python - Python 垃圾收集器有问题吗？

Question

我有一个简单的程序，它读取一个包含几百万行的大文件，解析每一行(numpy array)并转换成 double 组(python array)然后写入 hdf5 文件。我重复这个循环好几天。读取每个文件后，我删除所有对象并调用垃圾收集器。当我运行该程序时，第一天的解析没有任何错误，但第二天我得到了 MemoryError。我监控了程序的内存使用情况，在解析的第一天，内存使用量约为 1.5 GB。第一天的解析完成后，内存使用量下降到 50 MB。现在，当第 2 天开始时，我尝试从文件中读取行，我得到了 MemoryError。以下是程序的输出。

source file extracted at C:\rfadump\au\2012.08.07.txt
parsing started
current time: 2012-09-16 22:40:16.829000
500000 lines parsed
1000000 lines parsed
1500000 lines parsed
2000000 lines parsed
2500000 lines parsed
3000000 lines parsed
3500000 lines parsed
4000000 lines parsed
4500000 lines parsed
5000000 lines parsed
parsing done.
end time is 2012-09-16 23:34:19.931000
total time elapsed 0:54:03.102000
repacking file
done
> s:\users\aaj\projects\pythonhf\rfadumptohdf.py(132)generateFiles()
-> while single_date <= self.end_date:
(Pdb) c
*** 2012-08-08 ***
source file extracted at C:\rfadump\au\2012.08.08.txt
cought an exception while generating file for day 2012-08-08.
Traceback (most recent call last):
  File "rfaDumpToHDF.py", line 175, in generateFile
    lines = self.rawfile.read().split('|\n')
MemoryError

我非常确定 Windows 系统任务管理器显示此进程的内存使用量为 50 MB。看起来 Python 的垃圾收集器或内存管理器没有正确计算可用内存。应该有很多空闲内存，但它认为没有足够的。

有什么想法吗？

编辑

在这里添加我的代码

我会放上我的部分代码。我是 python 新手，请原谅我的 python 编码风格。

模块 1

def generateFile(self, current_date):
    try:
        print "*** %s ***" % current_date.strftime("%Y-%m-%d")
        weekday=current_date.weekday()
        if weekday >= 5:
            print "skipping weekend"
            return
        self.taqdb = taqDB(self.index, self.offset)
        cache_filename = os.path.join(self.cache_dir,current_date.strftime("%Y.%m.%d.h5"))
        outputFile = config.hdf5.filePath(self.index, date=current_date)
        print "cache file: ", cache_filename
        print "output file: ", outputFile

        tempdir = "C:\\rfadump\\"+self.region+"\\"  
        input_filename = tempdir + filename
        print "source file extracted at %s " % input_filename

        ## universe
        reader = rfaTextToTAQ.rfaTextToTAQ(self.tickobj)  ## PARSER
        count = 0
        self.rawfile = open(input_filename, 'r')
        lines = self.rawfile.read().split('|\n')
        total_lines = len(lines)
        self.rawfile.close()
        del self.rawfile
        print "parsing started"
        start_time = dt.datetime.now()
        print "current time: %s" % start_time
        #while(len(lines) > 0):
        while(count < total_lines):
            #line = lines.pop(0) ## This slows down processing
            result = reader.parseline(lines[count]+"|")
            count += 1
            if(count % 500000 == 0):
                print "%d lines parsed" %(count)
            if(result == None): 
                continue
            ric, timestamp, quotes, trades, levelsUpdated, tradeupdate = result
            if(len(levelsUpdated) == 0 and tradeupdate == False):
                continue
            self.taqdb.insert(result)

        ## write to hdf5 TODO
        writer = h5Writer.h5Writer(cache_filename, self.tickobj)
        writer.write(self.taqdb.groups)
        writer.close()

        del lines
        del self.taqdb, self.tickobj
        ##########################################################
        print "parsing done." 
        end_time = dt.datetime.now()
        print "end time is %s" % end_time
        print "total time elapsed %s" % (end_time - start_time)

        defragger = hdf.HDF5Defragmenter()
        defragger.Defrag(cache_filename,outputFile)
        del defragger
        print "done"
        gc.collect(2)
    except:
        print "cought an exception while generating file for day %s." % current_date.strftime("%Y-%m-%d")
        tb = traceback.format_exc()
        print tb

