homework-jianmu/tests/pytest/crash_gen/service_manager.py

738 lines
27 KiB
Python

import os
import io
import sys
import threading
import signal
import logging
import time
import subprocess
from typing import IO, List
try:
import psutil
except:
print("Psutil module needed, please install: sudo pip3 install psutil")
sys.exit(-1)
from queue import Queue, Empty
from .misc import Logging, Status, CrashGenError, Dice, Helper, Progress
from .db import DbConn, DbTarget
class TdeInstance():
"""
A class to capture the *static* information of a TDengine instance,
including the location of the various files/directories, and basica
configuration.
"""
@classmethod
def _getBuildPath(cls):
selfPath = os.path.dirname(os.path.realpath(__file__))
if ("community" in selfPath):
projPath = selfPath[:selfPath.find("communit")]
else:
projPath = selfPath[:selfPath.find("tests")]
buildPath = None
for root, dirs, files in os.walk(projPath):
if ("taosd" in files):
rootRealPath = os.path.dirname(os.path.realpath(root))
if ("packaging" not in rootRealPath):
buildPath = root[:len(root) - len("/build/bin")]
break
if buildPath == None:
raise RuntimeError("Failed to determine buildPath, selfPath={}, projPath={}"
.format(selfPath, projPath))
return buildPath
def __init__(self, subdir='test', tInstNum=0, port=6030, fepPort=6030):
self._buildDir = self._getBuildPath()
self._subdir = '/' + subdir # TODO: tolerate "/"
self._port = port # TODO: support different IP address too
self._fepPort = fepPort
self._tInstNum = tInstNum
self._smThread = ServiceManagerThread()
def getDbTarget(self):
return DbTarget(self.getCfgDir(), self.getHostAddr(), self._port)
def getPort(self):
return self._port
def __repr__(self):
return "[TdeInstance: {}, subdir={}]".format(
self._buildDir, Helper.getFriendlyPath(self._subdir))
def generateCfgFile(self):
# print("Logger = {}".format(logger))
# buildPath = self.getBuildPath()
# taosdPath = self._buildPath + "/build/bin/taosd"
cfgDir = self.getCfgDir()
cfgFile = cfgDir + "/taos.cfg" # TODO: inquire if this is fixed
if os.path.exists(cfgFile):
if os.path.isfile(cfgFile):
Logging.warning("Config file exists already, skip creation: {}".format(cfgFile))
return # cfg file already exists, nothing to do
else:
raise CrashGenError("Invalid config file: {}".format(cfgFile))
# Now that the cfg file doesn't exist
if os.path.exists(cfgDir):
if not os.path.isdir(cfgDir):
raise CrashGenError("Invalid config dir: {}".format(cfgDir))
# else: good path
else:
os.makedirs(cfgDir, exist_ok=True) # like "mkdir -p"
# Now we have a good cfg dir
cfgValues = {
'runDir': self.getRunDir(),
'ip': '127.0.0.1', # TODO: change to a network addressable ip
'port': self._port,
'fepPort': self._fepPort,
}
cfgTemplate = """
dataDir {runDir}/data
logDir {runDir}/log
charset UTF-8
firstEp {ip}:{fepPort}
fqdn {ip}
serverPort {port}
# was all 135 below
dDebugFlag 135
cDebugFlag 135
rpcDebugFlag 135
qDebugFlag 135
# httpDebugFlag 143
# asyncLog 0
# tables 10
maxtablesPerVnode 10
rpcMaxTime 101
# cache 2
keep 36500
# walLevel 2
walLevel 1
#
# maxConnections 100
"""
cfgContent = cfgTemplate.format_map(cfgValues)
f = open(cfgFile, "w")
f.write(cfgContent)
f.close()
def rotateLogs(self):
logPath = self.getLogDir()
# ref: https://stackoverflow.com/questions/1995373/deleting-all-files-in-a-directory-with-python/1995397
if os.path.exists(logPath):
logPathSaved = logPath + "_" + time.strftime('%Y-%m-%d-%H-%M-%S')
Logging.info("Saving old log files to: {}".format(logPathSaved))
os.rename(logPath, logPathSaved)
# os.mkdir(logPath) # recreate, no need actually, TDengine will auto-create with proper perms
def getExecFile(self): # .../taosd
return self._buildDir + "/build/bin/taosd"
def getRunDir(self): # TODO: rename to "root dir" ?!
