a97b8eafd4
* Use a fork of pq which supports userCurrent on wasm * Use sqlite3_js driver when running in JS * Add cmd/dendritejs to pull in sqlite3_js driver for wasm only * Update to latest go-sqlite-js version * Replace prometheus with a stub. sigh * Hard-code a config and don't use opentracing * Latest go-sqlite3-js version * Generate a key for now * Listen for fetch traffic rather than HTTP * Latest hacks for js * libp2p support * More libp2p * Fork gjson to allow us to enforce auth checks as before Previously, all events would come down redacted because the hash checks would fail. They would fail because sjson.DeleteBytes didn't remove keys not used for hashing. This didn't work because of a build tag which included a file which no-oped the index returned. See https://github.com/tidwall/gjson/issues/157 When it's resolved, let's go back to mainline. * Use gjson@1.6.0 as it fixes https://github.com/tidwall/gjson/issues/157 * Use latest gomatrixserverlib for sig checks * Fix a bug which could cause exclude_from_sync to not be set Caused when sending events over federation. * Use query variadic to make lookups actually work! * Latest gomatrixserverlib * Add notes on getting p2p up and running Partly so I don't forget myself! * refactor: Move p2p specific stuff to cmd/dendritejs This is important or else the normal build of dendrite will fail because the p2p libraries depend on syscall/js which doesn't work on normal builds. Also, clean up main.go to read a bit better. * Update ho-http-js-libp2p to return errors from RoundTrip * Add an LRU cache around the key DB We actually need this for P2P because otherwise we can *segfault* with things like: "runtime: unexpected return pc for runtime.handleEvent" where the event is a `syscall/js` event, caused by spamming sql.js caused by "Checking event signatures for 14 events of room state" which hammers the key DB repeatedly in quick succession. Using a cache fixes this, though the underlying cause is probably a bug in the version of Go I'm on (1.13.7) * breaking: Add Tracing.Enabled to toggle whether we do opentracing Defaults to false, which is why this is a breaking change. We need this flag because WASM builds cannot do opentracing. * Start adding conditional builds for wasm to handle lib/pq The general idea here is to have the wasm build have a `NewXXXDatabase` that doesn't import any postgres package and hence we never import `lib/pq`, which doesn't work under WASM (undefined `userCurrent`). * Remove lib/pq for wasm for syncapi * Add conditional building to remaining storage APIs * Update build script to set env vars correctly for dendritejs * sqlite bug fixes * Docs * Add a no-op main for dendritejs when not building under wasm * Use the real prometheus, even for WASM Instead, the dendrite-sw.js must mock out `process.pid` and `fs.stat` - which must invoke the callback with an error (e.g `EINVAL`) in order for it to work: ``` global.process = { pid: 1, }; global.fs.stat = function(path, cb) { cb({ code: "EINVAL", }); } ``` * Linting |
||
---|---|---|
.. | ||
api | ||
consumers | ||
routing | ||
storage | ||
sync | ||
types | ||
README.md | ||
syncapi.go |
Sync API Server
This server is responsible for servicing /sync
requests. It gets its data from the room server output log. Currently, the sync server will:
- Return a valid
/sync
response for the user represented by the providedaccess_token
. - Return a "complete sync" if no
since
value is provided, and return a validnext_batch
token. This contains all rooms the user has been invited to or has joined. For joined rooms, this includes the complete current room state and the most recent 20 (hard-coded) events in the timeline. - For "incremental syncs" (a
since
value is provided), as you get invited to, join, or leave rooms they will be reflected correctly in the/sync
response. - For very large state deltas, the
state
section of a room is correctly populated with the state of the room at the start of the timeline. - When you join a room, the
/sync
which transitions your client to be "joined" will include the complete current room state as per the specification. - Only wake up user streams it needs to wake up.
- Honours the
timeout
query parameter value.
Internals
When the server gets a /sync
request, it needs to:
- Work out which rooms to return to the client.
- For each room, work out which events to return to the client.
The logic for working out which rooms is based on Synapse:
- Get the CURRENT joined room list for this user.
- Get membership list changes for this user between the provided stream position and now.
- For each room which has membership list changes:
- Check if the room is 'newly joined' (insufficient to just check for a join event because we allow dupe joins). If it is, then we need to send the full room state down (and 'limited' is always true).
- Check if user is still CURRENTLY invited to the room. If so, add room to 'invited' block.
- Check if the user is CURRENTLY left/banned. If so, add room to 'archived' block.
- Add joined rooms (joined room list)
For each room, the /sync response returns the most recent timeline events and the state of the room at the start of the timeline.
The logic for working out which events is not based entirely on Synapse code, as it is known broken with respect to working out
room state. In order to know which events to return, the server needs to calculate room state at various points in the history of
the room. For example, imagine a room with the following 15 events (letters are state events (updated via '
), numbers are timeline events):
index 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 (1-based indexing as StreamPosition(0) represents no event)
timeline [A, B, C, D, 1, 2, 3, D', 4, D'', 5, B', D''', D'''', 6]
The current state of this room is: [A, B', C, D'''']
.
If this room was requested with ?since=14&limit=5
then 1 timeline event would be returned, the most recent one:
15
[ 6 ]
If this room was requested with ?since=9&limit=5
then 5 timeline events would be returned, the most recent ones:
11 12 13 14 15
[5, B', D''', D'''', 6]
The state of the room at the START of the timeline can be represented in 2 ways:
- The
full_state
from index 0 :[A, B, C, D'']
(aka the state between 0-11 exclusive) - A partial state from index 9 :
[D'']
(aka the state between 9-11 exclusive)
Servers advance state events (e.g from D'
to D''
) based on the state conflict resolution algorithm.
You might think that you could advance the current state by just updating the entry for the (event type, state_key)
tuple
for each state event, but this state can diverge from the state calculated using the state conflict resolution algorithm.
For example, if there are two "simultaneous" updates to the same state key, that is two updates at the same depth in the
event graph, then the final result of the state conflict resolution algorithm might not match the order the events appear
in the timeline.
The correct advancement for state events is represented by the AddsStateEventIDs
and RemovesStateEventIDs
that
are in OutputRoomEvents
from the room server.
This version of the sync server uses very simple indexing to calculate room state at various points.
This is inefficient when a very old since
value is provided, or the full_state
is requested, as the state delta becomes
very large. This is mitigated slightly with indexes, but better data structures could be used in the future.
Known Issues
m.room.history_visibility
is not honoured: it is always treated as "shared".- All ephemeral events are not implemented (presence, typing, receipts).
- Account data (both user and room) is not implemented.
to_device
messages are not implemented.- Back-pagination via
prev_batch
is not implemented. - The
limited
flag can lie. - Filters are not honoured or implemented. The
limit
for each room is hard-coded to 20. - The
full_state
query parameter is not implemented. - The
set_presence
query parameter is not implemented. - "Ignored" users are not ignored.
- Redacted events are still sent to clients.
- Invites over federation (if it existed) won't work as they aren't "real" events and so won't be in the right tables.
invite_state
is not implemented (for similar reasons to the above point).- The current implementation scales badly when a very old
since
token is provided. - The entire current room state can be re-sent to the client if they send a duplicate "join" event which should be a no-op.