Closed ferazambuja closed 5 years ago
It seems that openHAB now have gone from having a starting /
to have %sysname%
in as the base topic. I would suggest not to change to home/%sysname%/...
for the openHAB controller. We need to find a more generic way of using the same MQTT core and only apply templates.
https://docs.openhab.org/addons/bindings/mqtt1/readme.html
So based on these new info I'd say we could at least change the default to %sysname%/%tskname%/%valname%
(with no leading front slash).
If OpenHAB and HomeAssistant are different, we could add a new controller, which will share a lot of code between them, but have their own implementation where they differ. I guess that will be better than having some solution that's sub-optimal for both.
So please try to think of the best topic template for Home Assistant. (and for OpenHAB if the current one could be better)
Related to this topic. Eclipse Smarthome (OpenHab is based on this) gets a shinny new MQTT stack while I'm writing these lines. Some research went into auto-discovery of (IoT) devices and published sensors/actuators. The need for devices to describe themselves and their capabilities is inherent.
To form some sort of standard (convention on top of MQTT), members of the OpenHab community collaborated with the creator of the Homie convention.
It would be very convenient if ESPeasy could speak Homie 3.x and allows easy point and click integration in future OpenHab versions.
Sounds like a good idea. I have to read it first to let the concept shape in my head. The general idea to have a single description of the device you have to allow low level discovery is very good and I hope this description does not place too many limitations on the current (and future) ESPeasy design itself. I will have a look into this.
Too bad you have to read documents like this. An audiobook/podcast would be great, since I'm behind the wheels about 10 - 15 hours/week ;)
Has there been any discussion/progress among the team in implementing one of these standards (Homie or HomeAssistant) within the ESPEasy Controller? @davidgraeff 's newest MQTT binding in openHAB is now live (https://community.openhab.org/t/mqtt-binding-version-2-4/53811/55) and supports auto-discovery of MQTT topics, as long as they're defined in the Homie/HomeAssistant format, as I understand it. It would be excellent if I could convert all of my sensors to use this, rather than having to manually build them into my openHAB configuration...
Do you have an easy to read document on how such an auto-discovery should look like?
This is the official convention that I could find for Homie. There seems to be a lot of topics that need to be populated for auto-discovery, but, I did see an issue (https://github.com/homieiot/convention/issues/45) opened that aims to thin down those topics to a minimal set of required ones.
It looks like well documented and elaborate. Maybe a bit too elaborate at this moment to get an idea on where to start.
So can you give an example, for me to get a feeling about this discovery, to for instance let a switch input be discovered. What should the node do and what settings can be derived and what should be given by the user?
A "in a nutshell" paragraph is missing, indeed.
That's it!
@bkpsu Please refer to the web-page https://homieiot.github.io/ instead of the repository. The repo contains the latest/in-development specification, and is not meant to be used for implementers.
I currently start my home automation new from scratch. The "old" system was based on openHAB 1.8 and mosquitto MQTT server and a lot of openHAB code. I waited a wile until openHAB 2.x settled. Recently I discovered that with Version 2.4 openHAB committed to MQTT by integrating a MQTT server inside as a binding - great!. With some rules and a little mod of the MQTT-import plugin to receive true and false payloads It is possible to get sensors and actors running ESPEasy autodiscovered and working with no line of code, rules or conf files.
I'm planning to implement a controller plugin for homie and/or home assistant. For a "proof of concept" I did a test implementation which works out fine Basically a rule is triggered every minute to propagate the features. In the future this should be sent only on mqtt connect with retained flag set but for a quick test this works.