return self._buildDir + self._subdir
def getCfgDir(self): # path, not file
return self.getRunDir() + "/cfg"
def getLogDir(self):
return self.getRunDir() + "/log"
def getHostAddr(self):
return "127.0.0.1"
def getServiceCmdLine(self): # to start the instance
return [self.getExecFile(), '-c', self.getCfgDir()] # used in subproce.Popen()
def _getDnodes(self, dbc):
dbc.query("show dnodes")
cols = dbc.getQueryResult() # id,end_point,vnodes,cores,status,role,create_time,offline reason
return {c[1]:c[4] for c in cols} # {'xxx:6030':'ready', 'xxx:6130':'ready'}
def createDnode(self, dbt: DbTarget):
"""
With a connection to the "first" EP, let's create a dnode for someone else who
wants to join.
"""
dbc = DbConn.createNative(self.getDbTarget())
dbc.open()
if dbt.getEp() in self._getDnodes(dbc):
Logging.info("Skipping DNode creation for: {}".format(dbt))
dbc.close()
return
sql = "CREATE DNODE \"{}\"".format(dbt.getEp())
dbc.execute(sql)
dbc.close()
def getStatus(self):
return self._smThread.getStatus()
def getSmThread(self):
return self._smThread
def start(self):
if not self.getStatus().isStopped():
raise CrashGenError("Cannot start instance from status: {}".format(self.getStatus()))
Logging.info("Starting TDengine instance: {}".format(self))
self.generateCfgFile() # service side generates config file, client does not
self.rotateLogs()
self._smThread.start(self.getServiceCmdLine())
def stop(self):
self._smThread.stop()
def isFirst(self):
return self._tInstNum == 0
class TdeSubProcess:
"""
A class to to represent the actual sub process that is the run-time
of a TDengine instance.
It takes a TdeInstance object as its parameter, with the rationale being
"a sub process runs an instance".
"""
# RET_ALREADY_STOPPED = -1
# RET_TIME_OUT = -3
# RET_SUCCESS = -4
def __init__(self):
self.subProcess = None
# if tInst is None:
# raise CrashGenError("Empty instance not allowed in TdeSubProcess")
# self._tInst = tInst # Default create at ServiceManagerThread
def getStdOut(self):
return self.subProcess.stdout
def getStdErr(self):
return self.subProcess.stderr
def isRunning(self):
return self.subProcess is not None
def getPid(self):
return self.subProcess.pid
def start(self, cmdLine):
ON_POSIX = 'posix' in sys.builtin_module_names
# Sanity check
if self.subProcess: # already there
raise RuntimeError("Corrupt process state")
self.subProcess = subprocess.Popen(
cmdLine,
shell=False,
# svcCmdSingle, shell=True, # capture core dump?
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
# bufsize=1, # not supported in binary mode
close_fds=ON_POSIX
) # had text=True, which interferred with reading EOF
def stop(self):
"""
Stop a sub process, and try to return a meaningful return code.
Common POSIX signal values (from man -7 signal):
SIGHUP 1
SIGINT 2
SIGQUIT 3
SIGILL 4
SIGTRAP 5
SIGABRT 6
SIGIOT 6
SIGBUS 7
SIGEMT -
SIGFPE 8
SIGKILL 9
SIGUSR1 10
SIGSEGV 11
SIGUSR2 12
"""
if not self.subProcess:
print("Sub process already stopped")
return # -1
retCode = self.subProcess.poll() # ret -N means killed with signal N, otherwise it's from exit(N)
if retCode: # valid return code, process ended
retCode = -retCode # only if valid
Logging.warning("TSP.stop(): process ended itself")
self.subProcess = None
return retCode
# process still alive, let's interrupt it
print("Terminate running process, send SIG_INT and wait...")
