Ceph is a scalable distributed storage system. Seafile can use Ceph's RADOS object storage layer for storage backend.
By default, a typical Seafile server deployment uses file system as storage backend (e.g. Ext4). Later you may want to switch to more scalable storage solution like Ceph. This documentation shows you how to migrate your existing date from file system to Ceph and connect to Ceph. If you're deploying a fresh install with Ceph backend, please refer to setup with Ceph.
Seafile acts as a client to Ceph/RADOS, so it needs to access ceph cluster's conf file and keyring. You have to copy these files from a ceph admin node's /etc/ceph directory to the seafile machine.
seafile-machine# sudo scp [email protected]:/etc/ceph/ /etc
For best performance, Seafile requires install memcached and enable memcache for objects.
We recommend to allocate 128MB memory for memcached. Edit /etc/memcached.conf
# Start with a cap of 64 megs of memory. It's reasonable, and the daemon default# Note that the daemon will grow to this size, but does not start out holding this much# memory# -m 64-m 128
File search and WebDAV functions rely on Python Ceph library installed in the system.
sudo apt-get install python-ceph
sudo yum install python-rados
It's recommended to create separate pools for commit, fs, and block objects.
ceph-admin-node# rados mkpool seafile-blocksceph-admin-node# rados mkpool seafile-commitsceph-admin-node# rados mkpool seafile-fs
The migration process involves 3 steps:
Create a Seafile config folder for Ceph
Run the migration script
In the Seafile installation folder (e.g.
cd haiwenmkdir ceph-confcp conf/seafile.conf ceph-conf
ceph-conf/seafile.conf, add the following lines:
[block_backend]name = cephceph_config = /etc/ceph/ceph.confpool = seafile-blocksmemcached_options = --SERVER=localhost --POOL-MIN=10 --POOL-MAX=100[commit_object_backend]name = cephceph_config = /etc/ceph/ceph.confpool = seafile-commitsmemcached_options = --SERVER=localhost --POOL-MIN=10 --POOL-MAX=100[fs_object_backend]name = cephceph_config = /etc/ceph/ceph.confpool = seafile-fsmemcached_options = --SERVER=localhost --POOL-MIN=10 --POOL-MAX=100
Now there are two seafile.conf files, one under
conf folder and the other under
From Pro edition 6.0.0 on, the migration scripts are included in the package. For older versions, you have to download the two scripts into
You can run the migration script when your Seafile server is still running.
cd haiwen/seafile-server-latest./migrate-to-ceph.sh ../ceph-conf
If there is any error in the migration process, the script will stop. After you check and fix the errors, you can run the script again. The script is designed to be idempotent to multiple runs. It only copies non-existing objects to Ceph. The script won't delete any objects from the file system backend.
After the initial migration completes successfully, you need to shutdown the Seafile server and run the script again to migrate the data that's added when you run the initial migration. Since the script won't migrate objects that have been migrated, this phase should finish in a short time.
After migration is done. You need to update
conf/seafile.conf to make Seafile server use Ceph as backend in the future.
cp -R conf conf-backupcp ceph-conf/seafile.conf conf/seafile.conf
After restart, Seafile server will use Ceph as backend.
In a cluster environment, you may want to use a memcached cluster. In the above configuration, you have to specify all the memcached server node addresses in seafile.conf
memcached_options = --SERVER=192.168.1.134 --SERVER=192.168.1.135 --SERVER=192.168.1.136 --POOL-MIN=10 --POOL-MAX=100 --RETRY-TIMEOUT=3600
Notice that there is a
--RETRY-TIMEOUT=3600 option in the above config. This option is important for dealing with memcached server failures. After a memcached server in the cluster fails, Seafile server will stop trying to use it for "RETRY-TIMEOUT" (in seconds). You should set this timeout to relatively long time, to prevent Seafile from retrying the failed server frequently, which may lead to frequent request errors for the clients.
Since version 5.1.0 version, we upgraded the bundled Ceph rados library to 0.94.6. On Ubuntu 16.04, this causes some incompatibility. To work around this issue, you have to install librados 0.94.6 in the Ubuntu system (from Ceph's official repositories) and let Seafile use the library from system. To do this, you have to remove a few bundled libraries:
cd seafile-server-latest/seafile/librm librados.so.2 libstdc++.so.6 libnspr4.so
The above configuration will use the default (client.admin) user to connect to Ceph. You may want to use some other Ceph user to connect. This is supported in Seafile. To specify the Ceph user, you have to add a
ceph_client_id option to seafile.conf, as the following:
[block_backend]name = cephceph_config = /etc/ceph/ceph.conf# Sepcify Ceph user for Seafile hereceph_client_id = seafilepool = seafile-blocksmemcached_options = --SERVER=localhost --POOL-MIN=10 --POOL-MAX=100[commit_object_backend]name = cephceph_config = /etc/ceph/ceph.conf# Sepcify Ceph user for Seafile hereceph_client_id = seafilepool = seafile-commitsmemcached_options = --SERVER=localhost --POOL-MIN=10 --POOL-MAX=100[fs_object_backend]name = cephceph_config = /etc/ceph/ceph.conf# Sepcify Ceph user for Seafile hereceph_client_id = seafilepool = seafile-fsmemcached_options = --SERVER=localhost --POOL-MIN=10 --POOL-MAX=100
You can create a ceph user for seafile on your ceph cluster like this:
ceph auth add client.seafile \mds 'allow' \mon 'allow r' \osd 'allow rwx pool=seafile-blocks, allow rwx pool=seafile-commits, allow rwx pool=seafile-fs'
You also have to add this user's keyring path to /etc/ceph/ceph.conf:
[client.seafile]keyring = <path to user's keyring file>