on System#Boot do
Debug 4
timerSet,1,60
endon
On Rules#Timer=1 do //Update autodiscover
Publish homie/%sysname%/$homie,3.0.0
Publish homie/%sysname%/$name,%sysname%
Publish homie/%sysname%/$localip,%ip%
Publish homie/%sysname%/$mac,%mac% // not working but not used by openhab
Publish homie/%sysname%/$fw/name,ESP Easy Mega
Publish homie/%sysname%/$fw/version,2.0.0
Publish homie/%sysname%/$imlementation,ESP8266
Publish homie/%sysname%/$stats/interval,90 // publish a longer interval helps to avoid timeout messages
Publish homie/%sysname%/$nodes,Durchfluss,switchValve // to devices, one sensor one actor
Publish homie/%sysname%/Durchfluss/$name,Durchflussrate
Publish homie/%sysname%/Durchfluss/$type,Flowcounter
Publish homie/%sysname%/Durchfluss/$properties,Count,Total,Time // sensor offers 3 values
Publish homie/%sysname%/Durchfluss/Count/$name,Zaehler // Value 1
Publish homie/%sysname%/Durchfluss/Count/$datatype,integer
Publish homie/%sysname%/Durchfluss/Count/$unit,#
Publish homie/%sysname%/Durchfluss/Total/$name,Gesamt // Value 2
Publish homie/%sysname%/Durchfluss/Total/$datatype,integer
Publish homie/%sysname%/Durchfluss/Total/$unit,#
Publish homie/%sysname%/Durchfluss/Time/$name,Zeit // Value 3
Publish homie/%sysname%/Durchfluss/Time/$unit,sec
Publish homie/%sysname%/Durchfluss/Time/$datatype,integer
Publish homie/%sysname%/switchValve/$name,Stellventil
Publish homie/%sysname%/switchValve/$type,Valve
Publish homie/%sysname%/switchValve/$properties,Valve
Publish homie/%sysname%/switchValve/Valve/$name,Ventil
Publish homie/%sysname%/switchValve/Valve/$settable,true // because it is an actor
Publish homie/%sysname%/switchValve/Valve/$datatype,boolean // needs a modified flow2string function to convert true/false to 1/2
Publish homie/%sysname%/$state,ready //all up and ready
timerSet,1,60
endon
on Valve do // handle the incomming /set msg via MQTT import plugin
TaskValueSet 4,1,%eventvalue% // store value for debug reasons
if [switchValve#Valve]=1
gpio,5,1
Publish homie/%sysname%/switchValve/Valve,true // confirm back to homie
else
gpio,5,0
Publish homie/%sysname%/switchValve/Valve,false
endif
endon
after a few seconds a new thing is detected
So with the little mod of the source code to handle incomming true/false msgs by the MQTT import plugin the relais work. And it is usable after a few clicks.
Now I'm planning to implement this as a controller plugin to do most things automatically.
My "only" concern is adding/deleting devices as the retained flag should be set. There will be the need of a little bit of housekeeping to get rid of unused devices/nodes: With the above rules "New Thing" and "Thing updated" is working inside paper UI.
so with every reboot the header and node informations should be send/updated. If a user do modifications a reboot once he is happy is better than having openHAB handle "half backed" configurations with every change.
Finally I like the home assistant spec a little bit better because of less topics with JSON formatted payload but I first took a look on the homie as recomended by @davidgraeff
It cannot be stressed enough, to consider changing existing MQTT client devices to an MQTT convention like the mentioned Homie 3.x convention. That might not be possible in some cases though.
So my questions are:
BTW: I'm totally new in contributing to an open source project and so please be partitioned.
These look like they can be incorporated in an extension to the existing OpenHab controller. How often do these messages have to be sent to the broker? At least on a change I assume? We can create an event when settings are saved, then a rule can be triggered. Maybe later on also a command to publish the discovery texts, so it can also be queried from OpenHab I guess? Is there an existing format for?
Maybe we must also add something to the plugins themselves, or to a group of plugins. And about the info you may need, just ask here and someone will answer :)
I would not extend the openHAB controller, which should only stay as a legacy controller (or be removed, depending on your backwards compatibility policies).
openHAB is now fully compliant to the MQTT Homie specification and for OH 2.5 we aim to be fully compatible to the HomeAssistant MQTT-Components specification.
A Homie convention controller and a HA-controller is fully sufficient.
I like the home assistant spec a little bit better because of less topics with JSON formatted payload
Homie tries to stay json free, that's true. But HA MQTT has the issue that they are not generic. They have specified a set of components and that's it. If your device has more features or does not fit into one of the components you are out of luck with their specification. That's why I think both conventions should be considered.
openhab should delete all devices/nodes if they are not listed in the $nodes message any more
That is actually happening. But the OH core does not remove a once-seen thing or channel, no matter what the binding requests. There is an issue on our issue tracker and also a PR. Might be fixed for OH 2.5.
I agree to start with a new "homie" controller and perhaps a extra homeassistant controller later when openHAB 2.5 is ready. Looking into homeassistants definition I agree that there are some limitations.
as tools like mqtt-forget are available and openhab ignores old $nodes there should be no problem with "housekeeping"
Hello
@davidgraeff is it possible that the build in mqtt server do not set the retain flag as expected? I spend a few hours poking around in the ESPeasy source to find out why mqttFX don't show the retained flag until i found that messages I send by mqttFX with the retained flag set also not show up as retained. Sending the same message to my "old" mosquitto server work as expected,
openHAB server:
mosquitto server:
Retained flag always set:
or doese the build in mqtt server recognizes the retained flag and "consumes" it?
The mqtt specification allows MQTT servers to ignore the "retained" flag if QoS is 0. Might that be your problem?
The question is if "retained" is nessesary for autodiscover? And if so should I use a higher QoS? The homie convention says that most messages should be sent with retain flag set. Or can I simply ignore that the retain flag here. I plan to send all the information only once after reboot.