# sub process should end, then IPC queue should end, causing IO thread to end
# sig = signal.SIGINT
sig = signal.SIGKILL
self.subProcess.send_signal(sig) # SIGNINT or SIGKILL
self.subProcess.wait(20)
retCode = self.subProcess.returncode # should always be there
# May throw subprocess.TimeoutExpired exception above, therefore
# The process is guranteed to have ended by now
self.subProcess = None
if retCode != 0: # != (- signal.SIGINT):
Logging.error("TSP.stop(): Failed to stop sub proc properly w/ SIG {}, retCode={}".format(sig, retCode))
else:
Logging.info("TSP.stop(): sub proc successfully terminated with SIG {}".format(sig))
return - retCode
class ServiceManager:
PAUSE_BETWEEN_IPC_CHECK = 1.2 # seconds between checks on STDOUT of sub process
def __init__(self, numDnodes): # >1 when we run a cluster
Logging.info("TDengine Service Manager (TSM) created")
self._numDnodes = numDnodes # >1 means we have a cluster
self._lock = threading.Lock()
# signal.signal(signal.SIGTERM, self.sigIntHandler) # Moved to MainExec
# signal.signal(signal.SIGINT, self.sigIntHandler)
# signal.signal(signal.SIGUSR1, self.sigUsrHandler) # different handler!
self.inSigHandler = False
# self._status = MainExec.STATUS_RUNNING # set inside
# _startTaosService()
self._runCluster = (numDnodes > 1)
self._tInsts : List[TdeInstance] = []
for i in range(0, numDnodes):
ti = self._createTdeInstance(i) # construct tInst
self._tInsts.append(ti)
# self.svcMgrThreads : List[ServiceManagerThread] = []
# for i in range(0, numDnodes):
# thread = self._createThread(i) # construct tInst
# self.svcMgrThreads.append(thread)
def _createTdeInstance(self, dnIndex):
if not self._runCluster: # single instance
subdir = 'test'
else: # Create all threads in a cluster
subdir = 'cluster_dnode_{}'.format(dnIndex)
fepPort= 6030 # firstEP Port
port = fepPort + dnIndex * 100
return TdeInstance(subdir, dnIndex, port, fepPort)
# return ServiceManagerThread(dnIndex, ti)
def _doMenu(self):
choice = ""
while True:
print("\nInterrupting Service Program, Choose an Action: ")
print("1: Resume")
print("2: Terminate")
print("3: Restart")
# Remember to update the if range below
# print("Enter Choice: ", end="", flush=True)
while choice == "":
choice = input("Enter Choice: ")
if choice != "":
break # done with reading repeated input
if choice in ["1", "2", "3"]:
break # we are done with whole method
print("Invalid choice, please try again.")
choice = "" # reset
return choice
def sigUsrHandler(self, signalNumber, frame):
print("Interrupting main thread execution upon SIGUSR1")
if self.inSigHandler: # already
print("Ignoring repeated SIG...")
return # do nothing if it's already not running
self.inSigHandler = True
choice = self._doMenu()
if choice == "1":
self.sigHandlerResume() # TODO: can the sub-process be blocked due to us not reading from queue?
elif choice == "2":
self.stopTaosServices()
elif choice == "3": # Restart
self.restart()
else:
raise RuntimeError("Invalid menu choice: {}".format(choice))
self.inSigHandler = False
def sigIntHandler(self, signalNumber, frame):
print("ServiceManager: INT Signal Handler starting...")
if self.inSigHandler:
print("Ignoring repeated SIG_INT...")
return
self.inSigHandler = True
self.stopTaosServices()
print("ServiceManager: INT Signal Handler returning...")
self.inSigHandler = False
def sigHandlerResume(self):
print("Resuming TDengine service manager (main thread)...\n\n")
# def _updateThreadStatus(self):
# if self.svcMgrThread: # valid svc mgr thread
# if self.svcMgrThread.isStopped(): # done?
# self.svcMgrThread.procIpcBatch() # one last time. TODO: appropriate?
# self.svcMgrThread = None # no more
def isActive(self):
"""
Determine if the service/cluster is active at all, i.e. at least
one thread is not "stopped".
