MQTT call multi time on big message

[phuonglm]

Follow this document of async MQTT library, big message will be split to multi small MQTT message with difference start index. The library will call our handler multi time with payload is small message.

I think the best logic here is our _onMessage handler will check the message length, total length, start index, allocate the buffer to catch the all message before dispatch it to our _mqtt_callbacks.

I don't tested with pubsub lib but since pubsub don't give me any parameter like that so I disabled that code with pubsub library.

[xoseperez]

This would be useful for OTA updates over MQTT since the payload is managed by the Updater class and saved to flash on the fly. Any other use could be problematic.

What use cases, aside from OTA, would benefit from a payload bigger than 1460bytes but not big enough to make the device run out of memory (say 10k)?

[phuonglm]

In my case is HVAC raw ir codes. They can easily had 30-40 bytes data and mean 40*8*2*<numbers of char per raw timer> then mqtt message can reach 2000 bytes.

[abmantis]

@xoseperez I've seen some messages fail to send (the mqtt publish method returns zero) while adding some more HA discovery features, because the messages where big.

[mcspr]

@abmantis This issue is dealing with receiving big messages. As you may have noticed, HA discovery messages are getting stuck on initial connection when there are many of them. See PR #1370 and the related issue.

[mcspr]

long-time-no-see, but this seems to be already handled by the mqtt buffer size setting circa this commit
43c2c41#diff-adf1d7b355d1037ad971dd9bf6ca48bce5d8d033d2f1e77073951c5553a2997c

espurna/code/espurna/mqtt.cpp

Line 1004 in a69a705

static char message[((MQTT_BUFFER_MAX_SIZE + 1) + 31) & -32] = {0};

and the actual message is assembled from multiple onMessage calls

sure, this does not handle binary payloads, but none of the mqtt module funcs actually work with that. that may be another issue, but imo we also want to move away from the multiple-clients paradigm and just use a single one