I love that people in the comments here are like “have you tried” and “maybe we could”…as if I expected a server full of trans folks to NOT have a bunch of techie people on it :)
this post was submitted on 29 Jul 2023
131 points (100.0% liked)
Blahaj Lemmy Meta
2302 readers
6 users here now
Blåhaj Lemmy is a Lemmy instance attached to blahaj.zone. This is a group for questions or discussions relevant to either instance.
founded 2 years ago
MODERATORS
A few friendly recommendations that may help:
-
Remove the AWS load balancer. Your web application isn't horizontally scaled so you don't need that extra cost. Obviously you'll still have to pay for bandwidth moving all those images but you remove the ELB hourly usage cost from the equation. If you're using it for a reverse proxy, simply install a nginx container on the lemmy box. If really want a load balancer, you'll save money by hosting haproxy on a regular server.
-
Get off Amazon RDS. You'll save money by hosting your database on a regular server than using RDS. You're moving from a fully managed solution to one you have to manage your own. This means managing your own backups and replication if you want HA failover.
-
Switch to object storage for pictrs. The cost is much lower.
However, you're right that Amazon prices are high and you'll likely find cheaper alternatives elsewhere.
I'm not sure I agree with moving from managed to unmanaged for RDS, unless you're very familiar with databases and keeping them online, RDS (and other managed DB providers) are honestly one of the places I recommend spending more on.
The rest of your points are spot on though.
Oh yeah, I agree it's not the right choice for everyone. There's a whole lot that can go wrong self managing DBs. Also, you're losing out on availability as RDS gives you two availability zones for redundancy. It's a trade off I think based on your experience, budget, and uptime requirements. Just spit balling, but I believe it would be something like a 20% cost reduction for the db hosting.
Looks like they've got experience managing databases and are getting off of RDS as a result 🎉
I believe you can run multiple web ui containers to horizontally scale part of it
For just 50¢ a month you can can help save an instance like this one.
If you wish, I can try to migrate 196 to a different instance. That would most probably reduce the load considerably
I honestly really like that 196 is here
While your offer clearly comes from a place of compassion, I don't think that's what the admins are asking for. They stated elsewhere in the thread that the main point of their post was to inform about the downtime and why.
I think if you wanted to do something to make a difference, maybe you could put the ko-fi link in the sidebar of 196 and ask people to contribute if they're able.
It definitely seems like the instances that restrict image hosting are doing a lot better in terms of keeping costs manageable so far. Glad to see there's a plan, though, and no worries about downtime.
Just set up a monthly recurring donation to do my part. This isn't one of my main instances (this is just a backup moderator alt account) but some of my fave communities are here and the instance ethos is definitely something worth supporting!
*shark of the nation
State of the shork?
That's interesting to know the price per user is silver 50¢ per month, I wonder what other instances like beeaw, lemmy.world, etc. are at?
Either way, I hope more people donate to help keep costs manageable, especially reoccurring donations.
Also, thank you so much for the work you've done and continue to do with this, there's no where else in the fediverse I'd rather be than here!
Most people aren't running on AWS, and once we move off of it, our costs will come way down. We're just seeing the reality of a tiny instance for a few people that grew rapidly in a way we didn't see coming. We didn't implement things initially with this scale in mind :)
Out of curiosity, what are most people using currently? I've been really interested in getting involved with the tech behind Lemmy and how to improve it for instance admins.
It seems like the images are playing a big role so perhaps having images being offloaded to another service (like catbox.moe) or converting images to WebP with a healthy amount of compression would help.
Basically decrease the amount of storage you folks are having to pay for.
Or auto delete pictures after a week or 2 and add a disclaimer about it to when pictures are being uploaded.
Maybe also a file size limit. Nothing over 10MB or whatever.
I hope this whole thing can be sorted out I dig this instance.
To be clear, we have the solution. This post was explaining future downtime and why it's happening. We're not going anywhere.
In trans nerds we trust :)
Ah gotcha I misinterpreted it
That's what I get for skim reading on my break lol
Good luck on the server migration 👍
Can you remove image uploading as an option? How about using a third party website for users to upload their image to then link to that to offset costs.
I was going to mention something like this in another post that was concerned about the server slowing down. I particularly only use imgchest for my posts, as it replaced Imgur after their nsfw ban, and don't see why we can't limit uploads to that or other 3rd party services like catbox, imgpile, and such.
Hell it could redirect to catbox.moe
It works pretty good IMO
You will have issues accessing Catbox/files on Catbox if you are connecting from the following countries or providers:
The Australian government […] decided that Catbox was a terrorist propaganda spreading website
the love of footy has overtaken the sensibilites of the British and Irish governments, and they have decided to implement an IP level block
Sounds great.
From a friend I've heard that they can't access it via Tor, which is kinda sad. Under governments which are threatening LGBTQ+ people some may want some kind of protection looking at queer memes.
The pricing issue is unfortunate. Is there any way to donate to keep the instance running?
Maybe a more prominent placement of the donation site might make sense? i am running a small community on this instance and ofc want to help pay for the costs. this thread was the first time saw that donating was even an option.
I love the work you passionate people do on the fediverse. but i also think its just fair to get some compensation for your time and effort.
Best server admins ever <3
I used to work for Akamai for quite a while, so I've got a lot of experience in the CDN sphere. Let me know if you'd like guidance or assistance in coming up with cheaper solutions for handling the heavy lifting of your CDN needs, especially for highly-cachable content.
Set up a monthly donation, thanks for all the transparency
As someone who joined in the 196 migration I love this instance, I'm glad you're saying we're fine to stay <3
How much bandwidth does the instance use a day?
Over a terabyte
That is insane
Yep!
That's crazy is there a breakdown of the bandwidth? Like is it federation, images or ddos using so much bandwidth?
I know you are looking/found a solution but if the main culprit is images would it be possible to offload it to a few servers? I feel like running multiple image server instances would be cheaper for bandwidth. Oracle cloud offers 10tb outbound a month for free per region/month on the free teir. Or is an unmetered host viable? From my understanding Lemmy is pretty light in terms of CPU/memory or does that fall apart at scale?
If you don't mind and it's not too much trouble would you be willing to share stuff like load average/disk use?
Edit I looked this over again. Not an endorsement for giving oracle or Larry Ellison money, they are awful. I was just was trying to say the free stuff is generous. You can use 40tb outbound without paying on the free resources. As always aggressive support for you Ada! Thanks for the transparency in your community!
It's most images from lemmy. We've set up a caching server for pictrs, and that has got the bandwidth part under control.
Next step, get everything else off of AWS :)
Wow! Infrastructure surely is complicated! Thank you for doing this amazing work for us ❤️
ok
I see "An error has occurred!" when clicking through the above link :(