update mixin

Signed-off-by: Xiang Dai <764524258@qq.com>
thanos-io · Feb 20, 2020 · 4bda22a · 4bda22a
1 parent cd8a067
commit 4bda22a
Show file tree

Hide file tree

Showing 12 changed files with 115 additions and 22 deletions.
diff --git a/examples/alerts/alerts.md b/examples/alerts/alerts.md
@@ -428,6 +428,48 @@ rules:
     severity: warning
 ```
 
+## Replicate
+
+[embedmd]:# (../tmp/thanos-replicate.rules.yaml yaml)
+```yaml
+name: thanos-replicate.rules
+rules:
+- alert: ThanosReplicateIsDown
+  annotations:
+    message: Thanos Replicate has disappeared from Prometheus target discovery.
+  expr: |
+    absent(up{job=~"thanos-replicate.*"})
+  for: 5m
+  labels:
+    severity: critical
+- alert: ThanosReplicateErrorRate
+  annotations:
+    message: Thanos Replicate failing to run, {{ $value | humanize }}% of attempts
+      failed.
+  expr: |
+    (
+      sum(rate(thanos_replicate_replication_runs_total{result="error", job=~"thanos-replicate.*"}[5m]))
+    / on (namespace) group_left
+      sum(rate(thanos_replicate_replication_runs_total{job=~"thanos-replicate.*"}[5m]))
+    ) * 100 >= 10
+  for: 5m
+  labels:
+    severity: critical
+- alert: ThanosReplicateRunLatency
+  annotations:
+    message: Thanos Replicate {{$labels.job}} has a 99th percentile latency of {{
+      $value }} seconds for the replicate operations.
+  expr: |
+    (
+      histogram_quantile(0.9, sum by (job, le) (thanos_replicate_replication_run_duration_seconds_bucket{job=~"thanos-replicate.*"})) > 120
+    and
+      sum by (job) (rate(thanos_replicate_replication_run_duration_seconds_bucket{job=~"thanos-replicate.*"}[5m])) > 0
+    )
+  for: 5m
+  labels:
+    severity: critical
+```
+
 ## Extras
 
 ### Absent Rules

diff --git a/examples/alerts/alerts.yaml b/examples/alerts/alerts.yaml
@@ -439,3 +439,39 @@ groups:
     for: 5m
     labels:
       severity: critical
+- name: thanos-replicate.rules
+  rules:
+  - alert: ThanosReplicateIsDown
+    annotations:
+      message: Thanos Replicate has disappeared from Prometheus target discovery.
+    expr: |
+      absent(up{job=~"thanos-replicate.*"})
+    for: 5m
+    labels:
+      severity: critical
+  - alert: ThanosReplicateErrorRate
+    annotations:
+      message: Thanos Replicate failing to run, {{ $value | humanize }}% of attempts
+        failed.
+    expr: |
+      (
+        sum(rate(thanos_replicate_replication_runs_total{result="error", job=~"thanos-replicate.*"}[5m]))
+      / on (namespace) group_left
+        sum(rate(thanos_replicate_replication_runs_total{job=~"thanos-replicate.*"}[5m]))
+      ) * 100 >= 10
+    for: 5m
+    labels:
+      severity: critical
+  - alert: ThanosReplicateRunLatency
+    annotations:
+      message: Thanos Replicate {{$labels.job}} has a 99th percentile latency of {{
+        $value }} seconds for the replicate operations.
+    expr: |
+      (
+        histogram_quantile(0.9, sum by (job, le) (thanos_replicate_replication_run_duration_seconds_bucket{job=~"thanos-replicate.*"})) > 120
+      and
+        sum by (job) (rate(thanos_replicate_replication_run_duration_seconds_bucket{job=~"thanos-replicate.*"}[5m])) > 0
+      )
+    for: 5m
+    labels:
+      severity: critical
diff --git a/examples/alerts/rules.yaml b/examples/alerts/rules.yaml
@@ -121,3 +121,5 @@ groups:
     labels:
       quantile: "0.99"
     record: :thanos_objstore_bucket_operation_duration_seconds:histogram_quantile
+- name: thanos-replicate.rules
+  rules: []
diff --git a/examples/dashboards/dashboards.md b/examples/dashboards/dashboards.md
@@ -9,6 +9,7 @@ There exists Grafana dashboards for each component (not all of them complete) ta
 - [Thanos Receiver](thanos-receiver.json)
 - [Thanos Sidecar](thanos-sidecar.json)
 - [Thanos Ruler](thanos-ruler.json)
+- [Thanos Replicate](thanos-replicate.json)
 
 You can import them via `Import -> Paste JSON` in Grafana.