Retained is not necessary and only affects how the server processes and stores messages. openHAB will use whatever message it receives.
Fine, than I‘ll not wasting my time on this any further.
I finished 1st draft version of the homie protocol/controller plugin (I called it c014 for now if it's ok)
currently the basic discovery for sensors is implemented. Incomming messages are the next task.
I ran into several issues:
Every comments and help welcome. I committed the plugin and some changes in the framework to my repro.
@Christian-Me : Could you please link your repo here (with the correct branch preselected)? That saves people time, who are willing to have a look.
I also suggest that you create small PRs for required framework changes separately and try to get those in first. Your repo/branch should at the end only contain your new controller.
For now it is only possible to have 1 MQTT controller active in ESPeasy. So please be aware of that when testing yourself. This is still a legacy decision and I plan on removing that limitation, but that needs a lot of code changes.
About the queue system used for controllers. Each controller has its own delay queue system and I advice you have a short look at it to understand the idea behind it. The main purpose for writing this queue is to make sure handling messages is not stalling other tasks on the ESP. The MQTT controllers all use the same queue mechanism, which is also the simplest of all, since PubSub client is also using some buffer. But still that buffer is limited, so the MQTT delay queue is making ESPeasy behave a lot faster with a lot less delays since the PubSubClient buffer is quite small.
For testing purposes you can increase the limits of those queues, but keep in mind they tend to use a lot of memory if they retain the entire MQTT message. See also controller documentation about the available parameters for that buffer. If you need to extend these, please let me know, so other controllers can benefit from it too. But please don't try to do things different from other controllers, unless it is really needed. I've been trying very hard to make the code for all controllers look as much the same as possible so we can move stuff into separate helpers instead of the severe code duplication it was in the past.
If you need to process the items in the queue, please have a look at the function scheduleNextMQTTdelayQueue
and processMQTTdelayQueue
.
As can be seen there, the scheduler is having an entry for handling the MQTT delay queue and processing incoming messages.
For example by calling setIntervalTimerOverride(TIMER_MQTT, 10);
you will schedule to handle the MQTT flow at 10 msec from now. (don't set it to '0' or else you may cause lots of other scheduled tasks to become delayed)
By setting that timer, it will schedule a task to run runPeriodicalMQTT()
, which does also call the loop()
function of PubSubClient.
That loop function does the actual MQTT work.
So please try to use the existing MQTT functions as much as possible and if there's something missing, don't try to interact directly with PubSubClient, but try to find a way to use or extend the code in ESPeasy. If you're trying to interact directly with PubSubClient, you may cause other issues when someone is running another setup than the one you've been testing with.
Also note that we're not using the same code as on the official PubSubClient Github. I've made a few fixes myself and combined ideas of others that made pull requests to that repo. Currently there is a newer version than the one our lib is based on, but I didn't have time to check the differences.
you can find my 1st attempt here complete fork or the controller plugin directly here: _C014.ino
The autodiscover information for my test setup are more than 20 topics (10-12 for the system information, 3 for every device and 2 for every value) so expanding the queue is not an option I think. I used the following code for each topic (example for the system name):
// $name Device → Controller Friendly name of the device Yes Yes
tmppubname = pubname;
tmppubname.replace(F("#"), "$name");
log = F("C014: $name: ");
log += Settings.Name;
if (MQTTpublish(event->ControllerIndex, tmppubname.c_str(), Settings.Name, 1)) {addLog(LOG_LEVEL_INFO, log+" success!");}
else addLog(LOG_LEVEL_ERROR, log+" ERROR!");
processMQTTdelayQueue();
This works fine in most cases.
The plugin will need 4 new events
// new messages for autodiscover plugins (experimental)
#define CPLUGIN_CONNECTED 52 // call after connected to mqtt server to publich device autodicover features
#define CPLUGIN_FALLING_ASLEEP 53 // call before going to sleep
#define CPLUGIN_AWAKE 54 // call after wakeup (not sending autodicover feed)
#define CPLUGIN_SEND_STATS 55 // call every interval loop
I'm not sure were to place the function calls.
CPLUGIN_CONNECTED should be called after the MQTT connection is established but not after deep sleep. Currently placed in bool MQTTCheck(int controller_idx)
CPLUGIN_AWAKE should be called instead after deep sleep
CPLUGIN_FALLING_ASLEEP should be called before deep sleep
CPLUGIN_SEND_STATS should be called periodical Currently placed inside void runEach30Seconds()
Well at first I would suggest to make a separate function to show the extra plugin functions needed. Then we can call those later from the right points. In the loop function in ESPEasy.ino there is a function which is called right before going to sleep. You can add a call to this function from there. Later we will move these calls to the right place if we know more about their use. (it is now still in flux)
About the queue for all those topics. Do you think we can define a fixed set of topics? Then you could make some enum to keep these along with the values you try to send. Then we only store an enum and the values and generate the whole strings from the queue again as soon as they are being read from the queue. That's also being done in some other controller queues. Only store what is needed to recreate the message when needed.
