Initial /sync when filtered to a single room can be very slow

[squahtx]

Description

The /rooms/{roomId}/initialSync endpoint is deprecated and the documentation suggests using /sync instead.

However, using /sync with a room filter has drastically worse performance for users in very many rooms,
eg. taking up to 400 seconds for a user in 2,200 rooms and pulling 100,000+ events from the database.

Steps to reproduce

GET /_matrix/client/r0/sync?filter=%7B%22room%22:+%7B%22rooms%22:+%5B%22!XXXXXXXXXXXXXXXXXX:matrix.org%22%5D,+%22state%22:+%7B%22rooms%22:+%5B%22!XXXXXXXXXXXXXXXXXX:matrix.org%22%5D%7D,+%22timeline%22:+%7B%22rooms%22:+%5B%22!XXXXXXXXXXXXXXXXXX:matrix.org%22%5D,+%22senders%22:+%5B%22@xxxxxxxxxx:matrix.org%22%5D%7D,+%22ephemeral%22:+%7B%22limit%22:+0,+%22types%22:+%5B%5D%7D,+%22account_data%22:+%7B%22limit%22:+0,+%22types%22:+%5B%5D%7D%7D,+%22presence%22:+%7B%22limit%22:+0,+%22types%22:+%5B%5D%7D,+%22account_data%22:+%7B%22limit%22:+0,+%22types%22:+%5B%5D%7D%7D

Version information

• Homeserver: matrix.org

[ShadowJonathan]

Sync v3 looks to be addressing this, by adding filters and scoped syncing. Though my knowledge of the draft stops beyond that point.

Maybe this could be good feedback on the sync-v3 efforts then, though.

[reivilibre]

I believe this is just an implementation detail that /initialSync is optimised better than this for /sync and don't think it's specific to sync v2

[clokep]

Note that the filter given in the first comment is:

{
  "room": {
    "rooms": [
      "!XXXXXXXXXXXXXXXXXX:matrix.org"
    ],
    "state": {
      "rooms": [
        "!XXXXXXXXXXXXXXXXXX:matrix.org"
      ]
    },
    "timeline": {
      "rooms": [
        "!XXXXXXXXXXXXXXXXXX:matrix.org"
      ],
      "senders": [
        "@xxxxxxxxxx:matrix.org"
      ]
    },
    "ephemeral": {
      "limit": 0,
      "types": []
    },
    "account_data": {
      "limit": 0,
      "types": []
    }
  },
  "presence": {
    "limit": 0,
    "types": []
  },
  "account_data": {
    "limit": 0,
    "types": []
  }
}