Post AsZlvdRUGBsjxAxo6S by karppinen@mastodon.online
(DIR) More posts by karppinen@mastodon.online
(DIR) Post #AsZlvdRUGBsjxAxo6S by karppinen@mastodon.online
2025-03-30T07:00:54Z
0 likes, 0 repeats
Self-hosting a distributed S3 compatible object store is just as horrible as it was 10 years ago. I feel we’ve made great strides on the database side, for example, but this is just as bad as it always was
(DIR) Post #AsZlvepz4b4UHRQsvQ by karppinen@mastodon.online
2025-03-30T07:12:12Z
0 likes, 0 repeats
Ceph: do you have a Ph.D. In Ceph? If yes, it’s greatMinio: bad performance (each object is a minimum of two files on disk but typically more like 16) combined with weird requirements like a minimum of four servers that have an identical set of disks. Replication between clusters is best effort onlySeaweedFS: great in theory, not trustworthy in practice (especially on the filer layer)Deuxfleurs Garage: “you should make periodic snapshots of the metadata database to survive crashes”
(DIR) Post #AsZlvfbqCfLufrp6WG by tyil@fedi.tyil.nl
2025-03-30T09:28:06.069Z
0 likes, 0 repeats
@karppinen@mastodon.online I've tried SeaweedFS and Garage, minio seemed to be too "corporate" to run at home for fun. Guess I'll have to get a PhD in Ceph for my S3 needs/wants.