Hey folks!
I am looking for feedback from active lemm.ee users on what you all value when it comes to images on Lemmy. I’ll go into a bit of detail about what our options are, and then I would ask you to voice your opinion about the issue in the comments.
First, some context for those who don’t know. Lemmy software can be configured to handle images in three different ways:
- Store images locally - whenever an external image is posted somewhere, lemm.ee will download a permanent local copy. When you view posts, you are seeing our local copy of the image.
- Proxy all images - similarly to the first option, lemm.ee will download a local copy of external images, however, this copy is temporary. It will be automatically deleted shortly after, and if users open the relevant post/comment again in the future, there will be another attempt to download a temporary copy at that point.
- Pass through external images directly - lemm.ee never downloads any external images, users will always connect directly to the source servers to load the images.
There are pros and cons to each configuration.
Storing images locally
Benefits:
- Your IP address is never leaked to external image hosts, as you never connect directly to the source server. External image hosts only see the IP address of the lemm.ee server.
- External servers don’t become bottlenecks for opening lemm.ee posts. If an external server is slow, it won’t matter, because the image is always available locally
Downsides:
- As time goes on, our storage will fill up with hundreds of gigabytes of useless images, most of which will never be viewed again after the relevant posts fall off the front page.
- Many big external image hosts will rate limit bigger Lemmy servers, causing broken images when we fail to make a local copy.
- Crucially: some people love to spend their time uploading illegal content to online servers. There are tools to try and filter out such content, but these are not perfect. The end result is that there is a high chance of some content like this inadvertently reaching lemm.ee storage and staying there permanently. This downside is why lemm.ee has not, and will not, use this particular configuration.
Proxying images
Benefits: In addition to the same benefits as exist for the permanent local storage, by only temporarily making local copies for the moment they are requested by our users, we free up a ton of storage & remove the risk of permanently storing illegal content on our servers.
Downsides: The key downside is that external rate limits hit us much harder, as we will be requesting external images far more often. This results in a lot of constant broken images on lemm.ee.
Passing through external images
Benefits:
- Images are rarely broken, unless the source server goes down.
- The images never touch our servers, removing a lot of risk with illegal content as well as with storage costs.
Downsides:
- Our users lose a degree of privacy. Every external image that is loaded on your browser will result in the remote server getting a request directly from your computer to fetch that image - this is pretty much the same as you had visited that external server directly, which lets them log your IP address if they wish.
- When remote servers are slow, it can slow down the entire page load in some cases.
Current situation
Initially, lemm.ee was using the third option of passing through images. Ever since support for option 2, image proxying, was implemented in Lemmy code, we immediately switched to that option, mainly for the privacy benefits. However, after many months, and being blocked by more and more external servers, it is clear that image proxying is seriously degrading the user experience on lemm.ee. We often end up with broken images, and our users have to deal with the results.
I still believe image proxying is a really valuable feature, but I am starting to believe it is a better fit for small instances which make much less requests to external servers.
As a result, I am now seriously considering switching back to the previous method of passing through external images.
This is where you come in - I would ask you as users to please let me know which do you value more: the privacy that you get from image proxying, or the better user experience you get from directly passing through images from their source. Please let me know in the comments how you feel. If I get enough feedback about people being against image proxying, then I will be switching it off for lemm.ee soon. Thanks for reading & sharing your thoughs, and I hope you have a great weekend!
Storing permanently locally doesn’t sound like a good solution…
If you could adjust the length of time to keep cached/proxy’d images locally and increase it significantly, I’d think that would be the preferred solution.
No opinion for the moment, but thank you for the very detailed post
Not a lemm.ee user, but here’s my thoughts on #2 since it affects me via federation:
I am not a fan of how Lemmy chose to implement image proxying Specifically, federating the proxied URL.
That frequently prevents my instance from fetching a thumbnail locally (option 1 above). Which, ironically, increases the load on your server as my instance has to fetch it from your proxy every time instead of just once to generate a local copy here.
From a UI development standpoint, the proxied thumbnail URLs also make it harder to detect the image type (gif, static image, video) to handle rendering. It also complicates other proxying/caching methods I have in place. Ultimately, in the UI I develop, I’ve had to resort to passing thumbnail images through a function to un-proxy them so they can be handled sanely.
So I generally wish that admins avoid Lemmy’s proxying until it no longer federates the proxied URL and does something sane like just return that for the local API calls.
Wow didnt know they federated the proxied image kinda stupid ngl.
We really need some sort of distributed content hosting for images that allows everything to have a single unique address servable by anyone. Perhaps a bittorrent that has all federated media.Can still have the address to the media be a url for the local instance as not to break frontends but backends could recognise it as universal bittorrent resource and fetch it in a distributed manner.
Would also mean clients can implement their own retrieval as not to rely on the server but that wouldnt be required.
