Skip to main content
Version: 2.8.0

Monitoring

As a developer, you will encounter bugs. 😲

To help you in your development journey, Luos provides monitoring and debugging mechanisms that can give you the opportunity of having clearer visibility of your network.

Luos self-healing capabilities

caution

Make sure to read and understand how to create Luos services before reading this page.

Finding, understanding, and managing bugs on multiple boards running multiple services can be hard. To make your life easier, Luos engine allows you to get some basic information about any problems in your system, allowing you to adapt to them.

Service exclusion

Luos engine includes an acknowledgment management using the SERVICEIDACK target_mode. This mode guarantees the proper reception of critical messages.

If Luos fails to reach its target using SERVICEIDACK, it will retry 10 times. If the acknowledgment still fails, the targeted service is declared excluded. Excluded services are removed from the routing table to avoid messaging from any services, preserving bandwidth for the rest of the system.

note
  • Gates services can report service exclusion through JSON.
  • Pyluos can report service exclusion through gates. :::

Luos engine's statistics

Luos engine monitors some values representing the sanity of your nodes and services.

  • Gates services can report statistics through JSON.
  • Pyluos can display statistics through gates. :::

Node statistics

Inside any service, you can access the host node's statistics values using the luos_stats_t structure. This structure gives you access to several values:

  • memory: Memory statisctics information.
  • rx_msg_stack_ratio: Percentage of memory occupation of Rx message management tasks.
  • engine_msg_stack_ratio: Percentage of memory occupation of Luos engine's tasks.
  • tx_msg_stack_ratio: Percentage of memory occupation of Tx message management tasks.
  • buffer_occupation_ratio: Percentage of memory occupation of the message buffer.
  • msg_drop_number: Number of messages dropped due to a lack of memory (older messages are dropped to be replaced by new ones).
  • max_loop_time_ms: Maximum time in ms between Luos_loop executions.

You can access to node statistics by using service.node_statistics.

Service statistics

In any service, you have access to statistics values using the service_stats_t structure. This structure gives you access to a specific service's statistic value:

  • max_retry: Maximum number of sent retries due to a NAK or collision with another service.

You can access node statistics by using service.statistics.

Assert

Luos engine allows you to declare a critical failure on a service to an entire network. To handle it, Luos engine exposes a LUOS_ASSERT macro that will enable you to test some conditions on it to prevent wrong values. For example:

 LUOS_ASSERT(arg_ptr != NULL);

In this case, if arg_ptr is not initialized, Luos engine will crash the entire node and send a message to all other services with the file and line where the crash occurred. All other nodes will remove all the services from the crashed node from the routing table.

note
  • Gates services can [report asserting of other nodes through JSON].
  • Pyluos can display assert through gates.

Sniffer

An additional monitoring mechanism provided by Luos engine is the integration of a sniffer MCU into the network. The sniffer is responsible for gathering all the messages that are transferred into a Luos network, transmitting them serially to your computer, and displaying them in a logger, allowing you to examine the behavior of your nodes and services.

The sniffer, which consists in an application and a driver, can be easily ported on a simple MCU the same way as a serial gate. It can be connected to your network as any other node. The reception of the messages from the computer is achieved by transmitting the messages serially using a USB cable, while these messages are handled by Pyluos.

All you have to do is to connect the sniffer to your MCU network and to the computer, initialize the connection using Pyluos after you spotted the name of the USB port that the sniffer is connected (for example, COM13):

import pyluos
from pyluos import Sniffer
sniffer = Sniffer('COM13')

and then send the following command:

 sniffer.start
info

If you need more information on how to debug your application using a sniffer, contact us at hello@luos.io or connect to the Luos community Discord.

Inspector

In order to facilitate the debugging process, Luos offers a more advanced monitoring tool to help you investigate the behavior of your network. In other words, the inspector is capable of receiving the information of all the nodes and services in the network, as well as all the messages that are transmitted through it, and finally sending them to any other machine via a serial USB port.

The inspector includes the functionalities of the sniffer, with the main difference that it is a common Luos service that works like any other service in the network. That is to say, the inspector service is detected, and can interact with the other services existing in the same network. Also, we have added aditional functionalities, which enrich the capacities of monitoring the embedded systems.

The inspector, exactly like the gate, is a Luos engine's app service, so it is able to be hosted in any MCU running Luos engine.

In order to exploit the functionalities of the inspector app, we need to combine it with the communication driver service, which is called pipe.

The format of the messages that are transmitted serially from the inspector using the pipe driver follow the specified Luos engine communication protocol.

Inspector Features

  1. System's messages reception
  2. Routing table sharing
  3. Services' and nodes' statics occupation
  4. Services' and nodes' firmware revision occupation
  5. Services' and nodes' Luos engine's revision
  6. Verbose mode
  7. Assert messages occupation

The inspector and the pipe are two separate services, they can be placed on the same node or on separate nodes. However, if they are placed on the same node, they charge the network with additional messages, so it can have an impact on the normal behavior of the other services.

note

The inspector does not receive the messages that regard the detection, as it is a normal service which is not yet detected.

info

If you need more information on how to debug your application using the inspector, contact us at hello@luos.io or connect to the Luos community Discord.