e569668

e569668, 6 months ago (edited 6 months ago)

this might be https://github.com/MbinOrg/mbin/issues/35

it was fixed in the docs but requires manual change by instance owners that were affected, which this instance appears to be. tl;dr the docs recommended the nginx config add_header Referrer-Policy "no-referrer" always; which broke the style of redirects that were being used, it's now been changed to recommend add_header Referrer-Policy "same-origin" always;

more info here

edit: I promise I can read... Your description is slightly different, so it might not be this, but still worth trying it out to see if it helps

e569668, 6 months ago

That is very weird, their /instances page says kbin still... I wonder if they stopped federating / it got marked as an inactive instance by mistake? lemmy.world for instance shows fedia as mbin... I'm not quite sure what this means but they clearly haven't gotten node info in a while

e569668, 6 months ago (edited 6 months ago)

It looks like their instance is having some issues https://programming.dev/post/4308156 granted that says outbound but I just joined their matrix and there are other issues being discussed as well, I might wait on this to see if it's still happening once things get more sorted

edit: i posted in their matrix support channel to see if they noticed anything odd as well 🤞

e569668, 6 months ago

Minor problem: there's a new feature of 2fa, but at least for me, going to https://fedia.io/settings/2fa to set it up shows a broken qr code. looks like it's returning an nginx 404 at the moment

e569668, 6 months ago

Both servers are larger than the single server that infosec.pub runs on, yet infosec.pub has about 10x the traffic, and kbin is struggling under the load.

I wonder what the difference is. One thing I have always wondered is if the double comment / post issue is causing tons of double traffic, not to mention double storage / bandwidth / network costs. I really wish there was an easy solution but I haven't been able to find any answers. The random thing I thought was like, it looks like everything federates with both fedia.io and www.fedia.io, so I was like maybe everything is double sending on its side? (E.G. if you look at lemmy.world or lemm.ee etc they all list fedia twice.) But people didn't seem that was likely to be the cause, as hopefully everything is deduped by signing keys rather than hostnames. I don't know enough to investigate further myself.

e569668, 6 months ago

I don't know of every instance of course, but from the general purpose ones the order of up-to-date from most to least at the time of posting would be something like: kbin.run, fedia.io, kbin.cafe, artemis.camp (specific API branch), kbin.social.

But there's also other considerations. fedia has a double post/comment issue and a low amount of 500s. Artemis has been giving me a lot of 500s, can't upvote anymore (just reloads the page), commenting doesn't always federate. Many instance admins seem to have disappeared since 3 months ago. I'm not sure what life is like on instances I'm not actively using so there might be issues unknown to me.

Edit: the keys that I look for are localization / ui changes. "more" dropdown now has "more from domain" on updated instances. The UI for the sidebar is different, and most recently the footer is differently styled. There's also the federation list of instances page in the footer now.

e569668, 6 months ago

does the fediverse actually support the kind of block that's being asked for here?

Yea, that's where I'm thinking the hangup on this might be. A block could be implemented, but it'd come with the caveat of that all it's doing is giving you the idea they aren't continuing to engage with you on your instance. On their instance, and any instance that federates with them, they and others will continue to see the replies.

Personally, I would like to see block renamed to mute to be more accurate and a block from replying added with the note about the drawbacks of them being able to tell you blocked them and their posts still going out everywhere else. That at least empowers the user to make the decision themselves on what they're most ok with. My reasoning is: changing the UI for, let's say an aggressor, gives them a reason to retaliate. To me, either blocking method is a lose-lose; either it doesn't stop engagement which some users clearly want it to, or it makes it obvious someone is being blocked which start aggressors down the retaliation path. That's kind of why I'd want users to make their own risk assessment on actions.

Anyways, that's all very unlikely to happen. Most of all I'd like the bug about notifications fixed because that is clearly not working as intended.

e569668, 6 months ago

Yeah, Mastodon has separate block and mute functions.

Ah, good to know. I'll have to look into it a bit more, but reading https://docs.joinmastodon.org/user/moderating/#block

If you and the blocked user are on the same server, the blocked user will not be able to view your posts on your profile while logged in.

It looks like the limitation that Pamasich and I sort of expected is there. Blocks are basically only possible at your own instance. If the user is on another instance, there's no way to stop them in the fediverse. And that includes it going out to all other instances they federate with too.

