New LTE Firmware releases v1.18.1.r4 (stable) & 1.19.0.b5 (development) & NB-IoT firmware 40343

[iwahdan]

Hi @danielm the error you are getting
HTTP POST failed: 'NoneType' object has no attribute 'split' is because the socket operation socket.recv() has returned error code = EGAIN which could be when the operation timesout if you are are setting a socket timeout > 0 , where this error is considered non blocking by the readline function in MicroPython py/stream.c where in that case a None object is returned. maybe this should be handled in requests.py

on the other hand the problem of failing to create more sockets HTTP POST failed: 23 that happens at some point, is because lwip stack fails to allocate new sockets and this is because every time the exception 'NoneType' object has no attribute 'split' is raised it is not caught by the exception handler in requests.py

    except OSError:
        s.close()
        raise

As it is an AttributeError not an OSError exp, thus the socket is not closed and by turn the lwip stack does not free that socket as it considers conn still active since err code is EGAIN.
I know that in the normal case when the POST is successful you also don't close the socket afterwards but that is not a problem as the lwip will Reuse the oldest socket not closed yet [Note: that the Max active sockets is Configured to 10] .

so the flow goes as follow socket # 54 gets allocated for first message then 55 for second and so on till #64 .. then lwip will allocate socket #54 again for creating new socket then #55 and so on.
However when a socket connection timeout if the socket is not closed the socket # will be reserved forever.

so in your case if a socket operation timeout say every 1000 successful POST the esp32 will fail to allocate new sockets after 10000 POSTs (so it is a socket leak)

so to fix you problem you have to do the following modifications in requests.py:

Update exception handler to include all exceptions so that the socket is closed in any case:

    except :
        s.close()
        raise

maybe it is also a good idea to close the socket after successful POST as follows:

response = urequests.post(url, headers={ "Content-Type": "application/json", "Accept": "application/json"}, data='{"time": "%s"}' % (time.time()))
print("INFO", "%s. HTTP POST successful: %s" % (n, response.text))
response.close()

Thanks for your help.

[danielm]

Hi @iwahdan. Thank you for excelent analysis. I will perform recommended modifications and also long-term tests.
I believe that this issue is not related to issues you are describing in this post?
https://forum.pycom.io/topic/3985/new-lte-firmware-releases-v1-18-1-r4-stable-1-19-0-b5-development-nb-iot-firmware-40343/39

[iwahdan]

@danielm Thanks, No the problem I mentioned in that post is because the ssl handshake does not timeout, which I think is not what happens in your case.

[danielm]

@iwahdan
Is it possible to fix the ssl handshake not timing out?
What about socket.connect() timeout workaround implementation?

Due to nature of network communication any socket operation may fail but it should never hang if timeout is set. Otherwise it will not be possible to develop reliable applications.

[iwahdan]

Hi @danielm, I am investigating the handshake problem, and for a the socket.connect timeout I think it is possible to have a workaround in the Micropython, but it would need proper design to ensure a robust implementation .

I've already created an internal ticket for it.

[danielm]

@iwahdan
Great, please let me know if you need help with testing.

[tlanier]

@tlanier
The bug I was experiencing was caused by my inexperience using Python. I was allocating a new WLAN object every time I ran the code without destroying the old object.

Here is a listing of my new wifi.py module that fixes my problem. I'm sorry if I've caused anyone to lose confidence in the GPy or spend time looking for bugs that did not exist.

from network import WLAN
from util import log
import utime

wlan = None

def connect():
    global wlan

    if (wlan != None):
        if wlan.isconnected():
            print('Already connected')
            return

    wlan = WLAN(mode=WLAN.STA)
    
    while True:
        # scan for WIFI access points
        nets = wlan.scan()
        print(nets)
        
        for net in nets:
            if (net.ssid == 'NURSEX'):
                print('NURSEX WIFI found')
                # wlan.connect(net.ssid, auth=(net.sec, 'password'), timeout=5000)
                wlan.connect(net.ssid, auth=(net.sec, 'password'))
                while not wlan.isconnected():
                    utime.sleep_ms(1000)
                print('WLAN connection succeeded!')
                return
        
        utime.sleep_ms(1000)

def disconnect():
    global wlan

    if (wlan != None):
        wlan.disconnect()
        wlan = None```

[iwahdan]

@tlanier No problem!, thanks for letting us know

[danielm]

@iwahdan
Are there any news regarding solution for known socket issues?

[wghirakawa]

@danielm what socket issues are you referring too?

[danielm]

@wghirakawa
The issue is that socket.connect() and SSL handshake will not timeout also in case the socket timeout is set and hang the application forever.
https://forum.pycom.io/topic/3985/new-lte-firmware-releases-v1-18-1-r4-stable-1-19-0-b5-development-nb-iot-firmware-40343/45