I suppose u could also put websites content into the same system as a sort of archive. Make the fediverse more p2p distribute load to more smaller nodes improving resiliency.
Anyone know how peertube has done their bitorrent implementation?
While I appreciate you trying to make it so, my privacy isn’t your responsibility. Option 3 is the way to go to keep your costs down, which is the long-term best solution.
It wasn’t one of the options, but from a user perspective a hybrid solution is best. Making a local copy that has a 24-48 hour cache keeps your storage down but still gives us the benefit of option 1. But it sounds like this would require some changes to how the server software works, which would be cool, but again you shouldn’t feel compelled to attempt.
Keep your costs and stress low wherever possible and thank you for everything you do.
I’m on a lot, and I scroll from All/Hot. I rarely see broken images. They do pop up, but not enough to ever bother me. The only option I’d avoid is method 1, because of that image debacle a few months ago. Regarding methods 2 and 3, they both seemed to work fine. I leave it to smarter minds than myself.
Good work on next, btw. Enjoy your weekend and thank you for everything you do!
I say option 3. Sure, it’s annoying, but that’s our problem. Your problem is keeping the server operational, safe, and low-cost.
Considering the vote:content ratio, it looks like most users spend most of their time spectating. The spectators feed our egos and determine the trends in content by singular positive or negative votes. The spectator experience seems far more important to me and it should be the onus of the contributors to ensure their own privacy, just like they do in avoiding doxxing themselves via text.
Option 2 certainly follows in the vein of improved performance, but if real-world implementation is proving too unstable or creating too much overhead, then I say “fuck it, option 3 sounds great.”
2+3: Try to fetch image, if you get it proxy it, if your storage gets full use LRU eviction, once evicted or some amount of time has passed delete it and don’t fetch again, ever. Fall back to pure 3 if there’s ever any issues with anything, including you not particularly feeling like implementing smart caching: Our referrer privacy is not your responsibility.
Thanks for your hard work as always.
I’m in favor of moving away from proxying. Too many images break and proxying in general is very wasteful, having to download images from potentially small servers constantly would definitely get you ratelimited.
Passing through external images is OK. Many people often post external links anyways to sites like imgur and catbox because of the file size limits anyways.
I think the end goal would always to store images locally though - or at least caching them for extended periods of time. Don’t large instances like Lemmy World and huge Mastodon instances work this way? How do they manage the risk?
I’d say option 3. Personally, I don’t care if random websites get my IP among a list of hundreds of others, and if someone wants to keep their IP hidden from strangers, they should be using a VPN before browsing the net anyways. It’d also be nice not to have to open another instance when I come to a post with a broken image that I want to see, but that’s not hugely important to me.
If it were an instance specifically for privacy enthusiasts, that’d be a different story, but this is a general-purpose instance, and option 3 seems to be what’s best for both general users and the server itself.
Have you thought about solving this issue in the front end? The client I’m using (Mlem) implemented a feature to directly access the image if the proxy fails. This feature can either be triggered automatically or by pressing a button on the failed image. This allows users the benefit of the proxy while also having the option to give up their IP if they want to see a broken image.
If this is feasible, it sounds like an elegant solution. Does it work for the various mobile apps, or would each app need to do it on their end?
If there was a Lemmy account setting to automatically bypass the proxy on failure then it might not require any front end work, but the manual bypass button would definitely need to be implemented in each client
This is something I think would be the best solution. It seems like the best possible tradeoff between user privacy, and actual effectiveness.
I like 3 also. if you can’t get a good user experience privacy matters less because there are no users, and anyone can spin up a super privacy enhanced instance if they want.
I was thinking however that a super stand alone image server would be a great thing for Lemmy, and pixelfed and the Fediverse in general.
Imgur blew up and was the go to for image hosting on Reddit for years, until Reddit realized it was leaking traffic and users to them and started their own. But hosting images has a lot of potential headaches like copyright violations and big corps suing you into oblivion, in addition to inadvertently hosting illegal stuff. The Fediverse will need some good image hosting servers and video hosting servers as part of the plan in the long run though.
I would pick image pass through because I’m not necessarily concerned about the image hoster logging my IP. I have been more frustrated with broken links, so I am very much opposed to the current method.
I think it’s important for us to be mindful of content retention for posterity’s sake if we want Lemmy to compete with Reddit long-term. If possible, I’d hope we can avoid dead image links like we see with old forums and photobucket pics, for example.
Option 3.
Images have been a bit problematic for me lately, for sure. If storing them locally is not a solid option, the question I would have is how much of the other requests are proxied? As in, what other stuff apart from images/media is not being proxied? If the clients are leaking IPs anyway, maybe it’s okay to have them download the images, too. But if the server is proxying everything else then having some sort of a cache might not be that bad an idea
