groups: - name: redis-labs-alerts rules: - alert: Throughput Greater Than 15k expr: (bdb_total_req{bdb="10801408"}) offset 10m > 15000 for: 1m labels: severity: critical annotations: summary: "Throughput of (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 15000" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: Throughput Greater Than 25k expr: (bdb_total_req{bdb=~"10810299|10810324|10621595|11094293"}) offset 10m > 25000 for: 1m labels: severity: critical annotations: summary: "Throughput of (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 25000" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: Throughput Greater Than 50k expr: (bdb_total_req{bdb=~"11029271|10772171"}) offset 10m > 50000 for: 1m labels: severity: critical annotations: summary: "Throughput of (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 50000" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: Throughput Greater Than 89k expr: (bdb_total_req{bdb=~"10790599"}) offset 10m > 89000 for: 1m labels: severity: critical annotations: summary: "Throughput of (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 89000" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: Throughput Greater Than 200k expr: (bdb_total_req{bdb=~"11029236"}) offset 10m > 200000 for: 1m labels: severity: critical annotations: summary: "Throughput of (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 200000" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: bdb_avg_latency_greather_than_1ms_Default expr: (bdb_avg_latency) offset 10m > 0.1 for: 1m labels: severity: critical annotations: summary: "Throughput of Cluster {{ $labels.cluster }}, (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 100 Milli Second" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: bdb_avg_latency_greather_than_10ms expr: (bdb_avg_latency{bdb=~"11029271|11094293"}) offset 10m > 0.01 for: 1m labels: severity: critical annotations: summary: "Throughput of Cluster {{ $labels.cluster }}, (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 10 Milli Second" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: bdb_avg_latency_greather_than_50ms expr: (bdb_avg_latency{bdb=~"11029236|10621595|10790599"}) offset 10m > 0.05 for: 1m labels: severity: critical annotations: summary: "Throughput of Cluster {{ $labels.cluster }}, (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 50 Milli Second" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: bdb_avg_latency_greather_than_100ms expr: (bdb_avg_latency{bdb=~"10772171|10810299|10810324"}) offset 10m > 0.1 for: 1m labels: severity: critical annotations: summary: "Throughput of Cluster {{ $labels.cluster }}, (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 100 Milli Second" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: bdb_memory_usage_greaterthan_80% expr: avg by (cluster,instance)((bdb_used_memory{} / bdb_memory_limit{}) )* 100 offset 10m > 80 for: 1m labels: severity: critical annotations: summary: "Throughput of Cluster {{ $labels.cluster }}, (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 100 Milli Second" description: "Throughput of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: Node Free Memory Less than 2GB expr: (node_free_memory / 1073741824) offset 10m < 2 for: 1m labels: severity: critical annotations: summary: "node_free_memory of Cluster {{ $labels.cluster}}, (instance {{ $labels.instance }}) and for the node ( Node ID : {{ $labels.node }}) is less than 2GB" description: "node_free_memory of (Node ID : {{ $labels.node }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: Redis Shard CPU usage percent greater than 70%_Default expr: (redis_process_cpu_usage_percent) offset 10m > 70 for: 1m labels: severity: critical annotations: summary: "Redis Shard CPU Usage of (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 70%" description: "Redis Shard CPU Usage of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}" - alert: redis shard connections greater than 6000_Default expr: (redis_connected_clients) offset 10m > 6000 for: 1m labels: severity: critical annotations: summary: "redis shard connections of (instance {{ $labels.instance }}) and for the database (bdb ID : {{ $labels.bdb }}) is greater than 6000 connections" description: "redis shard connections of (bdb ID : {{ $labels.bdb }}) ( VALUE = {{ $value }}\n LABELS: {{ $labels }}"