系统将自动连接配置的 Arbitrator。如果副本数为奇数，即使配置了 Arbitrator，系统也不会去建立连接。）
+4. 在配置文件中配置了的 Arbitrator，会出现在 `SHOW DNODES;` 指令的返回结果中，对应的 role 列的值会是“arb”。
 

src/dnode/src/dnodeVnodes.c

@@ -198,6 +198,14 @@ void dnodeCleanupVnodes() {
 static void dnodeProcessStatusRsp(SRpcMsg *pMsg) {
   if (pMsg->code != TSDB_CODE_SUCCESS) {
     dError("status rsp is received, error:%s", tstrerror(pMsg->code));
+    if (pMsg->code == TSDB_CODE_MND_DNODE_NOT_EXIST) {
+      char clusterId[TSDB_CLUSTER_ID_LEN];
+      dnodeGetClusterId(clusterId);
+      if (clusterId[0] != '\0') {
+	dError("exit zombie dropped dnode");
+	exit(EXIT_FAILURE);
+      }
+    }
     taosTmrReset(dnodeSendStatusMsg, tsStatusInterval * 1000, NULL, tsDnodeTmr, &tsStatusTimer);
     return;
   }

tests/pytest/crash_gen/crash_gen_main.py

@@ -35,16 +35,19 @@ import os
 import signal
 import traceback
 import resource
-from guppy import hpy
+# from guppy import hpy
 import gc
 
 from crash_gen.service_manager import ServiceManager, TdeInstance
 from crash_gen.misc import Logging, Status, CrashGenError, Dice, Helper, Progress
 from crash_gen.db import DbConn, MyTDSql, DbConnNative, DbManager
+import crash_gen.settings 
 
 import taos
 import requests
 
+crash_gen.settings.init()
+
 # Require Python 3
 if sys.version_info[0] < 3:
     raise Exception("Must be using Python 3")
@@ -259,6 +262,7 @@ class ThreadCoordinator:
         self._execStats = ExecutionStats()
         self._runStatus = Status.STATUS_RUNNING
         self._initDbs()
+        self._stepStartTime = None  # Track how long it takes to execute each step
 
     def getTaskExecutor(self):
         return self._te
@@ -394,6 +398,10 @@ class ThreadCoordinator:
             try:
                 self._syncAtBarrier() # For now just cross the barrier
                 Progress.emit(Progress.END_THREAD_STEP)
+                if self._stepStartTime :
+                    stepExecTime = time.time() - self._stepStartTime
+                    Progress.emitStr('{:.3f}s/{}'.format(stepExecTime, DbConnNative.totalRequests))
+                    DbConnNative.resetTotalRequests() # reset to zero
             except threading.BrokenBarrierError as err:
                 self._execStats.registerFailure("Aborted due to worker thread timeout")
                 Logging.error("\n")
@@ -433,6 +441,7 @@ class ThreadCoordinator:
 
             # Then we move on to the next step
             Progress.emit(Progress.BEGIN_THREAD_STEP)
+            self._stepStartTime = time.time()
             self._releaseAllWorkerThreads(transitionFailed)                    
 
         if hasAbortedTask or transitionFailed : # abnormal ending, workers waiting at "gate"
@@ -691,7 +700,7 @@ class AnyState:
     def canDropDb(self):
         # If user requests to run up to a number of DBs,
         # we'd then not do drop_db operations any more
-        if gConfig.max_dbs > 0 : 
+        if gConfig.max_dbs > 0 or gConfig.use_shadow_db : 
             return False
         return self._info[self.CAN_DROP_DB]
 
@@ -699,6 +708,8 @@ class AnyState:
         return self._info[self.CAN_CREATE_FIXED_SUPER_TABLE]
 
     def canDropFixedSuperTable(self):
+        if gConfig.use_shadow_db: # duplicate writes to shaddow DB, in which case let's disable dropping s-table
+            return False
         return self._info[self.CAN_DROP_FIXED_SUPER_TABLE]
 
     def canAddData(self):
@@ -1037,7 +1048,7 @@ class Database:
     _clsLock = threading.Lock() # class wide lock
     _lastInt = 101  # next one is initial integer
     _lastTick = 0
-    _lastLaggingTick = 0 # lagging tick, for unsequenced insersions
+    _lastLaggingTick = 0 # lagging tick, for out-of-sequence (oos) data insertions
 
     def __init__(self, dbNum: int, dbc: DbConn): # TODO: remove dbc
         self._dbNum = dbNum # we assign a number to databases, for our testing purpose
@@ -1093,21 +1104,24 @@ class Database:
         t3 = datetime.datetime(2012, 1, 1)  # default "keep" is 10 years
         t4 = datetime.datetime.fromtimestamp(
             t3.timestamp() + elSec2)  # see explanation above
-        Logging.debug("Setting up TICKS to start from: {}".format(t4))
+        Logging.info("Setting up TICKS to start from: {}".format(t4))
         return t4
 
