test/cluster/lvmtestssh - main - githubmirror/lvm2

test/cluster/lvmtestssh
250 строк7 KB
David Teigland
add test/cluster framework: lvmtest, test scripts, configs, and groups
03 июл 2026, 01:24
03 июл 2026, 01:24f0a4ff3
#!/bin/bash
#
# lvmtestssh - SSH into a cluster test node or execute commands on all nodes
#
# Usage: lvmtestssh [-i cluster_id] <node_number> [command...]
#        lvmtestssh [-i cluster_id] -a <command...>
#        lvmtestssh [-i cluster_id] -p <command...>
#
# Examples:
#        lvmtestssh 1                          # SSH to node 1
#        lvmtestssh -i lvmtest-foo 1           # SSH to node 1 in cluster lvmtest-foo
#        lvmtestssh -a "lvs"                   # Execute lvs on all nodes serially
#        lvmtestssh -p "vgs"                   # Execute vgs on all nodes in parallel
#

set -e

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

# Source cluster library for cluster_validate_cluster_id
# shellcheck disable=SC1091
source "$SCRIPT_DIR/cluster-test-lib.sh"

cluster_init_privileges
SSH_KEY="${HOME}/.ssh/cluster_test_rsa"
SSH_USER="${CLUSTER_SSH_USER:-root}"

usage() {
    cat <<EOF
Usage: $0 [-i cluster_id] <node_number> [command...]
       $0 [-i cluster_id] -a <command...>
       $0 [-i cluster_id] -p <command...>

SSH into a cluster test node or execute commands on all nodes.

Options:
  -i cluster_id  Cluster ID (default: auto-detect most recent)
                 Simple names auto-expand: "foo" -> "lvmtest-foo"
  -a             Execute command on all test nodes serially (all nodes)
  -p             Execute command on all test nodes in parallel

Arguments (without -a/-p):
  node_number    Node number to connect to (0, 1, 2, ...)
                 Node 0 = storage exporter
                 Nodes 1..N = test nodes
  command        Optional command to execute on the node

Arguments (with -a or -p):
  command        Command to execute on all test nodes

Examples:
  # SSH to specific nodes
  $0 1                              # SSH to node 1 (auto-detect cluster)
  $0 -i foo 1                       # SSH to node 1 in lvmtest-foo cluster
  $0 -i lvmtest-bar 0               # SSH to node 0 in lvmtest-bar cluster
  $0 1 "systemctl status sanlock"  # Run command on node 1

  # Execute on all nodes
  $0 -a "lvs"                       # Execute lvs on all test nodes serially
  $0 -p "vgs"                       # Execute vgs on all test nodes in parallel
  $0 -i foo -a "hostname"           # Execute hostname on all nodes in cluster foo

EOF
    exit 1
}

# Parse options
CLUSTER_ID=""
EXEC_ALL_SERIAL=0
EXEC_ALL_PARALLEL=0
while getopts "i:aph" opt; do
    case $opt in
        i)
            CLUSTER_ID="$OPTARG"
            ;;
        a)
            EXEC_ALL_SERIAL=1
            ;;
        p)
            EXEC_ALL_PARALLEL=1
            ;;
        h)
            usage
            ;;
        *)
            usage
            ;;
    esac
done

shift $((OPTIND - 1))

# Check for conflicting flags
if [ $EXEC_ALL_SERIAL -eq 1 ] && [ $EXEC_ALL_PARALLEL -eq 1 ]; then
    echo "Error: Cannot use both -a and -p flags together" >&2
    usage
fi

# Parse arguments based on mode
if [ $EXEC_ALL_SERIAL -eq 1 ] || [ $EXEC_ALL_PARALLEL -eq 1 ]; then
    # Mode: Execute on all nodes
    if [ $# -lt 1 ]; then
        echo "Error: Command is required when using -a or -p" >&2
        usage
    fi
    # All remaining arguments are the command
    COMMAND="$*"
else
    # Mode: SSH to specific node
    if [ $# -lt 1 ]; then
        usage
    fi

    NODE_NUM="$1"
    shift  # Remove node number from arguments
fi

# Validate node number (only in single-node mode)
if [ $EXEC_ALL_SERIAL -eq 0 ] && [ $EXEC_ALL_PARALLEL -eq 0 ]; then
    if ! [[ "$NODE_NUM" =~ ^[0-9]+$ ]]; then
        echo "Error: Node number must be a number" >&2
        usage
    fi
fi

# Normalize cluster ID if specified (foo -> lvmtest-foo)
if [ -n "$CLUSTER_ID" ]; then
    normalized_id=$(cluster_validate_cluster_id "$CLUSTER_ID" 0 2>&1)  # 0 = don't check if exists
    if [ $? -ne 0 ]; then
        echo "Error: Invalid cluster ID: $CLUSTER_ID" >&2
        exit 1
    fi
    CLUSTER_ID="$normalized_id"
fi

