MQTT Client loses connection

[rbedemann]

Hi,

when having a MQTT client running for a while, I see a strange behaviour:

In irregular time ranges, following error shows up:

EXCEPTION error. 
Connection closed
  0: tcp_error_                system/modules/tcp.toit:205:3
  1: TcpSocket_.ensure_state_  system/modules/tcp.toit:78:14
  2: TcpSocket.read            system/modules/tcp.toit:154:14
  3: ClientSocketResource.read system/components/network.toit:120:20
  4: register_network.<lambda> system/components/network.toit:32:12
  5: KernelBroker.register_descriptor_procedure.<lambda> system/kernel/rpc.toit:44:14
  6: RpcRequest_.process.<block> <sdk>/rpc/broker.toit:98:26
  7: RpcRequest_.process       <sdk>/rpc/broker.toit:95:3
  8: RpcRequestQueue_.ensure_processing_task_.<lambda>.<block>.<block> <sdk>/rpc/broker.toit:214:20
  9: RpcRequestQueue_.ensure_processing_task_.<lambda>.<block> <sdk>/rpc/broker.toit:209:9
 10: RpcRequestQueue_.ensure_processing_task_.<lambda> <sdk>/rpc/broker.toit:204:56

Afterwards, following error shows:

EXCEPTION error. 
Connection closed
  0: Client.start.<lambda>     <pkg:mqtt>/client.toit:54:32
  1: Client.start.<lambda>     <pkg:mqtt>/client.toit:84:30

And finally:

Reset after exiting deep sleep mode.

I used the exact setup for the client as in the example files. I already tried different MQTT brokers, different WiFis and shrinked down the code to only have a client connecting to a broker without any subscription.

Is this something that can be avoided somehow? Or any clue why this could happen?

Thanks in advance!

[floitsch]

For now, I work under the hypothesis that the "reset after exiting deep sleep mode" is just a red herring, and just happens because the program closes and Toit automatically puts the device to deep sleep as it didn't to do any work anymore.

From the stacktrace, it looks like the TCP connection gets dropped. Since you have tried different WiFis and different MQTT brokers this is probably an issue with the TCP stack. Could be that we (Toit) do something wrong, or it might also be that the underlying Espressif code is not that stable, or it might be the combination of Toit with the Espressif code.

Could you maybe try the open-source version (https://github.com/jaguar) and see if that fixes the issue? Our work towards Toit 2.0 is happening there, and the latest Jaguar uses a more recent Espressif SDK (and also a newer garbage collector).

It might also help (although I doubt it will reveal more), to use a more verbose logger.

[rbedemann]

Thanks for the info! I just installed the 2.0.0 alpha version via jag and will observe the MQTT client running there.

Any plan when the 2.x.x version will be released?

[floitsch]

The 2.0.0 branch itself is already quite stable. We keep the "alpha" tag to give us room to do breaking changes, but the code is in pretty good shape.
Another feature we are missing is the fleet-management. We are currently going away from having our own proprietary protocol and servers, and use common MQTT servers instead. This requires a bit of a rewrite of the fleet-management software (basically the command-line toit tool). Once that's ready, we will probably tag a 2.0.0. At that point we will deprecate the v1 version and our servers, including the console.toit.io web-page.