模块 2 - taqdb - 将解析后的数据存储在数组中

class taqDB:
  def __init__(self, index, offset):
    self.index = index
    self.tickcfg = config.hdf5.getTickConfig(index)
    self.offset = offset
    self.groups = {}

  def getGroup(self,ric):
    if (self.groups.has_key(ric) == False):
        self.groups[ric] = {}
    return self.groups[ric]

  def getOrderbookArray(self, ric, group):
    datasetname = orderBookName
    prodtype = self.tickcfg.getProdType(ric)
    if(prodtype == ProdType.INDEX):
        return
    orderbookArrayShape = self.tickcfg.getOrderBookArrayShape(prodtype)
    if(group.has_key(datasetname) == False):
        group[datasetname] = array.array("d")
        orderbookArray = self.tickcfg.getOrderBookArray(prodtype)
        return orderbookArray
    else:
        orderbookArray = group[datasetname]
        if(len(orderbookArray) == 0):
            return self.tickcfg.getOrderBookArray(prodtype)
        lastOrderbook = orderbookArray[-orderbookArrayShape[1]:]
        return np.array([lastOrderbook])

  def addToDataset(self, group, datasetname, timestamp, arr):
    if(group.has_key(datasetname) == False):
        group[datasetname] = array.array("d")
    arr[0,0]=timestamp
    a1 = group[datasetname]
    a1.extend(arr[0])

  def addToOrderBook(self, group, timestamp, arr):
    self.addToDataset(self, group, orderBookName, timestamp, arr)

  def insert(self, data):
    ric, timestamp, quotes, trades, levelsUpdated, tradeupdate = data
    delta = dt.timedelta(hours=timestamp.hour,minutes=timestamp.minute, seconds=timestamp.second, microseconds=(timestamp.microsecond/1000))
    timestamp = float(str(delta.seconds)+'.'+str(delta.microseconds)) + self.offset
    ## write to array
    group = self.getGroup(ric)

    orderbookUpdate = False
    orderbookArray = self.getOrderbookArray(ric, group)
    nonzero = quotes.nonzero()
    orderbookArray[nonzero] = quotes[nonzero] 
    if(np.any(nonzero)):
        self.addToDataset(group, orderBookName, timestamp, orderbookArray)
    if(tradeupdate == True):
        self.addToDataset(group, tradeName, timestamp, trades)

模块 3 - 解析器

class rfaTextToTAQ:
  """RFA Raw dump file reader. Readers single line (record) and returns an array or array of fid value pairs."""
  def __init__(self,tickconfig):
    self.tickconfig = tickconfig
    self.token = ''
    self.state = ReadState.SEQ_NUM
    self.fvstate = fvstate.FID
    self.quotes = np.array([]) # read from tickconfig
    self.trades = np.array([]) # read from tickconfig
    self.prodtype = ProdType.STOCK
    self.allquotes = {}
    self.alltrades = {}
    self.acvol = 0
    self.levelsUpdated = []
    self.quoteUpdate = False
    self.tradeUpdate = False
    self.depth = 0

  def updateLevel(self, index):
    if(self.levelsUpdated.__contains__(index) == False):
        self.levelsUpdated.append(index)

  def updateQuote(self, fidindex, field):
    self.value = float(self.value)
    if(self.depth == 1):
        index = fidindex[0]+(len(self.tickconfig.stkQuotes)*(self.depth - 1))
        self.quotes[index[0]][fidindex[1][0]] = self.value
        self.updateLevel(index[0])
    else:
        self.quotes[fidindex] = self.value
        self.updateLevel(fidindex[0][0])
    self.quoteUpdate = True

  def updateTrade(self, fidindex, field):
    #self.value = float(self.value)
    if(self.tickconfig.tradeUpdate(self.depth) == False):
        return
    newacvol = float(self.value)
    if(field == acvol):
        if(self.value > self.acvol):
            tradesize = newacvol - self.acvol
            self.acvol = newacvol
            self.trades[fidindex] = tradesize
            if(self.trades.__contains__(0) == False):
                self.tradeUpdate = True
    else:
        self.trades[fidindex] = self.value
        if(not (self.trades[0,1]==0 or self.trades[0,2]==0)):
            self.tradeUpdate = True

