Synapse: garbage collect forgotten rooms

[ara4n]

if everyone on a server has left and forgotten a given room, we should GC it from the DB

[richvdh]

I'm surprised we don't already have this, but I can't find it if we do.

Related:

• #1307
• #2338
• https://github.com/vector-im/riot-web/issues/6978

[aaronraimist]

This should probably get a Mozilla label https://bugzilla.mozilla.org/show_bug.cgi?id=1622403

[schildbach]

This seems like no-brainer to me. I just deleted gigabytes of data from my postgres DB just by manually removing abandoned rooms.

[peterhoeg]

Do you have a (relatively) safe script to do that @schildbach ?

[ptman]

@peterhoeg you can get a list of rooms and remove them via the admin api: https://github.com/matrix-org/synapse/blob/develop/docs/admin_api/rooms.md

Directly modifying the database is error prone and should not be done.

Also synapse admin ui can let you click around instead of writing code for the API: https://github.com/Awesome-Technologies/synapse-admin

[schildbach]

@peterhoeg I used the instructions provided here. Yes, it uses the admin API for write operations. Still, I think synapse could automatically purge a room if the last local user left it.

[clokep]

I suspect to have this done automatically we would want to:

1. Add a background update that searches for rooms with 0 joined members.
2. Iterates over them and calls purge.

There's a couple of tricky pieces here:

1. Do we care how long ago this had no members in it? (Do you want to immediately purge the room or wait a few days?)
2. Some logic to ensure only a single background update is running at once, as purging can take a while.

[schildbach]

Alternatively, it could simply purge immediately, within the "last member left" operation. For me, waiting a few days doesn't add any value.