     @classmethod
-    def getNextTick(cls):        
+    def getNextTick(cls):       
+        '''
+            Fetch a timestamp tick, with some random factor, may not be unique.
+        ''' 
         with cls._clsLock:  # prevent duplicate tick
             if cls._lastLaggingTick==0 or cls._lastTick==0 : # not initialized
                 # 10k at 1/20 chance, should be enough to avoid overlaps
                 tick = cls.setupLastTick()
                 cls._lastTick = tick
-                cls._lastLaggingTick = tick + datetime.timedelta(0, -10000)                 
+                cls._lastLaggingTick = tick + datetime.timedelta(0, -60*2)  # lagging behind 2 minutes, should catch up fast
                 # if : # should be quite a bit into the future
 
-            if Dice.throw(20) == 0:  # 1 in 20 chance, return lagging tick
-                cls._lastLaggingTick += datetime.timedelta(0, 1) # Go back in time 100 seconds
+            if gConfig.mix_oos_data and Dice.throw(20) == 0:  # if asked to do so, and 1 in 20 chance, return lagging tick
+                cls._lastLaggingTick += datetime.timedelta(0, 1) # pick the next sequence from the lagging tick sequence
                 return cls._lastLaggingTick 
             else:  # regular
                 # add one second to it
@@ -1334,7 +1348,8 @@ class Task():
             elif self._isErrAcceptable(errno2, err.__str__()):
                 self.logDebug("[=] Acceptable Taos library exception: errno=0x{:X}, msg: {}, SQL: {}".format(
                         errno2, err, wt.getDbConn().getLastSql()))
-                print("_", end="", flush=True)
+                # print("_", end="", flush=True)
+                Progress.emit(Progress.ACCEPTABLE_ERROR)
                 self._err = err
             else: # not an acceptable error
                 errMsg = "[=] Unexpected Taos library exception ({}): errno=0x{:X}, msg: {}, SQL: {}".format(
@@ -1563,8 +1578,11 @@ class TaskCreateDb(StateTransitionTask):
             # numReplica = Dice.throw(gConfig.max_replicas) + 1 # 1,2 ... N
             numReplica = gConfig.max_replicas # fixed, always
             repStr = "replica {}".format(numReplica)
-        self.execWtSql(wt, "create database {} {}"
-            .format(self._db.getName(), repStr) )
+        updatePostfix = "update 1" if gConfig.verify_data else "" # allow update only when "verify data" is active
+        dbName = self._db.getName()
+        self.execWtSql(wt, "create database {} {} {} ".format(dbName, repStr, updatePostfix ) )
+        if dbName == "db_0" and gConfig.use_shadow_db:
+            self.execWtSql(wt, "create database {} {} {} ".format("db_s", repStr, updatePostfix ) )
 
 class TaskDropDb(StateTransitionTask):
     @classmethod
@@ -1774,13 +1792,13 @@ class TdSuperTable:
                     ]) # TODO: add more from 'top'
 
             
-                if aggExpr not in ['stddev(speed)']: #TODO: STDDEV not valid for super tables?!
-                    sql = "select {} from {}.{}".format(aggExpr, self._dbName, self.getName())
-                    if Dice.throw(3) == 0: # 1 in X chance
-                        sql = sql + ' GROUP BY color'
-                        Progress.emit(Progress.QUERY_GROUP_BY)
-                        # Logging.info("Executing GROUP-BY query: " + sql)
-                    ret.append(SqlQuery(sql))
+                # if aggExpr not in ['stddev(speed)']: # STDDEV not valid for super tables?! (Done in TD-1049)
+                sql = "select {} from {}.{}".format(aggExpr, self._dbName, self.getName())
+                if Dice.throw(3) == 0: # 1 in X chance
+                    sql = sql + ' GROUP BY color'
+                    Progress.emit(Progress.QUERY_GROUP_BY)
+                    # Logging.info("Executing GROUP-BY query: " + sql)
+                ret.append(SqlQuery(sql))
 
         return ret        
 
@@ -1988,7 +2006,7 @@ class TaskAddData(StateTransitionTask):
         numRecords = self.LARGE_NUMBER_OF_RECORDS if gConfig.larger_data else self.SMALL_NUMBER_OF_RECORDS        
         fullTableName = db.getName() + '.' + regTableName
 