  def updateResult(self):
    field = ''
    valid, field = field_dict.FIDToField(int(self.fid), field)
    if(valid == False):
        return
    if(self.value == '0'):
        return
    if(self.prodtype == ProdType.STOCK):
        fidindex = np.where(self.tickconfig.stkQuotes == field)
        if(len(fidindex[0]) == 0):
            fidindex = np.where(self.tickconfig.stkTrades == field)
            if(len(fidindex[0]) == 0):
                return
            else:
                self.updateTrade(fidindex, field)
        else:
            self.updateQuote(fidindex, field)
    else:
        fidindex = np.where(self.tickconfig.futQuotes == field)
        if(len(fidindex[0]) == 0):
            fidindex = np.where(self.tickconfig.futTrades == field)
            if(len(fidindex[0]) == 0):
                return
            else:
                self.updateTrade(fidindex, field)
        else:
            self.updateQuote(fidindex, field)

  def getOrderBookTrade(self):
    if (self.allquotes.has_key(self.ric) == False):
        acvol = 0
        self.allquotes[self.ric] = self.tickconfig.getOrderBookArray(self.prodtype)
        trades = self.tickconfig.getTradesArray()
        self.alltrades[self.ric] = [trades, acvol]
    return self.allquotes[self.ric], self.alltrades[self.ric]

  def parseline(self, line):
    self.tradeUpdate = False
    self.levelsUpdated = []
    pos = 0
    length = len(line)
    self.state = ReadState.SEQ_NUM
    self.fvstate = fvstate.FID
    self.token = ''
    ch = ''
    while(pos < length):
        prevChar = ch
        ch = line[pos]
        pos += 1
        #SEQ_NUM
        if(self.state == ReadState.SEQ_NUM):
            if(ch != ','):
                self.token += ch
            else:
                self.seq_num = int(self.token)
                self.state = ReadState.TIMESTAMP
                self.token = ''
        # TIMESTAMP
        elif(self.state == ReadState.TIMESTAMP):
            if(ch == ' '):
                self.token = ''
            elif(ch != ','):
                self.token += ch
            else:
                if(len(self.token) != 12):
                    print "Invalid timestamp format. %s. skipping line.\n", self.token
                    self.state = ReadState.SKIPLINE
                else:
                    self.timestamp = datetime.strptime(self.token,'%H:%M:%S.%f') 
                    self.state = ReadState.RIC
                self.token = ''
        # RIC
        elif(self.state == ReadState.RIC):
            if(ch != ','):
                self.token += ch
            else:
                self.ric = self.token
                self.token = ''
                self.ric, self.depth = self.tickconfig.replaceRic(self.ric)
                self.prodtype = self.tickconfig.getProdType(self.ric)
                if(self.tickconfig.subscribed(self.ric)):
                    self.state = ReadState.UPDATE_TYPE
                    self.quotes, trades = self.getOrderBookTrade()
                    self.trades = trades[0]
                    self.acvol = trades[1]
                else:
                    self.state = ReadState.SKIPLINE
        # UPDATE_TYPE
        elif(self.state == ReadState.UPDATE_TYPE):
            if(ch != '|'):
                self.token += ch
            else:
                self.update_type = self.token
                self.token = ''
                self.state = ReadState.FVPAIRS
        #SKIPLINE
        elif(self.state == ReadState.SKIPLINE):
            return None
        # FV PAIRS
        elif(self.state == ReadState.FVPAIRS):
            # FID
            if(self.fvstate == fvstate.FID):
                if(ch != ','):
                    if(ch.isdigit() == False):
                        self.token = self.value+ch
                        self.fvstate = fvstate.FIDVALUE
                        self.state = ReadState.FVPAIRS
                    else:
                        self.token += ch
                else:
                    self.fid = self.token
                    self.token = ''
                    self.fvstate = fvstate.FIDVALUE
                    self.state = ReadState.FVPAIRS
            # FIDVALUE
            elif(self.fvstate == fvstate.FIDVALUE):
                if(ch != '|'):
                    self.token += ch
                else:
                    self.value = self.token
                    self.token = ''
                    self.state = ReadState.FVPAIRS
                    self.fvstate = fvstate.FID
                    # TODO set value
                    self.updateResult()
    return self.ric, self.timestamp, self.quotes, self.trades, self.levelsUpdated, self.tradeUpdate

谢谢。

Answer 1

The only reliable way to free memory is to terminate the process.

所以，如果你的主程序spawns a worker process完成大部分工作(一天内完成的工作)然后当该工作进程完成时，将释放使用的内存:

import multiprocessing as mp

def work(date):
    # Do most of the memory-intensive work here
    ...

while single_date <= self.end_date:
    proc = mp.Process(target = work, args = (single_date,))
    proc.start()
    proc.join()