"""
for ti in self._tInsts:
if not ti.getStatus().isStopped():
return True
return False
def isRunning(self):
for ti in self._tInsts:
if not ti.getStatus().isRunning():
return False
return True
# def isRestarting(self):
# """
# Determine if the service/cluster is being "restarted", i.e., at least
# one thread is in "restarting" status
# """
# for thread in self.svcMgrThreads:
# if thread.isRestarting():
# return True
# return False
def isStable(self):
"""
Determine if the service/cluster is "stable", i.e. all of the
threads are in "stable" status.
"""
for ti in self._tInsts:
if not ti.getStatus().isStable():
return False
return True
def _procIpcAll(self):
while self.isActive():
Progress.emit(Progress.SERVICE_HEART_BEAT)
for ti in self._tInsts: # all thread objects should always be valid
# while self.isRunning() or self.isRestarting() : # for as long as the svc mgr thread is still here
status = ti.getStatus()
if status.isRunning():
th = ti.getSmThread()
th.procIpcBatch() # regular processing,
if status.isStopped():
th.procIpcBatch() # one last time?
# self._updateThreadStatus()
time.sleep(self.PAUSE_BETWEEN_IPC_CHECK) # pause, before next round
# raise CrashGenError("dummy")
print("Service Manager Thread (with subprocess) ended, main thread exiting...")
def _getFirstInstance(self):
return self._tInsts[0]
def startTaosServices(self):
with self._lock:
if self.isActive():
raise RuntimeError("Cannot start TAOS service(s) when one/some may already be running")
# Find if there's already a taosd service, and then kill it
for proc in psutil.process_iter():
if proc.name() == 'taosd':
print("Killing an existing TAOSD process in 2 seconds... press CTRL-C to interrupt")
time.sleep(2.0)
proc.kill()
# print("Process: {}".format(proc.name()))
# self.svcMgrThread = ServiceManagerThread() # create the object
for ti in self._tInsts:
ti.start()
if not ti.isFirst():
tFirst = self._getFirstInstance()
tFirst.createDnode(ti.getDbTarget())
ti.getSmThread().procIpcBatch(trimToTarget=10, forceOutput=True) # for printing 10 lines
def stopTaosServices(self):
with self._lock:
if not self.isActive():
Logging.warning("Cannot stop TAOS service(s), already not active")
return
for ti in self._tInsts:
ti.stop()
def run(self):
self.startTaosServices()
self._procIpcAll() # pump/process all the messages, may encounter SIG + restart
if self.isActive(): # if sig handler hasn't destroyed it by now
self.stopTaosServices() # should have started already
def restart(self):
if not self.isStable():
Logging.warning("Cannot restart service/cluster, when not stable")
return
# self._isRestarting = True
if self.isActive():
self.stopTaosServices()
else:
Logging.warning("Service not active when restart requested")
self.startTaosServices()
# self._isRestarting = False
# def isRunning(self):
# return self.svcMgrThread != None
# def isRestarting(self):
# return self._isRestarting
class ServiceManagerThread:
"""
A class representing a dedicated thread which manages the "sub process"
of the TDengine service, interacting with its STDOUT/ERR.