I sort of just experienced how this would work if implemented, in a way. A kbin social user posted to a beehaw org magazine. I replied to it, but my post does not seem to have made it to kbin social. However, it's on my instance, beehaw's, lemmy one, etc, because my instance federates with all of those instances. That's sort of what blocking would be like if the original page refused an incoming comment due to a block, all other instances would still accept it. It's possible there's something I'm missing as I'm not super knowledgable on activity pub or the fediverse, so I'll try to learn more about it

e569668, 7 months ago

The good news is this was fixed in latest https://codeberg.org/Kbin/kbin-core/issues/1019 the bad news is it hasn't made it to kbin.social yet so hopefully whenever this instance is updated it will be fixed

e569668, 7 months ago

I posted in https://kbin.social/m/kbinMeta/t/509976 as well but just to make sure you see it, ernest was pinged so hopefully this will be fixed when he sees it

e569668, 7 months ago

Peeps pinged ernest in matrix so hopefully he can take care of this when he gets the chance. Thanks (to squiblet as well) for bringing it up!

e569668, 7 months ago

Maybe gremlins, looks like it's working now. (I have noticed some 500s, but I assume jerry is still running his auto-null-image-fixer script and it will start working eventually). It's empty though as is the case with federation they don't automatically pull in things until added so the past content isn't there

e569668, 7 months ago

I don't want to give false information, but on fedia the majority of 500 pages seemed to stem from the image media cache causing issues (our instance admin detailed the issue here). People were able to mitigate it by disabling images everywhere they could, but one place they can't be disabled is the user avatars on the notification page. I could be wrong, but it's my belief it's those images causing the 500s, and that pushed me to add https://codeberg.org/Kbin/kbin-core/issues/1039 in hopes that if it was possible to disable them the 500s would stop. All this could be wrong though, maybe it's something else like deleted messages or anything else, or maybe there's a better way to solve issues with cached images

e569668, 7 months ago

I'm able to see that post, do you see it in private browsing / when logged out? If so, you might be blocking domains on kbin.social and removing them would make it appear again. That's been fixed in latest which hopefully kbin.social users should get soon, but wasn't affecting other instances I had checked so you should be fine to do so on artemis

e569668, 7 months ago

You can set the default landing page here https://fedia.io/settings/general it's the first option for homepage. The options are all, subscribed, favorites, and moderated, make sure to click save near the bottom and not the delete account button.

The closest you can get is blocking those domains, it won't remove everything from them but there isn't a block instance option at the moment, so remote link posts will still appear from them.

e569668, 7 months ago

he’s eventually going to make sure abandoned instances have admins

I think he meant magazines on kbin.social in this context. I don't think he has any way to control another instance and change who runs it, besides a kindly worded letter.

I do try to keep track of kbin instances and how up to date they are and their admins, but I can't really advertise in their place. One thing I've kind of been thinking about is it'd be nice if nodeinfo had information next to current user count of like: desired user count, max user count. Otherwise you have no idea what an instance admin is hoping for, whether it's just their personal instance or they're willing to support tens of thousands of users. Without that I don't want to send people to x instance and suddenly the admin being in for a surprise. But I just looked up nodeinfo and it looks like there's plenty of worms in that can about it being too complex already, so maybe I'll avoid opening my mouth about it... for now.

e569668, 7 months ago

It looks like all the spam I've seen are from newly registered accounts. Perhaps this is something you've already considered, but would shutting off registration on kbin.social and pointing people to other instances until things are more manageable help?

e569668, 7 months ago

I'm not sure if this is a good suggestion, are you able to delete it via moderation as you're the magazine owner? Someone mentioned having issues deleting their own posts but still could via moderation recently, but I'm not actually sure what that difference means as I haven't moderated any magazines myself.

e569668, 7 months ago

You should be able to see microblog posts of accounts you follow at https://kbin.social/sub/microblog

e569668, 7 months ago

Interesting, I've heard people say the same before. I do always sort by newest myself, but I sub to a bunch of magazines and also, say, three mastodon accounts, and that page is basically just solely the mastodon posts, very rarely do I see microblog posts in lemmy/kbin. I guess a filter of magazines vs users would be useful

Edit: Also, is the default view all time for you? I could see that causing a problem as well if it's hot / all time. I mainly have three bookmarks: /hot/12h, /sub/newest/24h, /sub/microblog/newest/1d. Getting all results for all of time by default seemed a bit weird to me, especially for microblog posts

e569668, 7 months ago

Definitely, you're right that new timeframe doesn't matter, I meant to type hot all time xD I would expect that to potentially grab like whatever was boosted most in the past but then again the algorithm does tend to show newer things even under hot.

