Hello everyone, I am maintaining a cost-optimized Ceph cluster at Codeberg.org, that is serving a lot of small random IO (Git operations). Currently, the data is stored in a single CephFS pool on mixed SSD and HDD cluster, where the HDDs have a primary affinity of 0 to shift the reads to the SSDs. In a quest to further optimize our setup, I discovered the subdatapool feature described at https://github.com/TheJJ/ceph-cheatsheet#subdatapool (assigning multiple pools to a filesystem and using setfattr to allow writes in a certain folder to a separate pool). It feels like this would allow us to cut cost by moving certain directories with large files to a cheaper pool than that used for Git operations easily, without doing data migrations such as splitting data into multiple filesystems. Is anyone using this feature in production and can comment on it? Are there best practices regarding the usage of this feature? Any caveats to be aware of, such as overhead, latency, long-term maintainability? Since certain "interesting" features have been deprecated in the past, such as inline_data, I thought it would be better to ask here for feedback first. Thank you for the comments and have a nice day! Kind Regards Otto --  https://codeberg.org Codeberg e.V.  –  Arminiusstraße 2-4  –  10551 Berlin  –  Germany Registered at registration court Amtsgericht Charlottenburg VR36929.