The stalls are because the database queries are suboptimal. There is one that occasionally runs that, on my instance (I have 337 active users), can sometimes run for 15 minutes and will lock tables. Everything stalls and backs up.

This query was discussed, and I believe in the next release (but unsure) it will be replaced. Instead of using a ton of "not in" clauses, it does a left join now. In testing, someone mentioned it went from multi minutes to multi seconds to run. But there are a lot more such queries.

I think what Friendica needs desperately is a MariaDB/MySql expert to clean up the queries.

Because Friendica supports groups, you can connect to Lemmy communities. This is what kills Freindica. It just cannot handle the hundreds of thousands of daily connections that come in just from lemmy.world alone. Basically, it then becomes a Lemmy/Piefed/MBIN instance plus a Mastodon instance. The database grows by leaps and bounces, queues back up, and it stalls. CPU pegs without relief.

On mine, I finally had to block the Lemmy User Agent at the Cloudflare firewall. I calculated I would have had to spend another $500/month to allow the server to handle the Lemmy traffic comfortably, excluding the continuing cost for DB space. So far, I haven't blocked Piefed and MBIN, but this could change.

Friendica groups were designed for small private groups or specialized groups. Not public forums. I don't think they ever anticipated someone connecting to Lemmy.world communities and that such groups would become so active. I've told people on my instance that if they want to connect to these groups, they should do them using Piefed/Lemmy/Mbin, not Friendica.