Yea the tag system might need a bit of work. It works great for some magazines, but for others not so much. For example m/help is supposed to be a magazine for kbin help but it pulls in everything under #help which can be anything in the fediverse. I did see someone suggest a way for a magazine to remove it defaulting to using the tag of its name, which doesn't seem like a bad idea to me

e569668, 7 months ago

In hindsight I remember why I filter newest to a timespan, so wanted to make note of it. Right now:
https://fedia.io/m/kbinMeta@kbin.social/microblog/newest 500s for me but
https://fedia.io/m/kbinMeta@kbin.social/microblog/newest/1w is able to load. so cutting off some of the older posts helps me run into less 500 issues, and I generally don't care too much about old posts unless I'm searching for something. that might be a fedia specific thing though, not sure how 500s are on other instances

e569668, 8 months ago

I believe you should be able to run it all locally without a hostname. The guides for doing so are here, depending on how you want to:
https://codeberg.org/Kbin/kbin-core/wiki/Admin-Bare-Metal-Guide
https://codeberg.org/Kbin/kbin-core/wiki/Admin-Docker-Guide

Edit: I should clarify, I meant to test things without activitypub. As the other poster mentioned you would need a valid hostname and valid certs for federation testing. Though it might be possible to run multiple local kbins that trust each other, not sure there is going to be much information on that

Someone just went through this themselves in the matrix kbin development channel so you could also potentially find help there if you're having issues. One thing that came up during that is https://codeberg.org/Kbin/kbin-core/pulls/132 has a bunch of changes to those docs which might make it easier

e569668, 8 months ago

It looks like ever since this maintenance, no kbin instances can get images from startrek.website:

https://kbin.run/d/startrek.website
https://kbin.social/d/startrek.website?p=3
https://kbin.cafe/d/startrek.website?p=3
https://fedia.io/d/startrek.website/newest?p=4

(the links atm go to ~3 days ago when maintenance happened but ofc will start to drift as more posts are made)

You can see the thumbnails disappear exactly at the mark of the maintenance and ever since then they're gone. If you see thumbnails after that, they're usually wrong ones that are being loaded from different posts (there's a separate issue in for that now). I was wondering if you were aware of any issues or changes that happened because of this maintenance that would prevent kbin instances from fetching thumbnails? I've raised this in the kbin matrix chat as well

e569668, 8 months ago

Just to follow up on this, contributors to kbin investigated and found that startrek.website appears to block user agents that contain "curl" in it now.

Example:

curl -v --user-agent "curl" https://startrek.website/pictrs/image/3ff1b940-75da-4e6b-8948-834ebd4b454c.jpeg

times out while:

curl -v --user-agent "anything-else" https://startrek.website/pictrs/image/3ff1b940-75da-4e6b-8948-834ebd4b454c.jpeg

works. kbin apparently sends User-Agent: Symfony HttpClient (Curl) in its requests for images, which is different than how it handles other fetch requests. they've made a PR to change kbin to send the user agent it uses elsewhere in those image requests https://codeberg.org/Kbin/kbin-core/pulls/1098 once that is approved / merged they should start working again as instances update. It'd be nice if, assuming this is all accurate, the block on curl could be removed so people don't have to wait for the PR / instances to update but I understand if that is difficult

e569668, 8 months ago

One thing to keep in mind is they don't federate with kbin.social; so they can't see your posts or replies, and can't reply to you. I believe your upvotes won't federate (maybe to other instances, but not to theirs at least). Downvotes work the same as they do for any other post as those don't federate in kbin to begin with. One-sided federation in the fediverse seems like it can be a bit confusing to users, I'm not sure in what case it's a good idea in since it can be like talking to a void.

e569668, 8 months ago

My guess would be it's due to you having a follower from kbin.social https://kbin.social/u/@cwagner@lemmy.cwagner.me/followers

e569668, 8 months ago

Check out https://codeberg.org/Kbin/kbin-core/issues/1019#issuecomment-1064117 and https://kbin.social/m/kbinMeta/t/361909/-/comment/1870631 as well.

