| 1 | # you can disable an alarm notification by setting the 'to' line to: silent |
| 2 | |
| 3 | template: rabbitmq_node_avail_status_down |
| 4 | on: rabbitmq.node_avail_status |
| 5 | class: Errors |
| 6 | type: Messaging |
| 7 | component: RabbitMQ |
| 8 | calc: $down |
| 9 | every: 10s |
| 10 | units: status |
| 11 | crit: $this > 0 |
| 12 | summary: RabbitMQ node is down (node ${label:node} cluster ${label:cluster_id}) |
| 13 | info: RabbitMQ node is down (node ${label:node} cluster ${label:cluster_id}) |
| 14 | delay: down 1m |
| 15 | to: sysadmin |
| 16 | |
| 17 | template: rabbitmq_node_network_partition_status |
| 18 | on: rabbitmq.node_network_partition_status |
| 19 | class: Errors |
| 20 | type: Messaging |
| 21 | component: RabbitMQ |
| 22 | calc: $detected |
| 23 | every: 10s |
| 24 | units: status |
| 25 | crit: $this > 0 |
| 26 | summary: RabbitMQ network partition detected (node ${label:node} cluster ${label:cluster_id}) |
| 27 | info: RabbitMQ network partition detected (node ${label:node} cluster ${label:cluster_id}) |
| 28 | delay: down 1m |
| 29 | to: sysadmin |
| 30 | |
| 31 | template: rabbitmq_node_mem_alarm_status_triggered |
| 32 | on: rabbitmq.node_mem_alarm_status |
| 33 | class: Errors |
| 34 | type: Messaging |
| 35 | component: RabbitMQ |
| 36 | calc: $triggered |
| 37 | every: 10s |
| 38 | units: status |
| 39 | warn: $this > 0 |
| 40 | summary: RabbitMQ mem alarm triggered (node ${label:node} cluster ${label:cluster_id}) |
| 41 | info: RabbitMQ mem alarm triggered (node ${label:node} cluster ${label:cluster_id}) |
| 42 | delay: down 1m |
| 43 | to: sysadmin |
| 44 | |
| 45 | template: rabbitmq.node_disk_free_alarm_status_triggered |
| 46 | on: rabbitmq.node_disk_free_alarm_status |
| 47 | class: Errors |
| 48 | type: Messaging |
| 49 | component: RabbitMQ |
| 50 | calc: $triggered |
| 51 | every: 10s |
| 52 | units: status |
| 53 | warn: $this > 0 |
| 54 | summary: RabbitMQ disk free alarm triggered (node ${label:node} cluster ${label:cluster_id}) |
| 55 | info: RabbitMQ disk free alarm triggered (node ${label:node} cluster ${label:cluster_id}) |
| 56 | delay: down 1m |
| 57 | to: sysadmin |
| 58 | |
| 59 | template: rabbitmq_vhost_status_unhealthy |
| 60 | on: rabbitmq.vhost_status |
| 61 | class: Errors |
| 62 | type: Messaging |
| 63 | component: RabbitMQ |
| 64 | calc: $stopped + $partial |
| 65 | every: 10s |
| 66 | units: status |
| 67 | warn: $this > 0 |
| 68 | summary: RabbitMQ vhost is unhealthy (vhost ${label:vhost} cluster ${label:cluster_id}) |
| 69 | info: RabbitMQ vhost is unhealthy (vhost ${label:vhost} cluster ${label:cluster_id}) |
| 70 | delay: down 1m |
| 71 | to: sysadmin |
| 72 | |
| 73 | template: rabbitmq_queue_status_minority |
| 74 | on: rabbitmq.queue_status |
| 75 | class: Errors |
| 76 | type: Messaging |
| 77 | component: RabbitMQ |
| 78 | calc: $minority |
| 79 | every: 10s |
| 80 | units: status |
| 81 | warn: $this > 0 |
| 82 | summary: RabbitMQ queue insufficient online members (queue ${label:queue} node ${label:node} cluster ${label:cluster_id}) |
| 83 | info: RabbitMQ queue insufficient online members (queue ${label:queue} node ${label:node} cluster ${label:cluster_id}) |
| 84 | delay: down 1m |
| 85 | to: sysadmin |
| 86 | |
| 87 | template: rabbitmq_queue_status_unhealthy |
| 88 | on: rabbitmq.queue_status |
| 89 | class: Errors |
| 90 | type: Messaging |
| 91 | component: RabbitMQ |
| 92 | calc: $stopped + $crashed + $down |
| 93 | every: 10s |
| 94 | units: status |
| 95 | crit: $this > 0 |
| 96 | summary: RabbitMQ queue is unhealthy (queue ${label:queue} node ${label:node} cluster ${label:cluster_id}) |
| 97 | info: RabbitMQ queue is unhealthy (queue ${label:queue} node ${label:node} cluster ${label:cluster_id}) |
| 98 | delay: down 1m |
| 99 | to: sysadmin |