> Also, there is no native support for sharding or cloud storage APIs like S3. Instead, PhotoPrism prefers a fast, local solid-state drive.
Won't use it myself though, as such solution should be centered around s3-like cloud storage with e2ee. Only the index should be in the server (and backup-ed).
Well I have to appreciate the design choice since where I am internet is slow and unreliable. I couldn't imagine having to fetch each photo over the internet just to browse from my own home network.
But there should be a feature to at least sync to a cloud service for backup purposes.
For example DigitalOcean: 1 Tb volume storage (without the instance) is $100/mo.
S3: $13/mo (most would be infrequent access) + traffic would prob be free (100Gb/mo, I don't think I would read more than that).
I am not interested in single servers that can crash on a low-cost provider that can catch fire, get my account deleted because of a software bug, get stolen at home, etc.
Hetzner Storage Box might fit you. Proper host, nice track record, hosted in Europe (probably matters, if you're not there, latency won't be great), unlimited traffic (as long as it's proper traffic) and cheap (no traffic cost nor setup).
- 1TB: 3.81 EUR (100USD cheaper than the DigitalOcean example you made)
I've been paying for two 5TBs boxes (+ rsync.net) for my backups for a very long time and had one issue one time when the connection to one of the boxes was degraded, but otherwise it's been painless.
> Ok, it's 2/3 times cheaper but the reliability is clunky as it's hosted on a single device.
Is it? One of those boxes have been running continuously for 5+ years, according to my monitoring that checks the status every 30 seconds. You can snapshot at your own leisure and it's setup with RAID to resist drive failures.
Or do you actually have something that points to it having troubles with reliability?
Having problems with OVH I don't know what has to do with Hetzner, they are separate companies...
How would you solve inference for things like face detection or object detection? I guess you could run it on the client while uploading but that would make updating models awkward (download every file, run inference, send requests to server to update index).
> Also, there is no native support for sharding or cloud storage APIs like S3. Instead, PhotoPrism prefers a fast, local solid-state drive.
Won't use it myself though, as such solution should be centered around s3-like cloud storage with e2ee. Only the index should be in the server (and backup-ed).