Edit: I guess those topics are depending on the plugin, right? Then the right place would eventually be to ask the plugin for the topic. But if it can be done via some helper to create generic topics for a type of sensor value, then we would have less code duplication. So that means we should query the plugin for a specific type of value it is trying to send. (e.g. temp/humidity/distance/time/CO2/etc)
The plan is to create a controller plugin to be as close as possible to the homie convention without changing code inside plugins and as less as possible in the firmware. In the beginning I thought I could uses the INIT message but this is to early in the boot process.
All required messages for sensors can be acquired without any change in plugins leaving the unit open. For actors a final solution is open. I think for test reasons i will "had code" LCD/OLED and simple NEOPixel because I have them here. Perhaps the two motor plugins too because I have both somewhere in my desk. Most interesting could be GPIOs for Relais. They can be handled by the controller directly catching incoming messages like homie/systemName/GPIO/Number/set 0 or 1 so no extra plugin necessary. Problem is how to determine when they should be in use. Perhaps if the set in the hardware tab to anything else than default or input. When set as input they can be estimated as buttons (homie/systeminfo/GPIO/Number without the settable flag set and handled by the controller.
Can you point me to an good example controller how a enum could be implemented?
Every topic acording to this scheme homie/%deviceName%/%nodeName%/%valueName% ie. homie/sensorDevice1/BMP280/Temperature
Sending Values work "of the shelf" by sending to "homie/%sysname%/%tskname%/%valname%" Receiving works by appending /set to set a value. i.e. "homie/motorDevice/motorController1/Speed/set". It is expected that the value will be set and sent back as a confirmation.
It would be good that beside the value name a value Unit can be selected currently °C Degree Celsius, °F Degree Fahrenheit, ° Degree, L Liter, gal Galon, V Volts, W Watt, A Ampere, % Percent, m Meter, ft Feet, Pa Pascal, psi PSI, # Count or Amount. I can thing about some more Hz, Ah, Wh, VA, LUX, .... Perhaps a set of fixed Units define by a byte can be enough. If I look in openHAB there are many more units defined.
Hey this started looking like a quick hack and going to be a real task, Its the first time in contribute to an opensource project ...
@Christian-Me You should have a look at Homie 4. This is what unofficially a lot of controllers have implemented. Have a look at the home page example:
homie / device123 / $homie → 3.0
homie / device123 / $name → My device
homie / device123 / $state → ready
homie / device123 / $nodes → my_thermostat
homie / device123 / my_thermostat / $name → My thermostat
homie / device123 / my_thermostat / $properties → temperature
homie / device123 / my_thermostat / temperature → 22
homie / device123 / my_thermostat / temperature / $name → Temperature
homie / device123 / my_thermostat / temperature / $unit → °C
homie / device123 / my_thermostat / temperature / $datatype → integer
So it's 4 topics for the device, 2 per node, and 3-4 publishes for a property, including the value. And one subscription per property on the "/set" topic. A LWT topic MUST be published for the $state. That's all.
Don't care about the $stats part. That is obsolete and not really thought trough. And $settable is optional in Homie 4, defaulting to false.
That's fine but still too many to fill the queue when I think of my current project - my garden water supply system (flow counter, 2 valve coils, 1 motor valve, 2 pressure sensors, 1 relay for the pump, a LCD or OLED display). I was not spending time with the deep sleep issue, I personal don't think ESPeasy is the best firmware for battery devices. Will check how to determine if the system start is caused by wakeup to avoid sending all info again every deep sleep cycle and wasting time and battery life.
Value changes are already working out of the box as I wrote in my first post, fine.
So next is to work on the actors, I think I go for the homie/systemName/gpio/pinX/set 0 or 1 if the default is set to low or high and as imput if it is set to input. For now without configurable Names. Here I will need the $settable flag, it worked with my proof of concept. The node showed up in openhab as a switch and via the MQTT import plugin I was able to handle the /set messages (after the true/false 1/0 mod)
$stats are already working good with uptime and quality. That's what I usually configure for all of my ESPs. $state ALERT can be a nice addon in the future when I think of the failed underwater sensors in my pond last year.
BTW were can I find Homie 4?
But still now idea of the best way to work with units.
I think having units defined in a plugin is also a great addition. Also because it could allow for some very simple transformations like C => F etc.
We will have to find a good standard for providing a units list for the plugins.
