Maintenance full taking an unbeleivable long time


We are trying a maintenance full on a bucket having a lot of snapshots (4000).
The maintenance operation is taking 72h, this is a real issue for us. (The underlying network and hardware is not a problem).

We are seing strange behavior in debug logs : the same Blob is being downloaded hundreds of times. A 20Mb file each time, fullly downloaded.

Can this be a bug ? Or at least a missing optimisation ?

The operation is running with : maintenance run --full --safety=none

I feel like I’m kicking someone who is already down on the ground…

We are aware of the implications, but here the behaviour is the same with our without this setting.

We see hundreds of downloads for the same blob.

I have open this issue, since we are beleiving this is a major Kopia issue.

BTW what is your backup back end? S3? rclone?

Scaleway Object Storage : Object Storage - Quickstart
It’s S3 compatible.

Ok I have noticed similar behaviour with rclone and sort of solved it by adding rclone read/write cache. But I suspected that it is because of not perfect rclone/kopia integration. Based on your findings it might be more general issue.