update some metrics names for alerts and dashboards

libra -> aptos mempool has no aptos prefix Closes: aptos-labs#187
GS- · Mar 15, 2022 · 5c83cef · 5c83cef
1 parent 03ab897
commit 5c83cef
Show file tree

Hide file tree

Showing 11 changed files with 40 additions and 40 deletions.
diff --git a/terraform/helm/validator/files/dashboards/components.json b/terraform/helm/validator/files/dashboards/components.json
@@ -2021,7 +2021,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_index_size{role=~\"$role\",owner=~\"$owner\",index=\"system_ttl\"}",
+          "expr": "core_mempool_index_size{role=~\"$role\",owner=~\"$owner\",index=\"system_ttl\"}",
           "format": "time_series",
           "hide": false,
           "instant": false,

diff --git a/terraform/helm/validator/files/dashboards/components.json.gz b/terraform/helm/validator/files/dashboards/components.json.gz
diff --git a/terraform/helm/validator/files/dashboards/fullnodes.json b/terraform/helm/validator/files/dashboards/fullnodes.json
@@ -451,14 +451,14 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_index_size{index=\"system_ttl\", kubernetes_pod_name=~\".*fullnode.*\", owner=~\"$owner\"}",
+          "expr": "core_mempool_index_size{index=\"system_ttl\", kubernetes_pod_name=~\".*fullnode.*\", owner=~\"$owner\"}",
           "interval": "",
           "legendFormat": "{{owner}}-{{kubernetes_pod_name}}",
           "refId": "A"
         },
         {
           "exemplar": true,
-          "expr": "aptos_core_mempool_index_size{index=\"system_ttl\", job=~\".*fullnode.*\", owner=~\"$owner\"}",
+          "expr": "core_mempool_index_size{index=\"system_ttl\", job=~\".*fullnode.*\", owner=~\"$owner\"}",
           "hide": false,
           "interval": "",
           "legendFormat": "{{owner}}-{{job}}",
@@ -537,14 +537,14 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_txn_commit_latency_sum{kubernetes_pod_name=~\".*fullnode.*\", owner=~\"$owner\"} / aptos_core_mempool_txn_commit_latency_count{kubernetes_pod_name=~\".*fullnode.*\", owner=~\"$owner\"}",
+          "expr": "core_mempool_txn_commit_latency_sum{kubernetes_pod_name=~\".*fullnode.*\", owner=~\"$owner\"} / core_mempool_txn_commit_latency_count{kubernetes_pod_name=~\".*fullnode.*\", owner=~\"$owner\"}",
           "interval": "",
           "legendFormat": "{{owner}}-{{kubernetes_pod_name}}",
           "refId": "A"
         },
         {
           "exemplar": true,
-          "expr": "aptos_core_mempool_txn_commit_latency_sum{job=~\".*fullnode.*\", owner=~\"$owner\"} / aptos_core_mempool_txn_commit_latency_count{job=~\".*fullnode.*\", owner=~\"$owner\"}",
+          "expr": "core_mempool_txn_commit_latency_sum{job=~\".*fullnode.*\", owner=~\"$owner\"} / core_mempool_txn_commit_latency_count{job=~\".*fullnode.*\", owner=~\"$owner\"}",
           "hide": false,
           "interval": "",
           "legendFormat": "{{owner}}-{{job}}",