Just wondering how to pass values form the controller to a plugin. i.e LCD Screen here is an example:
topic=homie/my_esp1/lcd/line1/set
payload=send from openHAB
or
topic=homie/my_esp1/5/line1/set
payload=send from openHAB
if the LCD is device Number 5.
so I have the name of the device given by the user (not the number, but this could be possible too as the human readable name can be given by the $name parameter) lcd
and the value line1
What is the best function to call or task structure to modify?
As mentioned before I'm still looking for a good way to find out which "settable" values a plugin provides. I go through a few known plugins like LCD or OLED but this means to update the controller code for every new or modified plugin and makes the code of the controller bigger every time. Also I could not find a way to determine features of the plugin like if it's a 2 or 4 line LCD display.
For simple GPIO tasks (true/false) I use the GPIO,<pin>,<0/1>
like if a command is received.
At this moment we only have the plugin MQTTimport, which can gather data from MQTT and present it just like a normal plugin would. So that's currently limited to up-to 4 float values. Right now, ESPeasy does not have an internal representation schema for other types than numerical values. Also the need for writing stuff (in a command for example) to a display is a much requested feature. It exists for the OLED plugin, but not the "Framed OLED" one. (LCD plugin also has its own write function) But right now we don't have a command in the rules to fetch data from a broker, so that's also a bit limited.
One of the things I would like to add in the (near) future, is a specific "display controller", which can have a number of spots on a display to show information.
Each spot will then be marked with either a configurable "IDX" value (is already present for the plugin and is used by Domoticz but can be re-used here too) or a marker like how it is used in the rules to access a variable. (e.g. [bme#T]
for the variable T in sensor called "bme", like the BME280)
This will make addressing displays much more uniform, since it is already a bit messy right now with the different displays. Also a display is not actually a plugin, since it is only accepting data, not producing data.
The mqtt specification allows MQTT servers to ignore the "retained" flag if QoS is 0. Might that be your problem?
The retain flag is still an issue. The pubsubclient lib used in ESPEasy does not support qos 1 or 2 (as far as i found out). When I send messages created by my controller plugin with qos 2 via mqttSpy all seams perfect in openHAB. Validation over the homie website says everything is should be ok! When ESPEasy is sending the messages only once on boot with qos0 and retain=true the device will be detected in PaperUI. When you add it you will get an error that the /$homie message is missing (but it was send before). After second reboot it says online but will not recognize the nodes ... It worked in my "proof of concept" above because the rule was sending all messages every minute ...
In my tests I used homie 4.0.0.
I do not know how to get ESPEasy to send retained messages with qos above 0. So very disappointing. It seams that openHAB need that the messages are "properly" send and retained for the multi step autodiscover process.
Any ideas? BTW mosquitto retains the messages even with qos=0. Perhaps I use mosquitto for now (does autodiscover works with an external broker too?)
moquette can be configured to also retain messages with qos=0 but I didn't do that as the current state is also standard conform.
I made also a few fixes just recently for the entire mqtt binding. The homeassistant autodiscovery also works with most devices now. This doesn't help with openHAB 2.4 of course.
I totally understand to stay on agreed standards that everybody can rely on.
So the only paths I currently see is either to get the pubsubclient to send qos>0 (1 or 2, what is actually needed for moquette?) or trying to use mosquitto for now. I do not want to go the homeassiastant path as to the limitations as @davidgraeff mentioned above.
A future way could be how openHAB and the build in mqtt broker/client handles the multi step registration process of a new thing. For me looking from the outside it looks like it relies on the fact that the broker retains the messages and publish them again every registration step (adding a thing, adding an item ...). Looking at the discussions of several pull requests for the pubsubclient trying to add qos>0 I even understand why this isn't implemented due to memory restrictions on microcontrollers or even on capable micros like the ESP running a complex firmware like ESPEasy.
Another Idea I just got a expansion of the homie standard. Instead of sending all information at once perhaps it is useful sending only basic information in the on boot (i.e. system name and list of device names) and sending detailed information of the device (with the list of values) and later the single values on request. This would keep the traffic caused by for already discovered items low.
Hope I cam find some time to test the mosquito solution soon.
later the single values on request
That's the point. MQTT is a pub/sub protocol, you can't request a value. In Homie we must therefore retain everything. There is also no start/finish signaling, so a controller must guess when it is allowed to actually create something out of the many received topics. Other conventions like HomeAssistant therefore have chosen json to package complex values into one topic, which is suboptimal, because they need to send the entire structure again if a single value changes.
For those reasons I think that MQTT is not the best solution for IoT and small embeddeds. I prefer CoAP (REST+subscriptions via UDP basically).
