cephadm agent (Technology Preview)
The cephadm agent is a service that runs on each host in a Ceph cluster and provides host and daemon metadata to the Ceph Manager.
By default, the Ceph Manager retrieves host information by using SSH connections. When the cephadm agent is enabled, each host collects metadata locally and sends updates to the manager, reducing the need for repeated SSH operations and improving responsiveness.
Benefits
- Reduces SSH-based communication overhead
- Improves scalability for large clusters
- Provides near real-time updates of host and daemon state
- Improves responsiveness of orchestration operations
How the cephadm agent works
When enabled, the cephadm agent integrates with the cephadm orchestrator and the Ceph Manager module to provide agent-based host monitoring and metadata collection.
- The
mgr/cephadm/use_agentconfiguration option is enabled - The cephadm module deploys the agent service on all hosts
- Each host runs the agent as a systemd service
- The agent collects host and daemon metadata
- The agent sends metadata updates to the Ceph Manager over HTTPS
- The manager processes the data and updates cluster state
Agent-based and SSH-based host management
In SSH-based mode, the Ceph Manager connects to each host on demand to retrieve host and daemon information. This approach relies on repeated SSH operations.
In agent-based mode, each host runs the cephadm agent and periodically sends metadata updates to the manager. This reduces the need for SSH connections and enables more efficient and timely host monitoring.
Metadata collection and synchronization
The cephadm agent replaces SSH-based metadata collection when enabled. Each host periodically collects and sends metadata to the manager at a defined refresh interval.
The Ceph Manager waits until host metadata is up to date before applying service specifications, ensuring consistent and accurate orchestration behavior.
Configuration behavior
The cephadm agent is controlled by configuration options stored in the monitor (MON) database. These options belong to the mgr subsystem.
When configuration values are updated, monitors notify active manager daemons, which then apply the changes across the cluster.