Ceph migration / delete old data

Seafile Server
1

Hi there,

I just followed the server manual in migrating my storage backend from on-server filesystem to ceph cluster. Everything worked like a charm.
One question remains, however:
which files in the seafile installation folder on my local disk can I safely delete now?

2

You can remove seafile-data/storage/* folders. These folders contains the objects that you’ve already migrated to Ceph. Other folders should be kept. But I suggest you still keep them for a while in case some data are not migrated unnoticed.

3

Thank you, Jonathan.

4

I just stumbled across this post. I’ve migrated my test server to ceph. We also want to migrate our production seafile cluster. The problem is, that we now use an NFS filesystem as storage backend. After migration, we want to get rid of the NFS mount. But that seems to be impossible, because you say that the other folders have to be kept. Seafile also doesn’t start without a seafile.ini file, where the location of the data is configured. So how can this be solved?

5

There are some limitations (e.g. for WebDAV or Seahub / API uploads it is going to be hard to remove the need for at least some local storage). But some parts could be improved. Thumbnails could be stored in another ceph pool, the same applies for resumeable uploads although I’m not sure if one would get rid of the need to fetch the data from ceph to local disk and upload the computed objects back to fs, commits and blocks pool. But these things would need additional code.

6

So then, what’s the best way to migrate our nfs storage to ceph?

  1. Stop seafile cluster
  2. Run migrate-to-ceph.sh on one of the cluster nodes
  3. Change seafile.conf on each cluster node
  4. Define a new local data directory on each cluster node (seafile.ini)
  5. Copy all contents from the nfs data directory (apart from the storage subdirectory) to the new local directory on each cluster node
  6. Start seafile cluster
7

It would be better if the temporary folder for uploads and thumbnails would still somehow be shared between the cluster nodes.

In case the thumbnails folder is not shared each node would generate the same thumbnail on request, so it’s just some overhead.

In case the upload folder is not shared continuing uploads does not work anymore (so this is imported unless the feature has been disabled).

8

Mmh, this seems to be more complicated than migrating a single server. Are you willing to supply an official howto on migrating a seafile cluster to ceph? I don’t want to experiment with our production cluster. If the migration won’t work, we can’t go back.

9

I’m not a member of Seafile Ltd.

10

Ah, ok :smile:
Maybe the Seafile staff can comment on this.

11

We have a ceph cluster in use, I will write you a PM, maybe I can help.