diff --git a/terraform/helm/validator/files/dashboards/fullnodes.json.gz b/terraform/helm/validator/files/dashboards/fullnodes.json.gz
diff --git a/terraform/helm/validator/files/dashboards/mempool.json b/terraform/helm/validator/files/dashboards/mempool.json
@@ -95,7 +95,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_index_size{index=\"system_ttl\", owner=~\"$owner\", role=~\"$role\"}",
+          "expr": "core_mempool_index_size{index=\"system_ttl\", owner=~\"$owner\", role=~\"$role\"}",
           "format": "time_series",
           "hide": false,
           "instant": false,
@@ -296,7 +296,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_core_mempool_removed_txns_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(core_mempool_removed_txns_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -393,7 +393,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "sum by (owner, role, network) (aptos_mempool_active_upstream_peers_count{owner=~\"$owner\", role=~\"$role\"})",
+          "expr": "sum by (owner, role, network) (mempool_active_upstream_peers_count{owner=~\"$owner\", role=~\"$role\"})",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}  ({{network}} network)",
           "refId": "A"
@@ -601,7 +601,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_service_latency_ms_count{result=\"fail\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_service_latency_ms_count{result=\"fail\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}} - {{type}}",
           "refId": "A"
@@ -698,7 +698,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_db_error_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_db_error_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -795,7 +795,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_json_rpc_callback_fail_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_json_rpc_callback_fail_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -892,7 +892,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_network_send_fail_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_network_send_fail_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -989,7 +989,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_vm_reconfig_update_fail_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_vm_reconfig_update_fail_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -1086,7 +1086,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_core_mempool_invariant_violated_count{owner=\"$owner\", role=\"$role\"}[$interval])",
+          "expr": "rate(mempool_core_mempool_invariant_violated_count{owner=\"$owner\", role=\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -1183,7 +1183,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_unexpected_network_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_unexpected_network_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}} <- {{network}}:{{peer}}",
           "refId": "A"
@@ -1294,7 +1294,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_core_mempool_gc_latency_count{type=\"client_expiration\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(core_mempool_gc_latency_count{type=\"client_expiration\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}} : {{status}}",
           "refId": "A"
@@ -1391,7 +1391,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_core_mempool_gc_latency_count{type=\"system_ttl\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(core_mempool_gc_latency_count{type=\"system_ttl\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}} : {{status}}",
           "refId": "A"
@@ -1488,7 +1488,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_index_size{index=\"parking_lot\", owner=~\"$owner\", role=~\"$role\"}",
+          "expr": "core_mempool_index_size{index=\"parking_lot\", owner=~\"$owner\", role=~\"$role\"}",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -1585,7 +1585,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_index_size{index=\"timeline\", owner=~\"$owner\", role=~\"$role\"}",
+          "expr": "core_mempool_index_size{index=\"timeline\", owner=~\"$owner\", role=~\"$role\"}",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -1682,7 +1682,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_index_size{index=\"priority\", role=\"validator\", owner=~\"$owner\"}",
+          "expr": "core_mempool_index_size{index=\"priority\", role=\"validator\", owner=~\"$owner\"}",
           "interval": "",
           "legendFormat": "{{owner}}",
           "refId": "A"
@@ -1876,7 +1876,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_core_mempool_gc_event_count{type=\"system_ttl\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(core_mempool_gc_event_count{type=\"system_ttl\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -1973,7 +1973,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_core_mempool_gc_latency_sum{owner=~\"$owner\", role=~\"$role\"}[$interval])/rate(aptos_core_mempool_gc_latency_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(core_mempool_gc_latency_sum{owner=~\"$owner\", role=~\"$role\"}[$interval])/rate(core_mempool_gc_latency_count{owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -2085,7 +2085,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_service_transactions_sum{type=\"commit_accepted\", owner=~\"$owner\", role=~\"$role\"}[$interval])\n",
