zulip/servers/puppet/modules/humbug/files/nagios3/conf.d/services.cfg

define service {
        use                              generic-service
        hostgroup_name                   web
        service_description              HTTP
        check_command                    check_http
        contact_groups                   paging_admins
}

define service {
        use                             generic-service
        hostgroup_name                  all
        service_description             SSH
        check_command                   check_ssh
        contact_groups                  paging_admins
}

define service {
        use                             generic-service
        hostgroup_name                  all
        service_description             remote disk
        check_command                   check_remote_disk!22!20%!10%!/
        contact_groups                  paging_admins
}

define service {
        use                             generic-service
        hostgroup_name                  all
        service_description             remote load
        check_command                   check_remote_load!22!5.0,4.0,3.0!10.0,6.0,4.0
        contact_groups                  paging_admins
}

define service {
        use                             generic-service
        service_description             zephyr mirror forwarding
        check_command                   check_zephyr_mirror_forwarding!22
        host                            zmirror
        normal_check_interval           2
# Note: the actual check is run via cron, so retry_check_interval
# should always equal normal_check_interval.
        retry_check_interval            2
        max_check_attempts              2
        contact_groups                  paging_admins
}

define service {
        use                             generic-service
        service_description             user zephyr mirror liveness
        check_command                   check_user_zephyr_mirror_liveness!22
        host                            app
        normal_check_interval           2
        retry_check_interval            1
        max_check_attempts              2
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Debian update availability
        check_command                   check_debian_packages!22
        hostgroup_name                  all
        max_check_attempts              1
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check NTP time
        check_command                   check_ntp_time!22
        hostgroup_name                  all
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check feedback bot
        check_command                   check_feedback_bot!22
        host                            bots
        # Feedback isn't lost, it just isn't reported through Humbug
        # if the bot is down, so don't page.
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check send receive time
        check_command                   check_send_receive_time!22
        host                            staging
        contact_groups                  admins
}

## check_postgres.pl services

define service {
        use                             generic-service
        service_description             Check postgres autovac_freeze
        check_command                   check_postgres!22!autovac_freeze
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres backends
        check_command                   check_postgres!22!backends
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres connection
        check_command                   check_postgres!22!connection
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres hitratio
        check_command                   check_postgres!22!hitratio
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres locks
        check_command                   check_postgres!22!locks
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres query_time
        check_command                   check_postgres_alert_args!22!query_time!20 seconds!40 seconds
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres sequence
        check_command                   check_postgres!22!sequence
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres timesync
        check_command                   check_postgres!22!timesync
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres txn_idle
        check_command                   check_postgres_alert_args!22!txn_idle!20 seconds!40 seconds
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             Check postgres txn_time
        check_command                   check_postgres_alert_args!22!txn_time!20 seconds!40 seconds
        host                            postgres
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             process_user_activity bot
        check_command                   check_process_user_activity!22
        hostgroup_name                  frontends
        # Activity isn't lost if this isn't running, but it is
        # critical that it arrive to avoid falsely telling MIT folks
        # their mirrors are down, so this should be changed to
        # page_admins sometime after a week of no false alerts
        # This service is also responsible for active/idle
        # status.
        contact_groups                  admins
}

define service {
        use                             generic-service
        service_description             zmirror subscriptons syncing
        check_command                   check_sync_public_streams!22
        hostgroup_name                  zmirror
        # Change this to page_admins after a week
        contact_groups                  admins
}