17.10.2023 12:22, Filip Hanes via dovecot пишет:
S3-compatible storage is very good for multi-server installations where you need redundancy, availability. S3 is basically HTTP server so you can code your own logic on stored emails, balancers, caches, deduplication, compression, encryption it does't need to be off-the-shelf storage.
is S3 better then cephfs?
The drawbacks of cephfs is you need to have the mds. If you scale the mds you could have some issues. I think even in newer ceph releases you need to start pin them on pools / directories. I still have issues with cephfs mounts locking up on a hyperconverged setup so I am not using it in production, but I am still on a older version.
The flip side to using the mds, is that it is caching a lot of meta data so in theory you could have a better performance with cephfs than writing directly to rados. Writing to rados directly seems to me the most stable.
What I thought was super strange about the s3/radowsgw layer is that if you rename a file, the file is actually copied to a new name. It is not renamed. I am not sure if this is a standard and still like this, but s3 is just developed for a different use. So it depends on how you use s3/radosgw, object storage directly or cephfs.