As controller and device is sending and receiving perhaps a simple two way communication could be possible. My idea for a multi step auto-discover can look like this:
On boot a system/ESP8266 (Name: homie-3) send for example this four messages:
homie/homie-3/$homie 3.0.0
homie/homie-3/$name homie-3
homie/homie-3/$nodes SYSTEM,mySystemInfo,myNeoPixel,myLCD,myCounter,myAnalogInput,myDummy
homie/homie-3/$state ready
So the controller/openHAB can recognize the system with all it's nodes/devices and can recognize changes on the devices if the known name change, appear or disappear.
then the controller can ask for the features of a node/device when necessary
by sending to homie/homie-3/$nodes/get
topic the name or list of names for nodes the controller whats more information.
homie/homie-3/$nodes/get SYSTEM
Then the system can send the detailed information:
reply
homie/homie-3/SYSTEM/$name SYSTEM
homie/homie-3/SYSTEM/$properties cmd,gpio4,gpio5
and then the controller can ask for the values
homie/homie-3/SYSTEM/$properties/get cmd,gpio5
(as an example gpio4 is not relevant)
reply
homie/homie-3/SYSTEM/cmd/$name Command
homie/homie-3/SYSTEM/cmd/$settable true
homie/homie-3/SYSTEM/gpio5/$name gpio5
homie/homie-3/SYSTEM/gpio5/$settable true
if the query of every single value is to complicated or not suitable (like in openHAB) then perhaps like this
homie/homie-3/$nodes/getAll SYSTEM
reply
homie/homie-3/SYSTEM/$name SYSTEM
homie/homie-3/SYSTEM/cmd/$name Command
homie/homie-3/SYSTEM/cmd/$settable true
homie/homie-3/SYSTEM/gpio4/$name gpio4
homie/homie-3/SYSTEM/gpio5/$name gpio5
homie/homie-3/SYSTEM/gpio5/$settable true
homie/homie-3/SYSTEM/$properties cmd,gpio4,gpio5
There could be some "keywords" defined like <ALL>
to get everything in one go instead by name or list of names. A simple checksum response could make sure that all messages are transferred as this should work on qos=0;
Only an idea, likely it is far to complicated and not fail proof! The goal is to reduce the permanent traffic, the amount of messages send in one go, the length of messages and to avoid outdated retained messages on the broker. I agree with your opinion that MQTT is not the best but one reason that it is so popular especially in the maker/diy scene is that MQTT is so simple and so easy to debug and seams to be quite fail proof. But I'm not an expert in this field.
Please suggest homie changes in the homie repo. This repo is about espEasy.
Your idea replicates a client/server architecture but mqtt is a pub/sub one. You probably have heard of the saying, "if all you have is a hammer, everything looks like a nail". Mqtt might just not be the right tool/protocol for 1:1 connections.
There are enough coap libraries out there and openHAB will soon also support an mqtt like binding for coap.
one reason that it is so popular especially in the maker/diy scene is that MQTT is so simple
That's not the reason. Someone just started to use mqtt, probably because Eclipse has open source mqtt libraries for all kind of languages since ever. If there would be one good coap library on platform.io for Arduino and the esps and controllers like openHAB start to pick coap up, that trend might as well change.
I agree with all above.
To “nail it down” or “make a long story short”: Due to the fact that homie relies on retained messages and ESPeasy is (currently) not able to send retained messages in a standardized way (qos>0) it is currently not possible to fulfill the homie standard with ESPEasy. Right?
Every other attempt like “hammering” the data out every minute will be against accepted standards and what not.
So I will try to get ESPEasy to send qos>0, even for my private fun or forget all my time I spend up to now.
So I will try to get ESPEasy to send qos>0, even for my private fun or forget all my time I spend up to now.
You would "violate" the homie spec, but you can as well use mosquitto and retain messages with qos 0 with your devices. That's not an issue. If you find the right knob for moquette, you can also add a patch PR for openHAB to accept qos 0 retained messages.
Just to have it all clear, what exactly is the problem with QoS levels > 0? And what has to be changed to support it and what are the most likely reasons it currently isn't supported?
I am not an expert on MQTT, but I do know the internals of ESPeasy, so I could maybe help thinking of a way how things can be done in ESPeasy if the "hammer" does not fit the nails we have here :)
@TD-er QoS has 3 Levels @davidgraeff please correct me if I'm wrong
0: fire and forget 1: make sure that the message arrives minimum once 2: make shure that a message arrives once (and only once)
The pubsubclient lib do only support sending with QoS 0 see limitations. This is done by purpose because to make QoS > 0 you have to store the topic and message and id to react on the confirmation send(or not) by the broker to resend in case of a problem. This eat up memory. More Details here ...
retain means that the flaged message is stored by the broker and will be sent to every subscriber as soon as they subscribe to a topic. The Homie convention demand most messages for autodiscover as to be flagged as retained. This make sense that ESPEasy has to send them only once (during boot) and they can be found by a client later or as often as needed.
It seams like a broker not necessarily "needs" to store messages flagged as retained if the QoS flag is 0.
