Connection interrupted with error AWS_ERROR_MQTT_UNEXPECTED_HANGUP after inactive period

[nickjmeyer]

I have an IoT application that publishes some data to a single MQTT periodically. If the frequency is short enough, everything works as expected, data is published, it's received by the server, and all is well. However, I've noticed that if I wait long enough, the next time I try to publish, then the call to Publish(...) succeeds, but the callback is never executed. When this happens, I see the following error messages:

Connection interrupted with error: libaws-c-mqtt: AWS_ERROR_MQTT_UNEXPECTED_HANGUP, The connection was closed unexpectedly.

A short time later, I see the connection resume with an error code of 0, but immediately gets interrupted again with the same error:

Connection resumed.
Connection interrupted with error: libaws-c-mqtt: AWS_ERROR_MQTT_UNEXPECTED_HANGUP, The connection was closed unexpectedly.

This continues, but the resume / interrupt sequence happens more infrequent as time goes on. Once this happens once, I'm never able to publish again. I'm not sure exactly how long I have to wait before publishing to get this error, but everything succeeds at 700 seconds or less and fails at 800 seconds or more.

I have found some articles suggesting that AWS_ERROR_MQTT_UNEXPECTED_HANGUP is due to insufficient permissions. I tried creating a policy allowing actions * for resources * to test that theory, but that didn't have any effect.

I also added the following to the client build config, but this didn't help either.

  builder.WithTcpKeepAlive();
  builder.WithTcpConnectTimeout(std::chrono::milliseconds(std::chrono::seconds(3)).count());
  builder.WithTcpKeepAliveTimeout(std::chrono::seconds(60).count());
  builder.WithTcpKeepAliveInterval(std::chrono::seconds(60).count());

Here are the parameters to the connect function that I'm using:

  bool const clean_session{false};
  uint16_t const keep_alive_time_s{std::chrono::seconds(60).count()};
  uint32_t const ping_timeout_ms{std::chrono::milliseconds(std::chrono::seconds(3)).count()};

Any suggestions for how to debug?

[nickjmeyer]

Well, it seems the underlying problem was sending a message that was too large. I'm assuming that the constant reconnect + interrupt was the publish method retrying to send. When retrying, it reconnected immediately, but because the payload is too large, it caused an interruption again.

[github-actions[bot]]

Hello! Reopening this discussion to make it searchable.