It takes a TdeInstance parameter at creation time, or create a default
"""
MAX_QUEUE_SIZE = 10000
def __init__(self):
# Set the sub process
self._tdeSubProcess = None # type: TdeSubProcess
# Arrange the TDengine instance
# self._tInstNum = tInstNum # instance serial number in cluster, ZERO based
# self._tInst = tInst or TdeInstance() # Need an instance
self._thread = None # The actual thread, # type: threading.Thread
self._status = Status(Status.STATUS_STOPPED) # The status of the underlying service, actually.
def __repr__(self):
return "[SvcMgrThread: status={}, subProc={}]".format(
self.getStatus(), self._tdeSubProcess)
def getStatus(self):
return self._status
# Start the thread (with sub process), and wait for the sub service
# to become fully operational
def start(self, cmdLine):
if self._thread:
raise RuntimeError("Unexpected _thread")
if self._tdeSubProcess:
raise RuntimeError("TDengine sub process already created/running")
Logging.info("Attempting to start TAOS service: {}".format(self))
self._status.set(Status.STATUS_STARTING)
self._tdeSubProcess = TdeSubProcess()
self._tdeSubProcess.start(cmdLine)
self._ipcQueue = Queue()
self._thread = threading.Thread( # First thread captures server OUTPUT
target=self.svcOutputReader,
args=(self._tdeSubProcess.getStdOut(), self._ipcQueue))
self._thread.daemon = True # thread dies with the program
self._thread.start()
self._thread2 = threading.Thread( # 2nd thread captures server ERRORs
target=self.svcErrorReader,
args=(self._tdeSubProcess.getStdErr(), self._ipcQueue))
self._thread2.daemon = True # thread dies with the program
self._thread2.start()
# wait for service to start
for i in range(0, 100):
time.sleep(1.0)
# self.procIpcBatch() # don't pump message during start up
print("_zz_", end="", flush=True)
if self._status.isRunning():
Logging.info("[] TDengine service READY to process requests")
Logging.info("[] TAOS service started: {}".format(self))
# self._verifyDnode(self._tInst) # query and ensure dnode is ready
# Logging.debug("[] TAOS Dnode verified: {}".format(self))
return # now we've started
# TODO: handle failure-to-start better?
self.procIpcBatch(100, True) # display output before cronking out, trim to last 20 msgs, force output
raise RuntimeError("TDengine service did not start successfully: {}".format(self))
def _verifyDnode(self, tInst: TdeInstance):
dbc = DbConn.createNative(tInst.getDbTarget())
dbc.open()
dbc.query("show dnodes")
# dbc.query("DESCRIBE {}.{}".format(dbName, self._stName))
cols = dbc.getQueryResult() # id,end_point,vnodes,cores,status,role,create_time,offline reason
# ret = {row[0]:row[1] for row in stCols if row[3]=='TAG'} # name:type
isValid = False
for col in cols:
# print("col = {}".format(col))
ep = col[1].split(':') # 10.1.30.2:6030
print("Found ep={}".format(ep))
if tInst.getPort() == int(ep[1]): # That's us
# print("Valid Dnode matched!")
isValid = True # now we are valid
break
if not isValid:
print("Failed to start dnode, sleep for a while")
time.sleep(600)
raise RuntimeError("Failed to start Dnode, expected port not found: {}".
format(tInst.getPort()))
dbc.close()
def stop(self):
# can be called from both main thread or signal handler
print("Terminating TDengine service running as the sub process...")
if self.getStatus().isStopped():
print("Service already stopped")
return
if self.getStatus().isStopping():
print("Service is already being stopped")
return
# Linux will send Control-C generated SIGINT to the TDengine process
# already, ref:
# https://unix.stackexchange.com/questions/176235/fork-and-how-signals-are-delivered-to-processes
if not self._tdeSubProcess:
raise RuntimeError("sub process object missing")
self._status.set(Status.STATUS_STOPPING)
# retCode = self._tdeSubProcess.stop()
try:
retCode = self._tdeSubProcess.stop()
# print("Attempted to stop sub process, got return code: {}".format(retCode))
if retCode == signal.SIGSEGV : # SGV
Logging.error("[[--ERROR--]]: TDengine service SEGV fault (check core file!)")
except subprocess.TimeoutExpired as err:
print("Time out waiting for TDengine service process to exit")
else:
if self._tdeSubProcess.isRunning(): # still running, should now never happen
print("FAILED to stop sub process, it is still running... pid = {}".format(
self._tdeSubProcess.getPid()))
else:
self._tdeSubProcess = None # not running any more
self.join() # stop the thread, change the status, etc.