I think when people get an error posting to kbin.social a text or photo post it doesn't federate until, as floppy / stopthatgirl7 mentioned, it is interacted with.

Edit: After reading your replies closer I see you were already aware, my bad for repeating information. There is a PR pending for the blocked domains so hopefully that is resolves soon

e569668, 8 months ago

I feel like I've been saying this for a while now but I'm hopeful full release will happen soon and that will hopefully make it easier to chase down bugs, at least different sites might be versioned so we get an idea of when problems manifest.

Just as an FYI there's a pretty bad bug right now https://codeberg.org/Kbin/kbin-core/issues/1019 that makes kbin.social users unable to see text and image posts to kbin.social if they have domains blocked. It seems to only be affecting kbin.social so I can't say for sure if it's in the develop branch or something specific to the setup ernest has. I've been basing my guesses off things like localization versions and, based on that, kbin.social is actually behind in commits compared to other kbins, so that makes me wonder if it's not something in the develop branch, but I could be mistaken, just a heads-up something to watch out for before updating in case of risks (at least there's a way to mitigate it, but as far as I can see it isn't happening on fedia.io)

e569668, 8 months ago

Regarding image thumbnails, it is quite a mess :( There must be some bugs happening somewhere with the pipeline for caching, and they started being mixed up on kbin.social as well which is worrying

For instance, look at this user https://fedia.io/u/@Facelikeapotato@lemmy.ml almost all of their thumbnails are wrong. Really you can browse https://fedia.io/m/cat@lemmy.world and just look for non-cat pictures. I've seen all kinds like nsfw, git repos, news thumbnails, but here are some examples [1] [2] [3].

I posted a theory here https://fedia.io/m/kbinMeta@kbin.social/t/221067/-/comment/1196287 when I noticed that: it seems to happen a lot on instances that go down often or don't respond quickly, so my guess was it's having trouble getting a response, and then it just fails and, potentially, uses the next thumbnail value it gets. That's based off of the thumbnails for nearby entries being similar, as I posted in that comment the one that was wrong was using the same image as a post 31 seconds after it was made

As I posted here https://fedia.io/m/fedia/p/66268 I notice it's easier to browse with show thumbnails turned off, as that stops the 500s when the image thumbnail has issues. Sadly it still happens on some pages and I ~think that's because avatars for users, which are also cached, are shown on that page, like the notifications page, and there is no way to disable displaying user avatars on those pages

e569668, 8 months ago

my profile, as well as hitstuns apparently - https://fedia.io/u/e569668 https://fedia.io/u/hitstun
my microblog sub feed - https://fedia.io/sub/microblog ... that doesn't help you I guess lol, let's see it appears to be https://fedia.io/m/kbin@kbin.social/microblog and https://fedia.io/m/FloatingIsFun/microblog for me

Definitely can confirm posts that were 500ing for me yesterday are working now though, so thank you for taking the time to fix things up

e569668, 7 months ago

So my profile as you can see still 500s. There's also the duplicate thread / comments issue which persists.

jerry just updated to latest yesterday, so fedia now has a lot of the new changes which should hopefully make it a little more stable, though there was a big effort to fix a lot of the 500s riiiiggghhhhtttt as he updated so I'm not sure if he snuck in the middle of those updates or caught them all. Some times I might get a 500 looking at my sub list, but it will fix itself later. I assume jerry is still running scripts which basically try to resolve those issues on a cron job so it might not work for a bit but then work later.

e569668, 7 months ago

If it's avatar uploads hopefully it's something simple like being caught inbetween it being broken and fixed for example https://codeberg.org/Kbin/kbin-core/issues/1022 and all it requires is an update

e569668, 8 months ago

This seems to be a popular topic today, I imagine it must be getting worse for kbin.social users (the errors when posting threads / images, if it is that)

I made this issue for what you described https://codeberg.org/Kbin/kbin-core/issues/1019

@ernest sorry to ping you but just want to make sure you're aware of this as the last ~4+ posts to this magazine have been about this issue