+          "expr": "rate(mempool_service_transactions_sum{type=\"commit_accepted\", owner=~\"$owner\", role=~\"$role\"}[$interval])\n",
           "format": "time_series",
           "interval": "",
           "intervalFactor": 1,
@@ -2186,7 +2186,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_service_latency_ms_sum{type=\"commit_accepted\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(aptos_mempool_service_latency_ms_count{type=\"commit_accepted\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_service_latency_ms_sum{type=\"commit_accepted\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(mempool_service_latency_ms_count{type=\"commit_accepted\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "format": "time_series",
           "interval": "",
           "intervalFactor": 1,
@@ -2285,7 +2285,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_service_transactions_sum{type=\"commit_rejected\", owner=~\"$owner\", role=~\"$role\"}[$interval])\n",
+          "expr": "rate(mempool_service_transactions_sum{type=\"commit_rejected\", owner=~\"$owner\", role=~\"$role\"}[$interval])\n",
           "interval": "",
           "legendFormat": "{{owner}}",
           "refId": "A"
@@ -2382,7 +2382,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_service_latency_ms_sum{type=\"commit_rejected\", owner=~\"$owner\"}[$interval])/rate(aptos_mempool_service_latency_ms_count{type=\"commit_rejected\", owner=~\"$owner\"}[$interval])",
+          "expr": "rate(mempool_service_latency_ms_sum{type=\"commit_rejected\", owner=~\"$owner\"}[$interval])/rate(mempool_service_latency_ms_count{type=\"commit_rejected\", owner=~\"$owner\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}",
           "refId": "A"
@@ -2480,7 +2480,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_service_transactions_sum{type=\"get_block\", owner=~\"$owner\"}[$interval]) / rate(aptos_mempool_service_transactions_count{type=\"get_block\", owner=~\"$owner\"}[$interval])",
+          "expr": "rate(mempool_service_transactions_sum{type=\"get_block\", owner=~\"$owner\"}[$interval]) / rate(mempool_service_transactions_count{type=\"get_block\", owner=~\"$owner\"}[$interval])",
           "format": "time_series",
           "interval": "",
           "intervalFactor": 1,
@@ -2580,7 +2580,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_service_latency_ms_sum{type=\"get_block\", owner=~\"$owner\"}[$interval]) / rate(aptos_mempool_service_latency_ms_count{type=\"get_block\", owner=~\"$owner\"}[$interval])",
+          "expr": "rate(mempool_service_latency_ms_sum{type=\"get_block\", owner=~\"$owner\"}[$interval]) / rate(mempool_service_latency_ms_count{type=\"get_block\", owner=~\"$owner\"}[$interval])",
           "format": "time_series",
           "interval": "",
           "intervalFactor": 1,
@@ -2793,7 +2793,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_bounded_executor_spawn_latency_sum{stage=\"start\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(aptos_mempool_bounded_executor_spawn_latency_count{stage=\"start\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_bounded_executor_spawn_latency_sum{stage=\"start\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(mempool_bounded_executor_spawn_latency_count{stage=\"start\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}} - {{task}}",
           "refId": "A"
@@ -3202,7 +3202,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_process_txn_breakdown_latency_sum{portion=\"vm_validation\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(aptos_mempool_process_txn_breakdown_latency_count{portion=\"vm_validation\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_process_txn_breakdown_latency_sum{portion=\"vm_validation\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(mempool_process_txn_breakdown_latency_count{portion=\"vm_validation\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"
@@ -3299,7 +3299,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_mempool_process_txn_breakdown_latency_sum{portion=\"storage_fetch\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(aptos_mempool_process_txn_breakdown_latency_count{portion=\"storage_fetch\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
+          "expr": "rate(mempool_process_txn_breakdown_latency_sum{portion=\"storage_fetch\", owner=~\"$owner\", role=~\"$role\"}[$interval]) / rate(mempool_process_txn_breakdown_latency_count{portion=\"storage_fetch\", owner=~\"$owner\", role=~\"$role\"}[$interval])",
           "interval": "",
           "legendFormat": "{{owner}}:{{role}}",
           "refId": "A"