# Check if it's really stopped
outputLines = 10 # for last output
if self.getStatus().isStopped():
self.procIpcBatch(outputLines) # one last time
Logging.debug("End of TDengine Service Output: {}".format(self))
Logging.info("----- TDengine Service (managed by SMT) is now terminated -----\n")
else:
print("WARNING: SMT did not terminate as expected: {}".format(self))
def join(self):
# TODO: sanity check
if not self.getStatus().isStopping():
raise RuntimeError(
"SMT.Join(): Unexpected status: {}".format(self._status))
if self._thread:
self._thread.join()
self._thread = None
self._status.set(Status.STATUS_STOPPED)
# STD ERR thread
self._thread2.join()
self._thread2 = None
else:
print("Joining empty thread, doing nothing")
def _trimQueue(self, targetSize):
if targetSize <= 0:
return # do nothing
q = self._ipcQueue
if (q.qsize() <= targetSize): # no need to trim
return
Logging.debug("Triming IPC queue to target size: {}".format(targetSize))
itemsToTrim = q.qsize() - targetSize
for i in range(0, itemsToTrim):
try:
q.get_nowait()
except Empty:
break # break out of for loop, no more trimming
TD_READY_MSG = "TDengine is initialized successfully"
def procIpcBatch(self, trimToTarget=0, forceOutput=False):
self._trimQueue(trimToTarget) # trim if necessary
# Process all the output generated by the underlying sub process,
# managed by IO thread
print("<", end="", flush=True)
while True:
try:
line = self._ipcQueue.get_nowait() # getting output at fast speed
self._printProgress("_o")
except Empty:
# time.sleep(2.3) # wait only if there's no output
# no more output
print(".>", end="", flush=True)
return # we are done with THIS BATCH
else: # got line, printing out
if forceOutput:
Logging.info(line)
else:
Logging.debug(line)
print(">", end="", flush=True)
_ProgressBars = ["--", "//", "||", "\\\\"]
def _printProgress(self, msg): # TODO: assuming 2 chars
print(msg, end="", flush=True)
pBar = self._ProgressBars[Dice.throw(4)]
print(pBar, end="", flush=True)
print('\b\b\b\b', end="", flush=True)
def svcOutputReader(self, out: IO, queue):
# Important Reference: https://stackoverflow.com/questions/375427/non-blocking-read-on-a-subprocess-pipe-in-python
# print("This is the svcOutput Reader...")
# for line in out :
for line in iter(out.readline, b''):
# print("Finished reading a line: {}".format(line))
# print("Adding item to queue...")
try:
line = line.decode("utf-8").rstrip()
except UnicodeError:
print("\nNon-UTF8 server output: {}\n".format(line))
# This might block, and then causing "out" buffer to block
queue.put(line)
self._printProgress("_i")
if self._status.isStarting(): # we are starting, let's see if we have started
if line.find(self.TD_READY_MSG) != -1: # found
Logging.info("Waiting for the service to become FULLY READY")
time.sleep(1.0) # wait for the server to truly start. TODO: remove this
Logging.info("Service is now FULLY READY") # TODO: more ID info here?
self._status.set(Status.STATUS_RUNNING)
# Trim the queue if necessary: TODO: try this 1 out of 10 times
self._trimQueue(self.MAX_QUEUE_SIZE * 9 // 10) # trim to 90% size
if self._status.isStopping(): # TODO: use thread status instead
# WAITING for stopping sub process to finish its outptu
print("_w", end="", flush=True)
# queue.put(line)
# meaning sub process must have died
Logging.info("\nEnd of stream detected for TDengine STDOUT: {}".format(self))
out.close()
def svcErrorReader(self, err: IO, queue):
for line in iter(err.readline, b''):
print("\nTDengine Service (taosd) ERROR (from stderr): {}".format(line))
Logging.info("\nEnd of stream detected for TDengine STDERR: {}".format(self))
err.close()