Error -23 in mqtt_connect

[Bastien42]

Good afternoon,

My device communicates using MQTT a few times a day. After many connections I noticed that communications stopped. Every connection attempt would fail with:

Error in mqtt_connect: -23.

Error code 23 corresponds to "file table overflow", therefore, I assumed that it was linked to a mishandling of sockets when connection failed.

I have now made sure that mqtt_abort() is called in any case where connection may fail and when disconnecting, and I am currently testing those changes.

Are there other causes to error -23 in mqtt_connect() ?

Best regards

Bastien

[Håkon Alseth]

Hi,

 

The socket is closed when you call mqtt_disconnect(), as this will then subsequently call close():

https://github.com/nrfconnect/sdk-zephyr/blob/main/subsys/net/lib/mqtt/mqtt_transport_socket_tcp.c#L167

Similar for TLS:

https://github.com/nrfconnect/sdk-zephyr/blob/main/subsys/net/lib/mqtt/mqtt_transport_socket_tls.c#L249

 

It is very important that you close a socket before initiating a new connection, to avoid overflowing the fd table.

 

Kind regards,

Håkon

[Bastien42]

Thanks for this rapid answer.

Okay, I will implement that.

Is mqtt_abort() even useful in my case ? Should I try to use mqtt_disconnect only ?

[Håkon Alseth]

Hi,

 

Bastien42 said:

Okay, I tried it, i still seem to have an edge case where sockets are not closed properly.

Please note that each mqtt_connect() will effectively setup a new fd:

https://github.com/nrfconnect/sdk-zephyr/blob/main/subsys/net/lib/mqtt/mqtt_transport_socket_tcp.c#L21-L30

 

I would recommend that you check if your firmware has any scenarios where it re-connects without closing/disconnecting the former socket successfully.

Bastien42 said:

Do you know how to properly monitor the fd table ? I think that this will help me to get to the problem.

Looking at zephyr/sys/fdtable.h, it does not seem that there is an API directly to do this, unfortunately.

 

Kind regards,

Håkon

[Bastien42]

Yes sure, do connections to nRF Cloud via MQTT can also be source of problems ? 

[Håkon Alseth]

Hi,

 

Yes, other concurrent connections should be considered.

Please see here for the restrictions of the modem firmware when it comes to concurrent connections, especially if they are secure sockets:

https://nsscprodmedia.blob.core.windows.net/prod/software-and-other-downloads/dev-kits/nrf9160-dk/release_notes_modemfirmware/mfw_nrf9160_1.3.7_release_notes.txt

 

What is your current "CONFIG_ZVFS_POLL_MAX" set to? You can try to increase this slightly, in case it is the kernel's fd table that is overflowing.

  

Kind regards,

Håkon

[Bastien42]

Hi,

I am currently using a nRF9160 with NCS 2.7.0 and I don't think that CONFIG_ZVFS_POLL_MAX exists in that version. I suppose that i have a default value for the corresponding parameter

Best regards

Bastien

[Håkon Alseth]

Hi,

 

Bastien42 said:

I am currently using a nRF9160 with NCS 2.7.0 and I don't think that CONFIG_ZVFS_POLL_MAX exists in that version.

Check the value of config "CONFIG_POSIX_MAX_FDS".

 

As you mention, if left untouched, it should not be a problem, however; if you are concurrently maintaining several asynchronous connections; please see the restrictions in the release notes.

 

Kind regards,

Håkon