diff --git a/terraform/helm/validator/files/dashboards/mempool.json.gz b/terraform/helm/validator/files/dashboards/mempool.json.gz
diff --git a/terraform/helm/validator/files/dashboards/overview.json b/terraform/helm/validator/files/dashboards/overview.json
@@ -1610,7 +1610,7 @@
           "steppedLine": false,
           "targets": [
             {
-              "expr": "aptos_core_mempool_index_size{index=\"system_ttl\", owner=~\"$owner\"}",
+              "expr": "core_mempool_index_size{index=\"system_ttl\", owner=~\"$owner\"}",
               "format": "time_series",
               "hide": false,
               "instant": false,
@@ -1701,7 +1701,7 @@
           "steppedLine": false,
           "targets": [
             {
-              "expr": "rate(aptos_core_mempool_txn_commit_latency_sum{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])/rate(aptos_core_mempool_txn_commit_latency_count{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])",
+              "expr": "rate(core_mempool_txn_commit_latency_sum{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])/rate(core_mempool_txn_commit_latency_count{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])",
               "format": "time_series",
               "interval": "",
               "intervalFactor": 1,

diff --git a/terraform/helm/validator/files/dashboards/overview.json.gz b/terraform/helm/validator/files/dashboards/overview.json.gz
diff --git a/terraform/helm/validator/files/dashboards/validator.json b/terraform/helm/validator/files/dashboards/validator.json
@@ -1060,7 +1060,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(aptos_core_mempool_txn_commit_latency_sum{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])/rate(aptos_core_mempool_txn_commit_latency_count{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])",
+          "expr": "rate(core_mempool_txn_commit_latency_sum{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])/rate(core_mempool_txn_commit_latency_count{stage=~\"commit_accepted|commit_rejected\", owner=~\"$owner\"}[1m])",
           "format": "time_series",
           "interval": "",
           "intervalFactor": 1,
@@ -1159,7 +1159,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "aptos_core_mempool_index_size{index=\"system_ttl\", owner=~\"$owner\"}",
+          "expr": "core_mempool_index_size{index=\"system_ttl\", owner=~\"$owner\"}",
           "format": "time_series",
           "interval": "",
           "intervalFactor": 1,

diff --git a/terraform/helm/validator/files/dashboards/validator.json.gz b/terraform/helm/validator/files/dashboards/validator.json.gz
diff --git a/terraform/helm/validator/files/rules/alerts.yml b/terraform/helm/validator/files/rules/alerts.yml
@@ -10,14 +10,14 @@ groups:
       summary: "The block commit rate is low"
     annotations:
   - alert: High local timeout rate
-    expr: rate(libra_consensus_timeout_count{role="validator"}[1m]) > 0.5
+    expr: rate(aptos_consensus_timeout_count{role="validator"}[1m]) > 0.5
     for: 20m
     labels:
       severity: warning
       summary: "Consensus timeout rate is high"
     annotations:
   - alert: High consensus error rate
-    expr: rate(libra_consensus_error_count{role="validator"}[1m]) / on (role) rate(consensus_duration_count{op='main_loop', role="validator"}[1m]) > 0.25
+    expr: rate(aptos_consensus_error_count{role="validator"}[1m]) / on (role) rate(consensus_duration_count{op='main_loop', role="validator"}[1m]) > 0.25
     for: 20m
     labels:
       severity: warning
@@ -42,21 +42,21 @@ groups:
 
     # Mempool alerts
   - alert: Mempool has no active upstream peers
-    expr: (sum by (owner, kubernetes_pod_name) (aptos_mempool_active_upstream_peers_count)) == 0
+    expr: (sum by (owner, kubernetes_pod_name) (mempool_active_upstream_peers_count)) == 0
     for: 3m
     labels:
       severity: error
       summary: "Mempool has no active upstream peers (unable to forward transactions to anyone!)"
     annotations:
   - alert: Mempool is at >80% capacity
-    expr: aptos_core_mempool_index_size{index="system_ttl"} > 800000 # assumes default mempool size 1_000_000
+    expr: core_mempool_index_size{index="system_ttl"} > 800000 # assumes default mempool size 1_000_000
     for: 5m
     labels:
       severity: warning
       summary: "Mempool is at >80% capacity (it may soon become full!)"
     annotations:
   - alert: Mempool is growing at a significant rate
-    expr: rate(aptos_core_mempool_index_size{index="system_ttl"}[1m]) > 30000
+    expr: rate(core_mempool_index_size{index="system_ttl"}[1m]) > 30000
     for: 10m
     labels:
       severity: warning