5.8 KiB
| id | title | sidebar_label |
|---|---|---|
| store-forward-plugin | Store and forward plugin | Store and forward |
About
:::caution This is a work in progress and is not yet available. :::
The Store Forward Plugin is an implementation of a Store and Forward system to enable resilient messaging in the event that a client device is disconnected from the main network.
Because of the increased network traffic for this overhead, it's not adviced to use this if you are duty cycle limited for your airtime usage nor is it adviced to use this for SF12 (Long range but Slow).
Requirements
Initial Requirements:
- Must be installed on a router node.
-
- This is an artificial limitation, but is in place to enforce best practices.
-
- Router nodes are intended to be always online. If this plugin misses any messages, the reliability of the stored messages will be reduced
- Esp32 Processor based device with external PSRAM. (tbeam v1.0 and tbeamv1.1, maybe others)
Implementation timeline
Not necessarily in this order:
UC 1) MVP - automagically forward packets to a client that may have missed packets.
UC 2) Client Interface (Web, Android, Python or iOS when that happens) to optionally request packets be resent. This is to support the case where Router has not detected that the client was away. This is because the router will only know you're away if you've been gone for a period of time but will have no way of knowing if you were offline for a short number of minutes. This will cover the case where you have ducked into a cave or you're swapping out your battery.
UC 3) router sends a periodic “heartbeat” to let the clients know they’re part of the main mesh
UC 4) support for a mesh to have multiple routers that have the store & forward functionality (for redundancy)
UC 5) Support for "long term" delayed messages and "short term" delayed messages. Handle the cases slightly different to improve user expierence. A short term delayed message would be a message that was resent becaue a node was not heard from for <5 minutes. A long term delayed message is a message that has not been delivered in >5 minutes.
UC 6) Eventually we could add a "want_store_and_forward" bit to MeshPacket and that could be nicer than whitelists in this plugin. Initially we'd only set that bit in text messages (and any other plugin messages that can cope with this). This change would be backward wire compatible so can add easily later.
UC 7) Currently the way we allocate messages in the device code is super inefficient. It always allocates the worst case message size. Really we should dynamically allocate just the # of bytes we need. This would allow many more MeshPackets to be kept in RAM.
UC 8) We'll want a "delayed" bit in MeshPacket. This will indicate that the message was not received in real time.
Things to consider
Not all these cases will be initially implemented. It's just a running stream of thoughts to be considered.
Main Mesh Network with Router
The store and forward plugin is intended to be enabled on a router that designates your "main" mesh network.
Store and Forward on Multiple Routers
If multiple routers with the plugin are enabled, they should be able to share their stored database amongst each other. This enable resilliancy from one router going offline.
Fragmented networks - No router
In this case, the mesh network has been fragmented by two client devices leaving the main network.
If two Meshtastic devices walk away from the main mesh, they will be able to message each other but not message the main network. When they return to the main network, they will receive the messages they have missed from the main mesh network.
Fragmented network - With routers
In this case, we have two routers separate by a great distance, each serving multiple devices. One of the routers have gone offline. This has now created two physically seaprated mesh networks using the same channel configuration.
Q: How do we rejoin both fragmented networks? Do we care about messages that were unrouted between fagments?
Identifing Delayed Messages
When a message is replayed for a node, identify the packet as "Delayed". This will indicate that the message was not received in real time.
Router Data Structures
Structure of received messages:
receivedMessages
Port_No
packetID
to
from
rxTimeMsec
data
Structure of nodes and last time we heard from them. This is a record of any packet type.
receivedRecord
From
rxTimeMillis
General Operation for UC1 - automagically forward packets to a client that may have missed packets
On every handled packet
- Record the sender from and the time we heard from that sender into senderRecord.
On every handled packet
- If the packet is a message, save the messsage into receivedMessages
On every handled packet, if we have not heard from that sender in a period of time greater than timeAway, let's assume that they have been away from the network.
- In this case, we will resend them all the messages they have missed since they were gone
Expected problems this implementation
- If the client has been away for less than 5 minutes and has received the previously sent message, the client will gracefully ignore it. This is thanks to PacketHistory::wasSeenRecently in PacketHistory.cpp.
-
- If the client has been away for more than 5 minutes and we resend packets that they have already received, it's possible they will see duplicate messages. This should be unlikely but is still possible.
Designed limitations
The Store and Forward plugin will subscribe to specific packet types and channels and only save those. This will both reduce the amount of data we will need to store and reduce the overhead on the network. Eg: There's no need to replay ACK packets nor is there's no need to replay old location packets.