There are many discussions if qos for a local network based on TCP/IP needs QoS>0 because the undelaying TCP/IP should take care that a message arrives at the broker by itself. I found pull requests for the pubsubclient to send messages with QoS 1 or 2 but not taking care of resending so not storing messages. That is in my opinion not "correct". Qos 0 is perfect for sensor data send in a permanent interval like ESPEasy (contoller can react if the messages don't arrive every timeframe), perhaps problematic if you only send data on changes (controller can have outdated data for a long time) and for actors (i.e. heater do not start because the ON message never arrived). Homie takes care of the last scenario because the device needs a response for every /set message.
All fine and dandy so far. The problem here is that the broker have to store the autodiscover data for the discovery process. That's totally ok because the controller not necessarily is subscribed when the device send the data. A little bit difficult if a device goes offline for ever or even is renamed by the user. Then the old data stays on the broker and must be deleted by "hand" or with a tool like mqtt-forget. Therefor the disconnected message is defined in homie.
I personally don't get why a broker should not store a messages sent with retained flag even when it was sent with qos = 0, perhaps somebody can explain.
@davidgraeff can you please explain why you think I violate homie when I'm able to get ESPEasy to send the retained messages with QoS > 0? I worked with QoS=2 messages on the STM32 platform. It could be painful but is doable.
you violate if you send with qos=0 as home requires qos>0
OK, I think I get it with these QoS levels 0 and 1 (2 was already too complex for me at this time of day ;) )
I understand why PubSubClient does not want to be held responsible for resending when a QoS level 1 message was not acknowledged. But if it is only needed for these autodiscovery things, then it is not really taking up much memory resources on the ESPeasy side. So if we would patch PubSubClient to accept QoS=1, then we must also implement a way to communicate back to see the message was sent. This means we must keep track of the message ID when sending a QoS=1 message, since PubSubClient should not have to.
Pseudo code:
The first way of communicating back such an acknowledgement could be some internal subscribe with a special acknowledge topic. Also you must be able to mark messages with QoS and DUP flag, since you don't want PubSubClient to be responsible for it.
@TD-er, I totally agree on your post. This gives ESPEasy the possibility to notice that the send messages don't arrive @ the broker even if the underlying protocol still works. In my opinion a rare situation because the TCP stack will spit out errors before. ESPeasy can send "alarm" to $state but in this case it is more than unlikely that this will reach the broker. Then the LWT can kick in.
But what I don't get (perhaps it was too late yesterday and still too early now is @davidgraeff eplanation why openHAB 2.4 build in broker not keep messages with retained flag set if they ares sent with QoS 0 and doing so should "violate" the Homie convention)
Point 2.3 of the homie convention says:
QoS and retained messages The nature of the Homie convention makes it safe about duplicate messages, so the recommended QoS for reliability is QoS 1. All messages MUST be sent as retained, UNLESS stated otherwise.
I read: QoS 1 is recommended (not a MUST) for reliability. Reliability is nice! But as explained /set messages from the controller has to be confirmed with a update of the value itself. All Data send by the device can be controlled by the heartbeat interval (if a message don't reach the controller the device can anyway don't do much about it except notifying locally)
The MQTT standard is quite clear (if I read it correctly and found the correct one (here 5.0 but also 3.1.1) Point 3.3.1.3:
If the Server receives a PUBLISH packet with the RETAIN flag set to 1, and QoS 0 it SHOULD store the new QoS 0 message as the new retained message for that topic, but MAY choose to discard it at any time. If this happens there will be no retained message for that topic.
I read this that the Broker SHOULD store the Message. It is not necessary to sore them in non volatile memory and MAY forget the topic if needed (i.e. outdated or memory issues). In our case it makes sense (for me) that the Broker actually keeps the messages (especially homie/$xxx Messages) because this (can) help to serve the purpose of autodetection. I naively thought the Broker was build into openHAB not only to make it convenient for the end user but perhaps also to serve the special needs of openHAB.
Before messing around with the pubsubclient lib I will first try to go back working with an external broker like mosquitto. Patching around and inside openHAB is far above my capabilities and do not fit into my initial approach to do a "as clean as possible, well documented, close to standard and easy to update" start from scratch for my private home automation system (running for more than two years on openHAB 1.8 with many ESPs running mostly ESPEasy without major issues)
Little Update: Downloaded recent mosquitto and run with default config (simply running the mosquitto.exe on my windows notebook) and added it to openHAB
Thing shows up after reboot of ESPEasy inside the Inbox
Step (which fails with the build in broker)
Add a Item/value:
Nice, default Naming is looking good
4.And you have a working Relais (old Wemos Relais Shield for testing purpose only)
5.Or showing a value from the System Info plugin as an example
For now I'm happy with that, especially form and "end User" perspective. (Still some task to do but I think the mqtt broker issue is solved for now)
BTW: Perhaps we should either rename the topic of the issue or raise a new one because this has nothing to do with Home Assistant any more.
