version: "1.0.0" name: drbd-ha-ops description: Operate a DRBD-HA cluster (DRBD + drbd-reactor high availability) through the drbd-ha MCP tools. Use when asked to inspect cluster health, create or manage HA service profiles (MySQL/PostgreSQL/NFS/generic systemd services on DRBD storage), perform failover drills, evict services from a node, or troubleshoot DRBD/drbd-reactor problems.

DRBD-HA Operations

You are operating a DRBD-HA cluster through its MCP server (mounted at /mcp on the drbd-ha backend, e.g. http://<node>:<port>/mcp). DRBD replicates a block device across 2-3 nodes; drbd-reactor promotes one node to Primary and starts the managed services (mount + systemd units + optional VIP) there. A "profile" is one HA service definition.

Ground rules (read first)

1. Read before write. Always check get_ha_profile_status / reactor_status /

list_nodes before any mutating call. Never guess IDs — list first.

1. One mutation at a time. Cluster operations are slow (SSH to every node). Wait for

each tool result before the next mutating call.

1. Evict = controlled failover. evict_ha_profile moves a service away from its

current Primary. It is disruptive but safe; confirm with the user before evicting a production service unless they explicitly asked for a failover.

1. Destructive operations (delete_ha_profile with delete_resource=true,

resource_action with down/invalidate/recover_split_brain) require explicit user confirmation — restate what will be destroyed.

1. After any mutation, verify: profile status should converge within ~30s

(get_ha_profile_status), DRBD should be UpToDate on all nodes (get_resource).

Tool catalog

Discovery / health:

• health, dashboard_summary — quick cluster overview
• list_nodes, check_node, list_available_disks — node inventory
• list_resources, get_resource, get_resource_logs — DRBD resources
• list_pools — LVM/ZFS storage pools
• list_ha_profiles, get_ha_profile, get_ha_profile_status, get_ha_profile_toml
• reactor_status, reactor_logs — drbd-reactor daemon state (per node, includes

enabled flag; a missing enabled field means enabled=true)

• list_resource_agents, list_available_services — OCF agents & systemd units

Mutations:

• add_node — register a node (key-based SSH must already work)
• create_ha_profile — full wizard: storage (LVM pool/volume optional) → DRBD resource

→ promoter config; see references/create-ha-service.md

• activate_ha_profile / deactivate_ha_profile / enable_ha_profile
• evict_ha_profile — failover away from current Primary
• delete_ha_profile — remove profile (optionally the DRBD resource + config file)
• resource_action — low-level drbdadm verbs (up/down/primary/secondary/connect/...)
• reactor_reload — reload/restart drbd-reactor

Workflows

• Create a new HA service end-to-end → references/create-ha-service.md
• Failover drill / move a service / recover a node → references/failover-and-recovery.md
• Diagnose a degraded profile or split-brain → references/troubleshooting.md