icinga2_hw_service_checks_example.conf

/*******************************
    Server Hardware Checks
*******************************/
apply Service for (service => request_command in {
        "HW Storage" = "storage",
        "HW Processor" = "proc",
        "HW DIMM" = "memory",
        "HW Power" = "power",
        "HW Temperature" = "temp",
        "HW Fan" = "fan",
        "HW NICs" = "nic",
        "HW BMC" = "bmc",
        "HW System Summary" = "info",
        "HW Firmware" = "firmware",
        "HW Management Log" = "mel",
        "HW System Event Log" = "sel",
        "HW Inventory" = "inventory"
    }){

    import "generic-service"

    check_command = "redfish"

    // increase timeout, older BMCs can be very slow
    check_timeout = 60
    vars.redfish_retries = 3
    vars.redfish_timeout = 13

    max_check_attempts = 10
    retry_interval = 1m

    vars.redfish_detailed = false

    // enable Perfdata
    if (request_command in [ "temp", "fan", "power" ]) {
        enable_perfdata = true
    }

    // add detailed view
    if (request_command in [ "bmc", "firmware", "mel", "sel", "storage" ]) {
        vars.redfish_detailed = true
    }

    // request logs only every 30 minutes
    if (request_command in [ "mel", "sel" ]) {
        check_interval = 30m
    }

    // update firmware info only four times a day
    if (request_command in [ "firmware" ]) {
        check_interval = 6h
    }

    // get system inventory only four times a day
    if (request_command in [ "inventory" ]) {
        check_interval = 6h
        check_timeout = 180

        vars.redfish_all = true
        vars.redfish_inventory = true
        vars.redfish_inventory_id = host.vars.inventory_source_id

        // inventory file to write inventory to
        vars.redfish_inventory_file = "/var/tmp/check_redfish_inventory/inventory_" + host.name + ".json"
    }

    /*
      set warning and critical
      log entries > info which are occurred within the last 24h are reported as CRITICAL
      log entries > info which occurred between the last 24h and 72h are reported as WARNING
    */
    if (request_command in [ "mel", "sel" ]) {
        vars.warning = 3
        vars.critical = 1
    }

    // credentials
    if (match("ntp*", host.name)) {
        vars.redfish_env_username = "icinga"
        vars.redfish_env_password = "SUPERSECRET"
    } else {
        vars.redfish_authfile = "/etc/icinga2/check_redfish_icinga.login"
    }

    vars.redfish_hostname = host.vars.console

    if (request_command == "storage") { vars.redfish_storage = true }
    if (request_command == "proc") { vars.redfish_proc = true }
    if (request_command == "memory") { vars.redfish_memory = true }
    if (request_command == "power") { vars.redfish_power = true }
    if (request_command == "temp") { vars.redfish_temp = true }
    if (request_command == "fan") { vars.redfish_fan = true }
    if (request_command == "nic") { vars.redfish_nic = true }
    if (request_command == "bmc") { vars.redfish_bmc = true }
    if (request_command == "info") { vars.redfish_info = true }
    if (request_command == "firmware") { vars.redfish_firmware = true }
    if (request_command == "mel") { vars.redfish_mel = true }
    if (request_command == "sel") { vars.redfish_sel = true }

    // ignore missing power supplies on Blade Server
    if (match("HP BL*",host.vars.hw_model)) {
        vars.ignore_missing_ps = true
    }

    assign where host.vars.hw_type == "HP_Proliant" && host.vars.console
}

// EOF