Hacker News Reader

Comment by robobro | Hacker News Reader

One thing I didn't like about Matrix: I had a single server with 3 or 4 other users. In a few months, the database ballooned to, like, 80GB. User-uploaded content, including text, was in the ballpark of 2 or 3GB. Is there any way to clear out the cache of data received from old servers, say, anything more than 2 weeks old, and re-retrieve it if needed?

View on HN

Yeah, there are admin APIs that can clear old images, local and remote ones. For your specific scenario you probably want this one:

https://matrix-org.github.io/synapse/v1.85/admin_api/media_a...

My server is not federated but fairly active, and we treat it as ephemeral. We've configured it so anything older than a week or two gets reclaimed automatically, text and media.

Are you using synapse?

Yes, it's room/concersation states taking up all the space, and you can compress them.

https://levans.fr/shrink-synapse-database.html

https://github.com/matrix-org/rust-synapse-compress-state

That it generates that much cruft for something so simple is disappointing. How long does the compression bandaid work for?

From the first link:

>I shrunk my homeserver database from 100GB to a little under 8GB, during a long maintenance cleanup.

I run the compressor on a cron job. I've been running for 5 years with ~ a dozen people on my server, all federated, in multiple large rooms, didn't run into space issues until this year. On Linode $10 and then Hetzner 4 CPU.

One thing to watch out for: disable registration. Otherwise spambots will register accounts, join the largest matrix.org rooms to spam, and then you'll have to store all that crap for all those rooms.

I feel registration is something which should come disabled by default and need to be explicitly enabled.

At least on Conduit and Dendrite that is the case iirc, but it's easy to forget to disable it once you register an account.

It may be. I setup my server half a decade ago and don't remember exactly how I configured it initially. I may have enabled it not realizing the implications.

This is one of those things that make me appreciate XMPP. There’s no expectation to duplicate the entire history of chatrooms. Many MUCs default to the last 20 messages, which is usually enough to let you catch & jump in on the conversation. Costs of running my server have been quite cheap.

Reply on news.ycombinator.com