After getting gpio working I whanted to try to implement other commands, like TaskValueSet or LCD.
I poked around some hours to get a good point to set values for the dummy device. As this plugin do not have a PLUGIN_WRITE function, I cannot find a good way access it`s values. I even cannot set the values of a dummy device via the openhab mqtt controller C005 via the cmd topic. I know I can do it calling a event,rule but I'm looking for a more elegant solution: sending values to the topic homie/%sysname%/dummyDevice/value1/set
I can add a PLUGIN_WRITE function but as it is working via HTTP there should be a better way.
Question: Is there a standard function handling all commands? Perhaps @TD-er has an idea.
EDIT: I implemented a PLUGIN_WRITE function in the dummy plugin and it's working fine. But this morning I woke up with a thought that this will only work with the first appearance of the dummy plugin. That's why TaskValueSet is not specific for the DUMMY plugin (perhaps). So is there a way to run PluginCall(PLUGIN_WRITE, ... ) for a specific Plugin either by Name or by the position in the task/device list?
TaskValueSet indeed needs to know the variable number and the plugin number. Just for reference, see also the Read the Docs documentation
Some of the commands have been "disabled" from outside access. For example the factory reset command should not be open for anyone to access. But I cannot think of a reason why TaskValueSet should not be made accessible for calls from outside.
One way could be to use the MQTT import to generate an event and then call TaskValueSet, but I can imagine it is a bit too much work to do it for numerous topics and values. And you may run out of resources like max rules size, max number of active plugins and perhaps some more. Maybe @Grovkillen can think of some other ways to do it with the current feature set.
The main problem here is probably that ESPeasy was originally designed to be collecting data from a sensor, do stuff with it and get the data out. So given that main design, there is no real proper way of getting the data in. MQTTimport is somewhat of a hack and also some controllers (Domoticz and OpenHAB) have some hooks to get some specific plugin related info from outside ESPeasy.
Thank you for the fast answer, sorry to answer so late, I was busy with many other things. And as usual when developing with one board I ran into some issues with openHAB yesterday which should not be ESPEasy related as MQTTSpy shows correct communication only that suddenly openHAB don't send any messages and not updating values. Have to solve this problem first
I welcome that critical commands should not be able to be sent via WIFI. I used the rules a lot in my current setup. Rules are on reason why I'm a big fan of ESPEasy. I try to handle many tasks locally on the ESP (like garden sprinkler systems) to make sure to stop the sprinkler even if WIFI is down). But my current approach is to avoid "workarounds" if possible.
No problems in finding plugin number, number of the device in the device list and variable number ... Currently if a /set message arrives (for example for a dummy device) all of this is somehow in the topic or can be found by name. So in the end the controller forms the command string i.e. TaskValueSet,6,2,12.5
(that's all done).
I only do not know to which is the best function to pass it to. I thought plugincall() would be best (like GPIO,5,1
or LCD,1,1,Test
works already perfect) but later found that the dummy plugin do not have a "PLUGIN_WRITE" function. That´s understood because there should be multiple dummy devices available and with "PLUGIN_WRITE" this process stops after the first hit, or is there another reason why TaskValueSet cannot be send by i.e. the openHAB MQTT controller like it is possible via http - instead you should pass this via rules.
So I wrote a "PLUGIN_WRITE" function for the dummy plugin that itself checks if it is the correct instance in the list. If not it passes to the next. That is working fine.
I only implement special solutions like GPIO for use cases where controls are available in i.e. openHAB (Switch) without coding.
The homie/device/SYSTEM/cmd/set
topic should be able to receive the same set of commands that can be sent to the ESP by http for any other use cases like gpio pulses or triggering events.
I only want to make sure that I followed the right path.
Well a command could have an extra parameter of course to indicate task ID. The PLUGIN_WRITE should only return success = true when it was able to handle it. If the command was not for that instance, it should not handle it and thus not return true.
So I don't think that's a route to ignore. That leaves the hardest part still open, to think of a proper name for this command. Maybe "DummyValueSet" ?
And if that's working fine, can you please create a pull request for it?
Currently, Home Assistant users use the Openhab MQTT template changing the Controller Subscribe and Publish. As stated in home assistant documentation: https://www.home-assistant.io/components/sensor.mqtt/#get-sensor-value-from-a-device-with-espeasy
Controller Subscribe: home/%sysname%/# (instead of /%sysname%/#) Controller Publish: home/%sysname%/%tskname%/%valname% (instead of /%sysname%/%tskname%/%valname%)
Since this is such a simple change wouldn't be possible to add the template?
Besides naming is just this two lines from _C005.ino: