| 1 | |
| 2 | # Warn if a list keys operation is running. |
| 3 | template: riakkv_list_keys_active |
| 4 | on: riak.core.fsm_active |
| 5 | class: Utilization |
| 6 | type: Database |
| 7 | component: Riak KV |
| 8 | calc: $list_fsm_active |
| 9 | units: state machines |
| 10 | every: 10s |
| 11 | warn: $list_fsm_active > 0 |
| 12 | summary: Riak KV active list keys |
| 13 | info: Number of currently running list keys finite state machines |
| 14 | to: dba |
| 15 | |
| 16 | |
| 17 | ## Timing healthchecks |
| 18 | # KV GET |
| 19 | template: riakkv_1h_kv_get_mean_latency |
| 20 | on: riak.kv.latency.get |
| 21 | class: Latency |
| 22 | type: Database |
| 23 | component: Riak KV |
| 24 | calc: $node_get_fsm_time_mean |
| 25 | lookup: average -1h unaligned of time |
| 26 | every: 30s |
| 27 | units: ms |
| 28 | info: average time between reception of client GET request and \ |
| 29 | subsequent response to client over the last hour |
| 30 | |
| 31 | template: riakkv_kv_get_slow |
| 32 | on: riak.kv.latency.get |
| 33 | class: Latency |
| 34 | type: Database |
| 35 | component: Riak KV |
| 36 | calc: $mean |
| 37 | lookup: average -3m unaligned of time |
| 38 | units: ms |
| 39 | every: 10s |
| 40 | warn: ($this > ($riakkv_1h_kv_get_mean_latency * 2) ) |
| 41 | crit: ($this > ($riakkv_1h_kv_get_mean_latency * 3) ) |
| 42 | summary: Riak KV GET latency |
| 43 | info: Average time between reception of client GET request and \ |
| 44 | subsequent response to the client over the last 3 minutes, \ |
| 45 | compared to the average over the last hour |
| 46 | delay: down 5m multiplier 1.5 max 1h |
| 47 | to: dba |
| 48 | |
| 49 | # KV PUT |
| 50 | template: riakkv_1h_kv_put_mean_latency |
| 51 | on: riak.kv.latency.put |
| 52 | class: Latency |
| 53 | type: Database |
| 54 | component: Riak KV |
| 55 | calc: $node_put_fsm_time_mean |
| 56 | lookup: average -1h unaligned of time |
| 57 | every: 30s |
| 58 | units: ms |
| 59 | summary: Riak KV PUT mean latency |
| 60 | info: Average time between reception of client PUT request and \ |
| 61 | subsequent response to the client over the last hour |
| 62 | |
| 63 | template: riakkv_kv_put_slow |
| 64 | on: riak.kv.latency.put |
| 65 | class: Latency |
| 66 | type: Database |
| 67 | component: Riak KV |
| 68 | calc: $mean |
| 69 | lookup: average -3m unaligned of time |
| 70 | units: ms |
| 71 | every: 10s |
| 72 | warn: ($this > ($riakkv_1h_kv_put_mean_latency * 2) ) |
| 73 | crit: ($this > ($riakkv_1h_kv_put_mean_latency * 3) ) |
| 74 | summary: Riak KV PUT latency |
| 75 | info: Average time between reception of client PUT request and \ |
| 76 | subsequent response to the client over the last 3 minutes, \ |
| 77 | compared to the average over the last hour |
| 78 | delay: down 5m multiplier 1.5 max 1h |
| 79 | to: dba |
| 80 | |
| 81 | |
| 82 | ## VM healthchecks |
| 83 | |
| 84 | # Default Erlang VM process limit: 262144 |
| 85 | # On systems observed, this is < 2000, but may grow depending on load. |
| 86 | template: riakkv_vm_high_process_count |
| 87 | on: riak.vm |
| 88 | class: Utilization |
| 89 | type: Database |
| 90 | component: Riak KV |
| 91 | calc: $sys_process_count |
| 92 | units: processes |
| 93 | every: 10s |
| 94 | warn: $this > 10000 |
| 95 | crit: $this > 100000 |
| 96 | summary: Riak KV number of processes |
| 97 | info: Number of processes running in the Erlang VM |
| 98 | to: dba |