 These dashboards require Grafana 5 or above, importing them in older versions are known not to work.

diff --git a/mixin/thanos/README.md b/mixin/thanos/README.md
@@ -89,6 +89,11 @@ This project is intended to be used as a library. You can extend and customize d
     selector: 'job=~"%s.*"' % self.jobPrefix,
     title: '%(prefix)sSidecar' % $.dashboard.prefix,
   },
+  replicate+:: {
+    jobPrefix: 'thanos-replicate',
+    selector: 'job=~"%s.*"' % self.jobPrefix,
+    title: '%(prefix)sReplicate' % $.dashboard.prefix,
+  },
   overview+:: {
     title: '%(prefix)sOverview' % $.dashboard.prefix,
   },

diff --git a/mixin/thanos/alerts/alerts.libsonnet b/mixin/thanos/alerts/alerts.libsonnet
@@ -4,4 +4,5 @@
 (import 'sidecar.libsonnet') +
 (import 'store.libsonnet') +
 (import 'rule.libsonnet') +
-(import 'absent.libsonnet')
+(import 'absent.libsonnet') +
+(import 'replicate.libsonnet')
diff --git a/mixin/thanos/alerts/replicate.libsonnet b/mixin/thanos/alerts/replicate.libsonnet
@@ -1,6 +1,6 @@
 {
   local thanos = self,
-  replicator+:: {
+  replicate+:: {
     jobPrefix: error 'must provide job prefix for Thanos Replicate dashboard',
     selector: error 'must provide selector for Thanos Replicate dashboard',
   },
@@ -13,7 +13,7 @@
             alert: 'ThanosReplicateIsDown',
             expr: |||
               absent(up{%(selector)s})
-            ||| % thanos.replicator,
+            ||| % thanos.replicate,
             'for': '5m',
             labels: {
               severity: 'critical',
@@ -33,7 +33,7 @@
               / on (namespace) group_left
                 sum(rate(thanos_replicate_replication_runs_total{%(selector)s}[5m]))
               ) * 100 >= 10
-            ||| % thanos.replicator,
+            ||| % thanos.replicate,
             'for': '5m',
             labels: {
               severity: 'critical',
@@ -50,7 +50,7 @@
               and
                 sum by (job) (rate(thanos_replicate_replication_run_duration_seconds_bucket{%(selector)s}[5m])) > 0
               )
-            ||| % thanos.replicator,
+            ||| % thanos.replicate,
             'for': '5m',
             labels: {
               severity: 'critical',

diff --git a/mixin/thanos/dashboards/dashboards.libsonnet b/mixin/thanos/dashboards/dashboards.libsonnet
@@ -4,5 +4,6 @@
 (import 'receive.libsonnet') +
 (import 'rule.libsonnet') +
 (import 'compact.libsonnet') +
+(import 'replicate.libsonnet') +
 (import 'overview.libsonnet') +
 (import 'defaults.libsonnet')
diff --git a/mixin/thanos/dashboards/replicate.libsonnet b/mixin/thanos/dashboards/replicate.libsonnet
@@ -2,35 +2,35 @@ local g = import '../lib/thanos-grafana-builder/builder.libsonnet';
 
 {
   local thanos = self,
-  replicator+:: {
+  replicate+:: {
     jobPrefix: error 'must provide job prefix for Thanos Replicate dashboard',
     selector: error 'must provide selector for Thanos Replicate dashboard',
     title: error 'must provide title for Thanos Replicate dashboard',
   },
   grafanaDashboards+:: {
     'replicate.json':
-      g.dashboard(thanos.replicator.title)
+      g.dashboard(thanos.replicate.title)
       .addRow(
         g.row('Replicate Runs')
         .addPanel(
           g.panel('Rate') +
           g.qpsErrTotalPanel(
-            'thanos_replicate_replication_runs_total{result="error", namespace="$namespace",%(selector)s}' % thanos.replicator,
-            'thanos_replicate_replication_runs_total{namespace="$namespace",%(selector)s}' % thanos.replicator,
+            'thanos_replicate_replication_runs_total{result="error", namespace="$namespace",%(selector)s}' % thanos.replicate,
+            'thanos_replicate_replication_runs_total{namespace="$namespace",%(selector)s}' % thanos.replicate,
           )
         )
         .addPanel(
           g.panel('Errors', 'Shows rate of errors.') +
           g.queryPanel(
-            'sum(rate(thanos_replicate_replication_runs_total{result="error", namespace="$namespace",%(selector)s}[$interval])) by (result)' % thanos.replicator,
