Today kbin.social is blocking a huge list of domains just to get federation working again.
The reason for this temporally block is not to defederate, but rather to get the large backlog of 500k messenger queue processed again. Anyway, this does mean that kbin.social is federating again with other instances.
This is a temporary measure. Several users / developers are looking into how to better optimize the failed message queue, as we speak. Hopefully Ernest has eventually time to dive into solutions as well instead of workarounds, once his instance is migrated to Kubernets. See my preview thread: https://kbin.melroy.org/m/updates/t/4257/Kbin-federation-issues-and-infra-upgrade
List of the domains causing trouble:
lemmygrad.ml, eientei.org, vive.im, lemmy.ml, lemmynsfw.com, kbin.lol, lemmy.webgirand.eu, tuna.cat, posta.no, lemmy.atay.dev, sh.itjust.works, kbin.stuffie.club, kbin.dssc.io, bolha.social, dataterm.digital, kbindev.lerman-development.com, test.fedia.io, mer.thekittysays.icu, lemmy.stark-enterprise.net, kbin.rocks, kbin.cocopoops.com, kbin.lgbt, lemmy.deev.io, lemmy.lucaslower.com, lemmy.norbz.org, social.jrruethe.info, digitalgoblin.uk, pwzle.com, lemmy.friheter.com, federated.ninja, lemmy.shtuf.eu, u.fail, arathe.net, lemmy.click, thekittysays.icu, lemmy.ubergeek77.chat, lemmy.maatwo.com, faux.moe, eslemmy.es, seriously.iamincredibly.gay, test.dataharvest.social, programming.dev, kbin.knocknet.net, pawb.social, lucitt.social, longley.ws, kbin.dentora.social, atay.dev, lemmy.kozow.com, ck.altsoshl.com, pawoo.net, techy.news, lemmy.vergaberecht-kanzlei.de, lemmyonline.com, beehaw.org, pouet.chapril.org, kbin.pcft.eu, fl0w.cc, lemmy.sdf.org, lemmy.zip, feddit.dk, fedi.shadowtoot.world, lemmy.noogs.me, lemmy.kemomimi.fans, social.agnitum.co.uk, fediverse.boo, hive.atlanten.se, forkk.me, lemmy.ghostplanet.org, lemmy.mayes.io, lemmy.mats.ooo, lemmy.world, lemmy.sdfeu.org, lemmy.death916.xyz, geddit.social, masto.fediv.eu
My understanding is that they don't have to handle all traffic from all instances, but rather all traffic that anyone in that instance interacts with. So if you made one for your own personal use, then requirements will only scale up with the number of instances you interact with.
It does seem like it's going to be a big issue specific interfacing with the really large instances though. We'll see how it goes.
Well, actually.. big instances like kbin.social (but that is the same with other big instances or software), need to process not only the outbox but also the inbox of the activitypub. So this thread with comments, need to be sent to several instances. And each instance need to process this comment now. And the other way around is also true, I created this thread but that needed to be sent towards multiple instances (which is converted the inbox from their server perspective). Kbin.social had mainly issues with sending out messages actually (the outbox), because many instances are not responding or are blocked, causing retries and eventually also an increase in the queue.
Right, but if you have a small instance for personal use, only stuff from communities/magazines you have interacted with are sent, no? The complete traffic from those communities shouldn't be that hard to handle in my understanding. Unless you subscribe to 196 in which case good luck. ;)
(I just blocked them, my first block so far. Even though it doesn't show up in /sub it was too much on the "all" page.)
I thought so as well.. But I see threads in my random magazine, from users I do not follow nor I follow this magazine or domain. For instance this thread: https://kbin.melroy.org/m/random/t/4864/A-cache-of-90s-internal-documents-at-Sega-has-suddenly
It seems like once the remote user is added in kbin (due to some post or thread from a magazine I do follow), I will get all the comments, threads and posts from those users as well. On top of the magazines and people I do follow...
But it does seem like for the largest instances the economics will favour centralisation. Decentralisation is typically expensive and inefficient as it often requires duplication of resources.
ActivityPub protocol seems to create big instances (like the big and populair mastodon, kbin and lemmy instances). I hope that spreading content across and finding information across all fediverse will improve over the coming years. And users should be understand the concept of decentralization much better. Users moving to other (smaller) instances will so bring the load down.
Bottom line, I hope AP will be further improved. And users will understand the power of decentralization better (in currently a very very centralized www).