Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

MSC3489: Sharing streams of location data with history #3489

Open
wants to merge 14 commits into
base: main
Choose a base branch
from
167 changes: 167 additions & 0 deletions proposals/3489-location-streaming.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,167 @@
# MSC3489 - m.beacon: Sharing streams of location data with history

## Problem

A common communication feature is sharing live location data with other users.
Use cases include helping users find each other and asset tracking. Matrix
doesn't currently provide a way of doing this.

Therefore, we'd like to provide a way for users to efficiently advertise the
time-limited real-time location of themselves or arbitrary assets to other
users in a room - while by default preserving the history of the location
data in the room by storing it as relational timeline events.

The rationale for persisting location data is to support the use cases of:

* Publishing things like Strava-style exercise data
* Bridging to things like ADS-B sources for sharing plane flight path data.
* Logistics use cases where you want to track where your fleet of vehicles/ships
has been, or the packages within them
* Supporting scenarios where you're trying to rendezvous with someone/something
which will be aided by seeing their historical whereabouts (e.g. search-and-rescue
operations; enhanced Find-my-iPhone style use cases).

For purely ephemeral location sharing, see
[MSC3672](https://github.com/matrix-org/matrix-doc/pull/3672)

## Proposal

This MSC adds the ability to publish real-time location beacons to Matrix by
building on [MSC3488](https://github.com/matrix-org/matrix-doc/pull/3488)
(m.location: Extending events with location data).

We introduce two types of events to describe beacons: the first, a state event, contains
the metadata about the beacons advertised by a given user: `m.beacon_info.*`.
andybalaam marked this conversation as resolved.
Show resolved Hide resolved
As Matrix doesn't yet have a way to stop other users overwriting an event
other than setting its state_key to be the owner's mxid, we work around this
by letting the final part of the event type be a unique ID for the beacon.
(Ideally we would put the unique ID as its `state_key` and flip a bit on the
event to indicate that it can only ever be written by its owner - we may get
this via MSC3414 encrypted state events).
andybalaam marked this conversation as resolved.
Show resolved Hide resolved

This lets us track an arbitrary number of beacons per user, and avoids beacon
metadata being duplicated as location data is shared.

An example `m.beacon_info.*` event is:

```json5
{
"type": "m.beacon_info.@matthew:matrix.org",
andybalaam marked this conversation as resolved.
Show resolved Hide resolved
"state_key": "@matthew:matrix.org",
"content": {
"m.beacon_info": {
"description": "The Matthew Tracker", // same as an `m.location` description
"timeout": 86400000, // how long from the last event until we consider the beacon inactive in milliseconds
},
"m.ts": 1436829458432, // creation timestamp of the beacon on the client
"m.asset": {
"type": "m.self" // the type of asset being tracked as per MSC3488
}
}
}
```

Separately, we have the actual location data from the beacon in question
as `m.beacon` events forming a reference relationship to the original `m.beacon_info` event.

Storing the location data as references as opposed to in a state event has multiple advantages:
* Location data is easily encrypted (no dependency on MSC3414)
* Doesn't thrash state resolution by storing new location points every time they change
* Paginated history easily accessible through the `relations` endpoint. If history is not wanted, then one can use
data retention policies(e.g. exploding messages) to ensure it does not
andybalaam marked this conversation as resolved.
Show resolved Hide resolved
accumulate unnecessarily.
* Allows other users to request data from a beacon


`m.beacon` events should be sent at a variable rate which is meaningful for
the use case for the asset being tracked. This rate should not be more
frequent than sending an event every 2 seconds. If the asset is not moving,
the location should still be refreshed on a regular basis to convey that
information - e.g. every 30 seconds.

