Python-light Memory Leak/ Python garbage collector not being called

christian lacdael

A background python task is resulting in the process VMZ value going up and up, till there's a crash after ~ 30 min.

I'm using spidev and onionGpio libraries to read an SPI device.

Any ideas?

JP Norair

Do you have the code to the python task that's running?

Seems to me like there's a memory leak in the code of one of the underlying libraries. Maybe you can figure out exactly the line(s) of python that causes the leak, then we can fix the C library.

christian lacdael

This is what I'm doing:

import spidev
import onionGpio

...

    def __init__(self):
          try:
                    self.spi = spidev.SpiDev(32766,1)
                    self.spi.max_speed_hz=100000;
                    self.spi.mode = 1;
                    self.ready = onionGpio.OnionGpio(0,1);
                    r = self.ready.setInputDirection();

     ...

    def _drdy(self):
            try:
                    while self.ready.getValue()[0] == '1':
                            continue;
            except:
                    self.ready._freeGpio();
                    print(traceback.format_exc());

    def getReadings(self):
            self._drdy();
            v1 = self.spi.xfer3([self.CMD_WMUX,0x00,self.MUX_AN0_AN1,self.CMD_SYNC,self.CMD_WAKEUP,self.CMD_RDATA],3);
            self._drdy();
            v2 = self.spi.xfer3([self.CMD_WMUX,0x00,self.MUX_AN2_AN3,self.CMD_SYNC,self.CMD_WAKEUP,self.CMD_RDATA],3);
            self._drdy();
            v3 = self.spi.xfer3([self.CMD_WMUX,0x00,self.MUX_AN4_AN5,self.CMD_SYNC,self.CMD_WAKEUP,self.CMD_RDATA],3);
            self._drdy();
            v4 = self.spi.xfer3([self.CMD_WMUX,0x00,self.MUX_AN6_AN7,self.CMD_SYNC,self.CMD_WAKEUP,self.CMD_RDATA],3);
            return [self._getVal(v1),self._getVal(v2),self._getVal(v3),self._getVal(v4)];



...

####  THE LOOP  ####

....

while True:
    try:
        if not lock:
            time.sleep(0.005);
            weights = adc.getReadings();
            display_count += 1;
            if display_count > 20:
                    display_count = 0;
                    client.publish(MQTT_TOPIC,"{\"weight\":%6.2f,\"adc1\":%d,\"adc2\":%d,\"adc3\":%d,\"adc4\":%d}" % (weight,weights[0],weights[1],weights[2],weights[3]));
    except KeyboardInterrupt:
        print(stamp()+" :Exiting, interrupt");
        sys.exit(1);
    except:
        print(traceback.format_exc());

christian lacdael

https://stackoverflow.com/questions/21524477/python-memory-leak

"So here is what I have found out:

the py-spidev module (or some part of spidev itself) seems to be the problem.    
besides the memory leak, py-spidev is incredibly slow
I've now got rid of py-spidev and write to /dev/spidev0.0 directly via file handle. No more excessive memory     usage and SPI communication now needs about two seconds, that's about a tenth of the time it needed before." - 

answered Feb 3 '14 at 20:14 tungl

JP Norair

@christian-lacdael Will you be able to follow suit and write to /dev/spidev_ directly? This seems simple enough.

christian lacdael

I compiled a simple spi C program, as per: http://linux-sunxi.org/SPIdev#In_the_user_space

 PID  PPID USER     STAT   VSZ %VSZ %CPU COMMAND
 3768  1816 root     R      900   1%  65% ./spi

The %VSZ value remains constant.

while (1==1){
    buffer=(char *)spi_read(0x10,0x0A,11,file); //reading the address 0x100A
    for (i =0; i < 11; i++) {
        printf("%X ",buffer[i]);
    }
}

I am planning to make a daemon C program that stores reads from the SPI device in a /tmp/buffer file. Then I can read the file with Python.