Federated instances list marking dead instances

[harcesz]

Not sure if it's really a backend thing or more lemmyui, but if a federated instance is down for a longer period of time (week/month?) it could be marked as such on the linked instances list. Or maybe that could be manually done by admins.

Otherwise users who end up there might feel like in a ghost town sometimes. Currently we have 11 on the list out of which 5 are dead for various reasons. Graying them out or creating a separate column, would inform users that there might be something wrong with these particular links.

[Nutomic]

The simplest option for this would be to use the site.updated column, and if it is over some maximum (maybe a week, or a month), add an indicator that the instance is dead. This could be done in lemmy-ui, without any changes in lemmy. But note that this will give false positives, in case an instance is up but not actively federating with our instance. This could be solved with some automated check which regularly tries to update every instance, or in lemmy-ui by indicating that the instance might be dead.

[dessalines]

Kind of a dupe of LemmyNet/lemmy#2221 . Just as important as removing them from the front end, is the backend not retrying to send requests to them after them being dead for potentially years.

[Nutomic]

I will reopen this and move it to lemmy-ui, as thats a better place for it. The other issue is explicitly about some check in the backend.

[dessalines]

Both issues are about dead instances. IMO the back-end should identify dead instances in order to not send things to them, and not display them, so that the GetSite { federated_instances list does'nt contain them. I'd rather this be a comprehensive BE fix rather than a front end hack.

[Nutomic]

There is no definite way to identify dead instances, because an instance can be down for a year, and then come back. And sending activities to dead instances doesnt cause any problems, so i wouldnt bother with a fix for now. So even though both issues are related, I dont think it makes sense to fix them together.

For this issue specificaly, i stil think the best solution is simply to hide instances where updated > 7 days or such. Doing this in the frontend seems better, because then you can have a button "show inactive" which also lists instances that havent been updated recently.

[jacklynnepenick]

An instance can be down for a year and come back, but we could handle that case pretty easily. Perhaps we assume an instance is dead if it's offline for at least a day, and then have a protocol in place so that when an instance comes back online, it notifies the instances it's federated with that it's back. The "I'm not dead anymore" message.