An example `m.beacon` event is:

```json5
{
"type": "m.beacon",
"sender": "@matthew:matrix.org",
"content": {
"m.relates_to": { // from MSC2674: https://github.com/matrix-org/matrix-doc/pull/2674
"rel_type": "m.reference", // from MSC3267: https://github.com/matrix-org/matrix-doc/pull/3267
"event_id": "$beacon_info"
},
"m.location": {
"uri": "geo:51.5008,0.1247;u=35",
"description": "Arbitrary beacon information"
},
"m.ts": 1636829458432,
}
}
```

The `m.location` section of the `m.beacon` event can include an optional
`description` field to provide user facing updates e.g. current address

## Encryption

Location data should be stored end-to-end encrypted for obvious data privacy
reasons - given the beacon data is stored as simple events that will be automatic.

## Alternatives
andybalaam marked this conversation as resolved.
Show resolved Hide resolved
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There is another alternative that appears not to have been discussed yet, although the idea is quite obvious:

Instead of introducing a new event m.beacon that for the most part duplicates an MSC3488 m.location:

  • use m.location
  • when used for location streams (i.e. in this context and MSC3672), require the m.relates_to mixin to be set

Pros compared to the MSC as it currently is:

  • automatic fallback for clients that support static m.location
  • further fallback to m.text

Cons:

  • this could result in clients without live location support having their timelines spammed. while one could still view the movement trail similar to the original intention, this might be considered undesireable


Initially this MSC considered storing location data in a separate state event but
that had multiple downsides:
* No encryption without MSC3414
* Difficult access to historical data, timeline pagination required
* State resolution thrasing on every location update. By storing a state event for every location datapoint,
we put significant load on servers' state management implementations. Current implementations
may not handle this well.

Another option would be using ephemeral data units to broadcast location updates but they
do come with downsides of their own:
* they are not persisted and cannot provide historical data
* there's no per-room API for them
* they are not end to end encrypted
However, a privacy-preserving "never store these" approach is still desirable, hence [MSC3672](https://github.com/matrix-org/matrix-doc/pull/3672)

Alternatively, we could behave like MSC3401 and announce users' beacons in
`m.beacon_info.*` (similar to MSC3401's `m.call`), and then send location
data via to-device messages to interested devices.
* Pros:
* Doesn't thrash state resolution by storing new location points every time they change
* Gets you easy E2EE automatically
ara4n marked this conversation as resolved.
Show resolved Hide resolved
* Cons:
* designs out historical geo-data use cases
* means you are reinventing the pubsub fan-out routing you get for free with Matrix events
* means you have to reinvent decentralised ACLs to check that the subscribing users are valid
* means new joiners to a room won't even be able to see the most recent location for someone
* MSC1763 data retention lets us control the historical data anyway.

Alternatively, we could send location data as normal E2EE messages. However,
clients would have to back-paginate in an unbounded fashion to confidently
say what the various beacons are doing rather than simply fishing it out of
room state.
andybalaam marked this conversation as resolved.
Show resolved Hide resolved

Alternatively, we could negotiate a WebRTC data channel using MSC3401 and
stream low-latency geospatial data over the participating devices in the room.
This doesn't give us historical data, however, and requiring a WebRTC stack
andybalaam marked this conversation as resolved.
Show resolved Hide resolved
is prohibitively complicated for simple clients (e.g. basic IOT devices
reporting spatial telemetry).
andybalaam marked this conversation as resolved.
Show resolved Hide resolved

We could include velocity data as well as displacement data here(especially as
the W3C geolocation API provides it); this has been left for a future MSC in
the interests of avoiding scope creep.

andybalaam marked this conversation as resolved.
Show resolved Hide resolved
## Security considerations

Location data is high risk for real-world abuse, especially if stored
unencrypted or persisted. The same security considerations apply as for
`m.location` data in general.

## Unstable prefix

* `m.beacon_info.*` should be referred to as `org.matrix.msc3489.beacon_info.*` until this MSC lands.
* `m.beacon` should be referred to as `org.matrix.msc3489.beacon` until this MSC lands.
* `m.location` should be referred to as `org.matrix.msc3488.location.*` until MSC3488 lands.
* `m.ts` should be referred to as `org.matrix.msc3488.ts.*` until MSC3488 lands.
* `m.asset` should be referred to as `org.matrix.msc3488.asset.*` until MSC3488 lands.