We Saved $500k per Year by Rolling Our Own "S3"
https://engineering.nanit.com/how-we-saved-500-000-per-year-by-rolling-our-own-s3-6caec1ee1143
#HackerNews #WeSaved500k #RollingOurOwn #S3 #CostSavings #Engineering #HackersNews
jbz
boosted
⚠️ MinIO Again Under Fire for Source-Only Decision / Linuxiac
｢ Because the MinIO team has once again stirred community debate after quietly discontinuing pre-built binaries for its Community Edition. According to the project, users must now build MinIO entirely from source — a move that many see as a setback for accessibility and open-source collaboration ｣
https://linuxiac.com/minio-with-yet-another-controversial-move/
⚠️ MinIO Again Under Fire for Source-Only Decision / Linuxiac
｢ Because the MinIO team has once again stirred community debate after quietly discontinuing pre-built binaries for its Community Edition. According to the project, users must now build MinIO entirely from source — a move that many see as a setback for accessibility and open-source collaboration ｣
https://linuxiac.com/minio-with-yet-another-controversial-move/
So, at Amazon, we have a huge ecosystem of internal systems that have all sorts of goofy names and weird interdependencies. A colleague of mine left, I inherited a bunch of his technical assets, and now I have to go around shutting things down, deleting things, etc. But the process is always like:
- I get an email "You are the owner of Frobulator Bob1 and it hasn't been baselined in 90 days." Ok. I know Frobulators. This is Bob's old Frobulator. Let me just go to FrobulatorHub and delete it.
- "Error: Frobulator Bob1 cannot be deleted because it is still associated with 3 Gronks: g1, g2, g3". Fine. I've used Gronks but I've never deleted one. How do you do that?
- Gronks g2 and g3 delete just fine. But I get "Error: Gronk g1 is still commissioned. Decommission it before deleting it." Dammit. How do you decommission Gronks? I go look up the docs.
- "As of 2017, Gronks are the new way of managing all your Splorts in one easy place." What the hell are Splorts? I've never even heard of them. Gronk g1 has just 1 Splort in it: s5. I click delete.
- "Error: Splort s5 is a legacy Splort that cannot be deleted via Gronks management. Visit the old Splort Fleet page to delete s5." The "new" version launched in 2017, but I have to use the "old" system to delete this thing.
- I find s5 on the old Splort Fleet page and delete it.
- I go back to Gronks to "decommission" g1. The "decommission" button is greyed out. I look up their FAQ. "FAQ 73: I deleted all my Splorts, why can't I decommission my Gronk?" And the answer is "Splorts may take up to 48 hours to delete. Wait until their status shows as UNGLUNKED in Splort Fleet, then try your delete again."
- I go back to Splort Fleet. s5 is in status "UNWINDING." What does that mean!? How many statuses are there? Where is that on the journey to UNGLUNKING? Before? After? Why can't we say "deleting" and "deleted"?
So I give up.
A week later I get another email "You still haven't baselined your Frobulator Bob1 and if you don't do it right now, we will page your director and send an email to your mother." Fuck it. I click the 'baseline' button and everyone stops nagging me.
...90 days later...
I get another email and I just click 'baseline' and move on.
Years later it's like "Paco, why do you have a fleet of 34 Gronks? Most people have 1 or 2" Don't ask me, man. I just work here.
And now I have found an S3 bucket with 15M objects in it. That's not that much in the grand scheme of things. I am sure lots of people have bigger buckets. But deleting that many is a pain. The credentials I'm issuing are valid for 60 minutes. And one deletion process seems to do about 1000 objects per API call and about 1 API call/second. If that's accurate, it'll take about 4 hours.
S3 rate limits are per bucket, not per connection or per account. And they're about 3300 per second. So I'm nowhere near that limit. I just have 3 processes all running deleting as much as they can. I only have to do it once, so I'm not bothering to automate it.
#AWS #S3
@stefano, I just watched the recording of Patrick Hausen's @EuroBSDCon distributed storage session and at the end of the session somebody said you might have compared #SeaweedFS with #Garage and that you found SeaweedFS more suitable.
I know that you blogged about running SeaweedFS but I wasn't aware you compared different #S3 platforms, if you indeed did, would you mind sharing your experiences?
Considering switching the #MinIO backend of #mstdndk to #Garage by #deuxfleurs. 3 replicas on #Kubernetes. Anyone with real life experience and/or tips? :-)
Michael Dexter
boosted
Thanks to SeaweedFS, I've set up two replicas of the BSD Cafe's media. One is on a backup server (for disaster recovery purposes only, also containing the hourly zfs-autobackup backups), and the other is on a Raspberry Pi 4 running FreeBSD, which is one of my home backup servers - just one meter away from my desk.
I've also configured Nginx in a jail, so when I'm at home, the BSD Cafe's media is served directly from the local replica.
The performance is lightning fast.
#BSDCafe #SeaweedFS #S3 #Mastodon #SelfHosting #Homelab #FreeBSD #RunBSD
Thanks to SeaweedFS, I've set up two replicas of the BSD Cafe's media. One is on a backup server (for disaster recovery purposes only, also containing the hourly zfs-autobackup backups), and the other is on a Raspberry Pi 4 running FreeBSD, which is one of my home backup servers - just one meter away from my desk.
I've also configured Nginx in a jail, so when I'm at home, the BSD Cafe's media is served directly from the local replica.
The performance is lightning fast.
#BSDCafe #SeaweedFS #S3 #Mastodon #SelfHosting #Homelab #FreeBSD #RunBSD
Michael Dexter
boosted
After conducting some performance tests, I believe #SeaweedFS will be the new S3-compatible software for BSD Cafe media storage. The results have been promising
After conducting some performance tests, I believe #SeaweedFS will be the new S3-compatible software for BSD Cafe media storage. The results have been promising
Michael Dexter
boosted
OK, Minio has decided to drop all the useful GUI features from the open-source (CE) version, making them exclusive to the paid version. More, they'll only provide "security fixes for severe issues" and no new features/development.
Personally, I prefer software that doesn't offer a certain feature at all, rather than one that, at some point, removes it and reserves it for "paying" customers.
It almost feels like a betrayal, in my eyes.
It's probably time to move the #BSDCafe media over to Garage or Seaweedfs.
OK, Minio has decided to drop all the useful GUI features from the open-source (CE) version, making them exclusive to the paid version. More, they'll only provide "security fixes for severe issues" and no new features/development.
Personally, I prefer software that doesn't offer a certain feature at all, rather than one that, at some point, removes it and reserves it for "paying" customers.
It almost feels like a betrayal, in my eyes.
It's probably time to move the #BSDCafe media over to Garage or Seaweedfs.
I don't know if anybody noticed #ZeroFS yet, but it seems there is a completely user space-implementation of #NFS and #blockstorage on top of #S3 #objectstorage: https://github.com/Barre/zerofs
Including a demo running #ZFS on top of it which essentially allows geo-redundant ZFS volumes: https://asciinema.org/a/728234 & https://github.com/Barre/zerofs?tab=readme-ov-file#geo-distributed-storage-with-zfs
I don't see no #FreeBSD port yet, but if that really works it would be absolutely awesome.
Biason::Julio::new();
boosted
A new major version of #Garage had been released: 2.0.0
Garage is an #S3Compatible distributed object storage service designed for #selfhosting at a small-to-medium scale.
It runs on old PCs. It's build to be resilient with geographical redundancy.
A new major version of #Garage had been released: 2.0.0
Garage is an #S3Compatible distributed object storage service designed for #selfhosting at a small-to-medium scale.
It runs on old PCs. It's build to be resilient with geographical redundancy.