# Check if -i option appears in remaining arguments (common mistake)
for arg in "$@"; do
    if [ "$arg" = "-i" ]; then
        echo "Error: The -i option must come before the node number" >&2
        echo "Correct usage: $0 -i cluster_id node_number [command...]" >&2
        exit 1
    fi
done

# Find cluster ID
if [ -z "${CLUSTER_ID:-}" ]; then
    # Auto-detect most recent cluster
    if [ ! -d "$CLUSTER_STATE_DIR" ]; then
        echo "Error: No cluster state directory found: $CLUSTER_STATE_DIR" >&2
        echo "Have you created a cluster yet?" >&2
        exit 1
    fi

    # Find most recently modified state file
    CLUSTER_ID=$(ls -t "$CLUSTER_STATE_DIR"/*.state 2>/dev/null | head -1 | xargs -r basename -s .state)

    if [ -z "$CLUSTER_ID" ]; then
        echo "Error: No cluster state files found in $CLUSTER_STATE_DIR" >&2
        echo "Please create a cluster first with: ./cluster-test-main.sh create" >&2
        exit 1
    fi

    echo "Auto-detected cluster: $CLUSTER_ID" >&2
fi

# Load cluster state
STATE_FILE="${CLUSTER_STATE_DIR}/${CLUSTER_ID}.state"
if [ ! -f "$STATE_FILE" ]; then
    echo "Error: Cluster state file not found: $STATE_FILE" >&2
    exit 1
fi

# Source the state file to get node IPs
# shellcheck disable=SC1090
source "$STATE_FILE"

# Check if CLUSTER_NODE_IPS is set
if [ "${#CLUSTER_NODE_IPS[@]}" -eq 0 ]; then
    echo "Error: No node IPs found in cluster state" >&2
    exit 1
fi

# Check if CLUSTER_NUM_NODES is set (needed for -a/-p mode)
if [ -z "${CLUSTER_NUM_NODES:-}" ]; then
    # Try to infer from array size (subtract 1 for node 0)
    CLUSTER_NUM_NODES=$((${#CLUSTER_NODE_IPS[@]} - 1))
    if [ $CLUSTER_NUM_NODES -lt 1 ]; then
        echo "Error: Invalid cluster configuration - no test nodes found" >&2
        exit 1
    fi
fi

# Check if SSH key exists
if [ ! -f "$SSH_KEY" ]; then
    echo "Error: SSH key not found: $SSH_KEY" >&2
    exit 1
fi

# Execute on all nodes (serial or parallel mode)
if [ $EXEC_ALL_SERIAL -eq 1 ] || [ $EXEC_ALL_PARALLEL -eq 1 ]; then
    # Source cluster-executor.sh to get cluster_all_exec and cluster_nodes_exec
    # shellcheck disable=SC1091
    source "$SCRIPT_DIR/cluster-executor.sh" || {
        echo "Error: Failed to load cluster-executor.sh" >&2
        exit 1
    }
    # Source cluster-vm-manager.sh for SSH helper functions
    # shellcheck disable=SC1091
    source "$SCRIPT_DIR/cluster-vm-manager.sh" || {
        echo "Error: Failed to load cluster-vm-manager.sh" >&2
        exit 1
    }

    # Export required variables for executor functions
    export CLUSTER_ID
    export CLUSTER_NUM_NODES
    export CLUSTER_NODE_IPS
    export CLUSTER_SSH_USER
    export CLUSTER_SSH_KEY_DIR="${HOME}/.ssh"

    if [ $EXEC_ALL_SERIAL -eq 1 ]; then
        echo "Executing command on all test nodes (serially)..." >&2
        cluster_all_exec "$COMMAND"
        exit $?
    else
        echo "Executing command on all test nodes (in parallel)..." >&2
        cluster_nodes_exec "$COMMAND"
        exit $?
    fi
fi

# Single node mode: Get the IP for the requested node
NODE_IP=""
if [ "$NODE_NUM" -lt "${#CLUSTER_NODE_IPS[@]}" ]; then
    NODE_IP="${CLUSTER_NODE_IPS[$NODE_NUM]}"
else
    echo "Error: Node $NODE_NUM does not exist in cluster $CLUSTER_ID" >&2
    echo "Available nodes: 0 to $((${#CLUSTER_NODE_IPS[@]} - 1))" >&2
    exit 1
fi

if [ -z "$NODE_IP" ]; then
    echo "Error: No IP address found for node $NODE_NUM" >&2
    exit 1
fi

# SSH to the node
echo "Connecting to node $NODE_NUM ($NODE_IP) in cluster $CLUSTER_ID..." >&2
exec ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -o LogLevel=ERROR \
    -i "$SSH_KEY" "${SSH_USER}@${NODE_IP}" "$@"