+            'sum(rate(thanos_replicate_replication_runs_total{result="error", namespace="$namespace",%(selector)s}[$interval])) by (result)' % thanos.replicate,
             '{{result}}'
           ) +
           { yaxes: g.yaxes('percentunit') } +
           g.stack
         )
         .addPanel(
           g.panel('Duration', 'Shows how long has it taken to run a replication cycle.') +
-          g.latencyPanel('thanos_replicate_replication_run_duration_seconds', 'result="success", namespace="$namespace",%(selector)s' % thanos.replicator)
+          g.latencyPanel('thanos_replicate_replication_run_duration_seconds', 'result="success", namespace="$namespace",%(selector)s' % thanos.replicate)
         )
       )
       .addRow(
@@ -39,20 +39,19 @@ local g = import '../lib/thanos-grafana-builder/builder.libsonnet';
           g.panel('Metrics') +
           g.queryPanel(
             [
-              'sum(rate(thanos_replicate_origin_iterations_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicator,
-              'sum(rate(thanos_replicate_origin_meta_loads_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicator,
-              'sum(rate(thanos_replicate_origin_partial_meta_reads_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicator,
-              'sum(rate(thanos_replicate_blocks_already_replicated_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicator,
-              'sum(rate(thanos_replicate_blocks_replicated_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicator,
-              'sum(rate(thanos_replicate_objects_replicated_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicator,
+              'sum(rate(thanos_replicate_origin_iterations_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicate,
+              'sum(rate(thanos_replicate_origin_meta_loads_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicate,
+              'sum(rate(thanos_replicate_origin_partial_meta_reads_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicate,
+              'sum(rate(thanos_replicate_blocks_already_replicated_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicate,
+              'sum(rate(thanos_replicate_blocks_replicated_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicate,
+              'sum(rate(thanos_replicate_objects_replicated_total{namespace="$namespace",%(selector)s}[$interval]))' % thanos.replicate,
             ],
             ['iterations', 'meta loads', 'partial meta reads', 'already replicated blocks', 'replicated blocks', 'replicated objects']
           )
         )
       )
       +
       g.template('namespace', 'kube_pod_info') +
-      g.template('job', 'up', 'namespace="$namespace",%(selector)s' % thanos.replicator, true, '%(jobPrefix)s.*' % thanos.replicator),
+      g.template('job', 'up', 'namespace="$namespace",%(selector)s' % thanos.replicate, true, '%(jobPrefix)s.*' % thanos.replicate),
   },
-} +
-(import 'defaults.libsonnet')
+}
diff --git a/mixin/thanos/defaults.libsonnet b/mixin/thanos/defaults.libsonnet
@@ -29,6 +29,11 @@
     selector: 'job=~"%s.*"' % self.jobPrefix,
     title: '%(prefix)sSidecar' % $.dashboard.prefix,
   },
+  replicate+:: {
+    jobPrefix: 'thanos-replicate',
+    selector: 'job=~"%s.*"' % self.jobPrefix,
+    title: '%(prefix)sReplicate' % $.dashboard.prefix,
+  },
   overview+:: {
     title: '%(prefix)sOverview' % $.dashboard.prefix,
   },

diff --git a/mixin/thanos/rules/replicate.libsonnet b/mixin/thanos/rules/replicate.libsonnet
@@ -1,6 +1,6 @@
 {
   local thanos = self,
-  replicator+:: {
+  replicate+:: {
     selector: error 'must provide selector for Thanos Replicate dashboard',
   },
   prometheusRules+:: {

diff --git a/mixin/thanos/rules/rules.libsonnet b/mixin/thanos/rules/rules.libsonnet
@@ -1,3 +1,4 @@
 (import 'query.libsonnet') +
 (import 'receive.libsonnet') +
-(import 'store.libsonnet')
+(import 'store.libsonnet') +
+(import 'replicate.libsonnet')