-        sql = "insert into {} values ".format(fullTableName)
+        sql = "INSERT INTO {} VALUES ".format(fullTableName)
         for j in range(numRecords):  # number of records per table
             nextInt = db.getNextInt()
             nextTick = db.getNextTick()
@@ -2016,12 +2034,24 @@ class TaskAddData(StateTransitionTask):
                 # print("_w" + str(nextInt % 100), end="", flush=True) # Trace what was written
 
             try:
-                sql = "insert into {} values ('{}', {}, '{}');".format( # removed: tags ('{}', {})
+                sql = "INSERT INTO {} VALUES ('{}', {}, '{}');".format( # removed: tags ('{}', {})
                     fullTableName,
                     # ds.getFixedSuperTableName(),
                     # ds.getNextBinary(), ds.getNextFloat(),
                     nextTick, nextInt, nextColor)
                 dbc.execute(sql)
+
+                # Quick hack, attach an update statement here. TODO: create an "update" task
+                if (not gConfig.use_shadow_db) and Dice.throw(5) == 0: # 1 in N chance, plus not using shaddow DB
+                    nextInt = db.getNextInt()
+                    nextColor = db.getNextColor()
+                    sql = "INSERt INTO {} VALUES ('{}', {}, '{}');".format( # "INSERt" means "update" here
+                    fullTableName,
+                    nextTick, nextInt, nextColor)
+                    # sql = "UPDATE {} set speed={}, color='{}' WHERE ts='{}'".format(
+                    #     fullTableName, db.getNextInt(), db.getNextColor(), nextTick)
+                    dbc.execute(sql)
+
             except: # Any exception at all
                 if gConfig.verify_data:
                     self.unlockTable(fullTableName)     
@@ -2070,7 +2100,8 @@ class TaskAddData(StateTransitionTask):
         random.shuffle(tblSeq) # now we have random sequence
         for i in tblSeq:
             if (i in self.activeTable):  # wow already active
-                print("x", end="", flush=True) # concurrent insertion
+                # print("x", end="", flush=True) # concurrent insertion
+                Progress.emit(Progress.CONCURRENT_INSERTION)
             else:
                 self.activeTable.add(i)  # marking it active
             
@@ -2373,6 +2404,11 @@ class MainExec:
             '--larger-data',
             action='store_true',
             help='Write larger amount of data during write operations (default: false)')
+        parser.add_argument(
+            '-m',
+            '--mix-oos-data',
+            action='store_false',
+            help='Mix out-of-sequence data into the test data stream (default: true)')
         parser.add_argument(
             '-n',
             '--dynamic-db-table-names',
@@ -2414,6 +2450,11 @@ class MainExec:
             '--verify-data',
             action='store_true',
             help='Verify data written in a number of places by reading back (default: false)')
+        parser.add_argument(
+            '-w',
+            '--use-shadow-db',
+            action='store_true',
+            help='Use a shaddow database to verify data integrity (default: false)')
         parser.add_argument(
             '-x',
             '--continue-on-exception',
@@ -2422,6 +2463,11 @@ class MainExec:
 
         global gConfig
         gConfig = parser.parse_args()
+        crash_gen.settings.gConfig = gConfig # TODO: fix this hack, consolidate this global var
+
+        # Sanity check for arguments
+        if gConfig.use_shadow_db and gConfig.max_dbs>1 :
+            raise CrashGenError("Cannot combine use-shadow-db with max-dbs of more than 1")
 
         Logging.clsInit(gConfig)
 

tests/pytest/crash_gen/db.py

@@ -18,6 +18,8 @@ import datetime
 import traceback
 # from .service_manager import TdeInstance
 
+import crash_gen.settings 
+
 class DbConn:
     TYPE_NATIVE = "native-c"
     TYPE_REST =   "rest-api"
@@ -244,7 +246,7 @@ class MyTDSql:
         self._conn.close() # TODO: very important, cursor close does NOT close DB connection!
         self._cursor.close()
 
