Deploying the Client

Ceph client deployment enables storage access between the client nodes and the Ceph cluster so that users or applications can read data from or write data to the cluster. This section describes how to perform container-based client deployment.

1. Start a container on a client node.

   1 2 3

   podman load -i ceph_release.tar podman tag [IMAGE_ID] localhost/ceph_release:v17.2.7 podman run --name client1 --hostname client1 --privileged --net=host --ipc=host -dti localhost/ceph_release:v17.2.7 /usr/sbin/init

   

   Replace [IMAGE_ID] with the actual image ID. You can run the podman images command to query.

2. Synchronize the configuration file from the server to the client.

   1 2 3

   mkdir -p /etc/ceph scp -r ceph1:/etc/ceph/ceph.conf /etc/ceph/ scp -r ceph1:/etc/ceph/ceph.client.admin.keyring /etc/ceph/

3. On the client, modify ms_async_ucx_device in the /etc/ceph/ceph.conf file based on the devices in show_gids of the local host.

   1	ms_async_ucx_device = mlx5_xxx:1 # Replace mlx5_xxx with the device(s) in show_gids.

   

   You can enter multiple network device names in ms_async_ucx_device. The selected device(s) must contain the management IP address and cluster network IP address configured for the client node.

   Add the following configurations to /etc/ceph/ceph.conf:

   ms_async_ucx_event_polling = false
   ms_async_op_threads = 5
   librados_thread_count = 3

   

4. Copy the file to the client container.

   1 2 3

   podman exec client1 mkdir -p /etc/ceph podman cp /etc/ceph/ceph.conf client1:/etc/ceph/ podman cp /etc/ceph/ceph.client.admin.keyring client1:/etc/ceph/

5. Go to the client container and query the cluster status.
   

   By default, Ceph 17 automatically scales the number of placement groups (PGs) according to the number of OSDs. Ceph administrators can manually adjust the number of PGs to optimize the system and avoid unexpected short-term performance fluctuation. If auto scaling is not required, you can run the following command to disable it:

   1	ceph osd pool set ${pool_name} pg_autoscale_mode off

   1 2	podman exec -it client1 /bin/bash ceph -s