undefined | Better HN

0 pointsdrdaeman2y ago0 comments

I'm a technical user, I've spent last 3 days trying to figure out what should I really do to make a setup that I wouldn't want to nuke in a month (or a year). I'm still at loss how to get this deployed the right way (and not a lazy "install this in 5 minutes, feel the pains forever" way, thanks, my Matrix Synapse server is bad enough).

My current list of grievances:

- No standards of any kind on sharing the same identity, e.g. between Lemmy and Mastodon.

- No support for any distributed multi-master database (I'm currently trying to see if I can hack Lemmy to work with CockroachDB, but I'm probably going to give up).

0 comments

paulmd2y ago

> thanks, my Matrix Synapse server is bad enough

OT but what pains did you experience with matrix / what would you have done differently? this'd make a good blog post/HN article at this moment imo.

drdaemanOP2y ago

I would've explored alternative servers.

The issue with Synapse is the same as I currently have with Lemmy - lack of HA. One particular machine goes down and the system is 100% unavailable. And most of my machines are hardware I own, located at homes (mine, parents, etc) in different countries, so they're slightly more prone to random power or uplink outages than AWS' us-east-1. Hobbyist-grade geo-distributed cluster, hah :)

I've solved this for my email (two mailhosts on different continents, Dovecot dsync replication for mailboxes) and my storage (Minio works, and I'm slowly experimenting with Tahoe-LAFS), but I haven't solved it for Matrix.

Dendrite supports distributed deployments and reportedly works with CockroachDB (albeit not officially supported). Not sure what Conduit supports (last time I've checked it was all about embedded databases like sqlite, rocksdb, sled and persy), but it has pluggable database backend architecture so when I'll finally decide I want to fix my stuff, I'd certainly check it out as well.

Arathorn2y ago

Solving active/backup HA for Matrix is pretty straightforward: postgres replication for the db and a shared filesystem for the media repo. When the primary postgres+synapse goes down, promote the secondary to primary, and continue.

We’re also working on account portability (MSC4014) that would eventually support active/active at the matrix layer, but it’s not ready tet.

2 more replies

j / k navigate · click thread line to collapse