-    def _execInternal(self, sql):
+    def _execInternal(self, sql):        
         startTime = time.time() 
         # Logging.debug("Executing SQL: " + sql)
         ret = self._cursor.execute(sql)
@@ -257,6 +259,27 @@ class MyTDSql:
                 cls.longestQuery = sql
                 cls.longestQueryTime = queryTime
                 cls.lqStartTime = startTime
+
+        # Now write to the shadow database
+        if crash_gen.settings.gConfig.use_shadow_db:
+            if sql[:11] == "INSERT INTO":
+                if sql[:16] == "INSERT INTO db_0":
+                    sql2 = "INSERT INTO db_s" + sql[16:]
+                    self._cursor.execute(sql2)
+                else:
+                    raise CrashGenError("Did not find db_0 in INSERT statement: {}".format(sql))
+            else: # not an insert statement
+                pass
+
+            if sql[:12] == "CREATE TABLE":
+                if sql[:17] == "CREATE TABLE db_0":
+                    sql2 = sql.replace('db_0', 'db_s')
+                    self._cursor.execute(sql2)
+                else:
+                    raise CrashGenError("Did not find db_0 in CREATE TABLE statement: {}".format(sql))
+            else: # not an insert statement
+                pass
+        
         return ret
 
     def query(self, sql):
@@ -302,12 +325,18 @@ class DbConnNative(DbConn):
     _lock = threading.Lock()
     # _connInfoDisplayed = False # TODO: find another way to display this
     totalConnections = 0 # Not private
+    totalRequests = 0 
 
     def __init__(self, dbTarget):
         super().__init__(dbTarget)
         self._type = self.TYPE_NATIVE
         self._conn = None
-        # self._cursor = None        
+        # self._cursor = None   
+        
+    @classmethod
+    def resetTotalRequests(cls):        
+        with cls._lock: # force single threading for opening DB connections. # TODO: whaaat??!!!            
+            cls.totalRequests = 0
 
     def openByType(self):  # Open connection
         # global gContainer
@@ -356,6 +385,8 @@ class DbConnNative(DbConn):
         Logging.debug("[SQL] Executing SQL: {}".format(sql))
         self._lastSql = sql
         nRows = self._tdSql.execute(sql)
+        cls = self.__class__
+        cls.totalRequests += 1
         Logging.debug(
             "[SQL] Execution Result, nRows = {}, SQL = {}".format(
                 nRows, sql))
@@ -369,6 +400,8 @@ class DbConnNative(DbConn):
         Logging.debug("[SQL] Executing SQL: {}".format(sql))
         self._lastSql = sql
         nRows = self._tdSql.query(sql)
+        cls = self.__class__
+        cls.totalRequests += 1
         Logging.debug(
             "[SQL] Query Result, nRows = {}, SQL = {}".format(
                 nRows, sql))

tests/pytest/crash_gen/misc.py

@@ -176,11 +176,13 @@ class Progress:
     SERVICE_START_NAP           = 7
     CREATE_TABLE_ATTEMPT        = 8
     QUERY_GROUP_BY              = 9
+    CONCURRENT_INSERTION        = 10
+    ACCEPTABLE_ERROR            = 11
 
     tokens = {
         STEP_BOUNDARY:      '.',
-        BEGIN_THREAD_STEP:  '[',
-        END_THREAD_STEP:    '] ',
+        BEGIN_THREAD_STEP:  ' [',
+        END_THREAD_STEP:    ']',
         SERVICE_HEART_BEAT: '.Y.',
         SERVICE_RECONNECT_START:    '<r.',
         SERVICE_RECONNECT_SUCCESS:  '.r>',
@@ -188,8 +190,14 @@ class Progress:
         SERVICE_START_NAP:           '_zz',
         CREATE_TABLE_ATTEMPT:       'c',
         QUERY_GROUP_BY:             'g',
+        CONCURRENT_INSERTION:       'x',
+        ACCEPTABLE_ERROR:           '_',
     }
 
     @classmethod
     def emit(cls, token):
         print(cls.tokens[token], end="", flush=True)
+
+    @classmethod
+    def emitStr(cls, str):
+        print('({})'.format(str), end="", flush=True)

tests/pytest/crash_gen/settings.py

+from __future__ import annotations
+import argparse
+
+gConfig:    argparse.Namespace 
+
+def init():
+    global gConfig
+    gConfig = []
\ No newline at end of file