Things have been incredibly unstable there. Until things stabilise, they should force the traffic elsewhere.
Things have been incredibly unstable there.
I wish lemmy.ml (also unstable) or lemmy.world would hand out a (nearly) full copy of the database so we can get more analysis done on PostgreSQL performance behaviors. Remove the private comments and password /2fa/user, or whitelist only comments/posts/communities/person tables - but most everything else should already be public information that’s shared via the API or federation anyway. it’s the quantity, grouping, and the age of the data that’s hard to reproduce in testing. And knowledge of other federated servers, even data that may have been generated by older versions of Lemmy that new versions can’t reproduce.
It’s been over 60 days of constant PostgreSQL overload problems and last week Lemmy.ca made a clone of their database to study offline with AUTO_EXPLAIN which surfaced a major overload on new comments and posts related to site_aggregates counting (it was counting each new post/comment against every known server, not just the single database row for a server).
I have an account over on World too, and every major Lemmy server I use throws errors with casual usage. It’s been discouraging, I haven’t visited a website with this many errors in years. Today (Sunday) has actually been better than yesterday, but I do not see many new postings being created on lemmy.ml today.
[This comment has been deleted by an automated system]
Wasn’t this how kolektiva.social got in trouble 😂
What is the story? never heard of it.
deleted by creator
ok, I did know about that, just didn’t memorize the name. I’m assuming only private messages and user account info (email address) are the real concern in terms of exposure? It’s mostly a public posting thing, or not?
deleted by creator
Are you really trying to gatekeep the internet?
No, I’m trying to find a solution to a very real problem. World has had at least partial outages every day for a week now. It affects everyone when it gives the impression that things don’t work and people aren’t able to create content to attract new people to the wider platform. But have an upvote anyway.
What?
People can subscribe to other instances or even host their own. If the majority of users end up on the same instance then what’s the point of decentralization? Even worse if the traffic it generates for that instance makes it unstable.
Should the admin do nothing until the instance is mostly down with windows of uptime throughout the day?
A big part of the problem is that a new instance starts with zero database content and PostgreSQL performs fine with the way Lemmy organizes the data. But then there isn’t anything for people to read, and search is only going to pick up local stuff.
There’s no perfect solution but I certainly didn’t move to a decentralized platform only to see it be intentionally centralized through inaction.
The quantity of users on Lemmy I still consider to be pretty low, the performance bugs need to be addressed on a big server. Bugs like not having a WHERE clause on an UPDATE hitting 1500 rows in a table (one row per server) instead of 1 single row… these need to be shaken out.
The errors of the overload themselves have been a way to throttle growth of the big servers. People were not able to insert new posts and comments into Lemmy.ml - reducing outbound federation activity too, and they went to other servers. This went on all of June and July.
You are welcome to go to another instance to reduce lemmy.world load.
I’m not on world. I don’t even have a world login. 🤷🏾♂️
I don’t think you understand how the fediverse works of you think that solves the problem - lemmy.world being unstable effects its ability to federate properly with other instances
Badly managed popular instances unfortunately affects the whole fediverse
There has also been problems with federated copies of communities not getting all the actions. I added testing code to demonstrate that comment deletes were not going out to federation peers. Comparing copies of data between instances for the same community shows some overlooked problems. Still have more tests to add.
deleted by creator
