diff --git a/examples/weave-net-example.jsonnet b/examples/weave-net-example.jsonnet
index 9d708761bc7008134abd7d9f28b7cd0a001462ab..a7eca242c2402e53190d6d8906ed87785c519480 100644
--- a/examples/weave-net-example.jsonnet
+++ b/examples/weave-net-example.jsonnet
@@ -1,12 +1,12 @@
local kp = (import 'kube-prometheus/main.libsonnet') +
- (import 'kube-prometheus/addons/weave-net.libsonnet') + {
+ (import 'kube-prometheus/addons/weave-net/weave-net.libsonnet') + {
values+:: {
common+: {
namespace: 'monitoring',
},
},
prometheus+: {
- prometheusRule+: {
+ prometheusRuleWeaveNet+: {
spec+: {
groups: std.map(
function(group)
diff --git a/jsonnet/kube-prometheus/addons/weave-net.libsonnet b/jsonnet/kube-prometheus/addons/weave-net.libsonnet
deleted file mode 100644
index 0755e60c83143931fb6861b34db7ca87e25ea68f..0000000000000000000000000000000000000000
--- a/jsonnet/kube-prometheus/addons/weave-net.libsonnet
+++ /dev/null
@@ -1,196 +0,0 @@
-{
- prometheus+: {
- serviceWeaveNet: {
- apiVersion: 'v1',
- kind: 'Service',
- metadata: {
- name: 'weave-net',
- namespace: 'kube-system',
- labels: { 'app.kubernetes.io/name': 'weave-net' },
- },
- spec: {
- ports: [
- { name: 'weave-net-metrics', targetPort: 6782, port: 6782 },
- ],
- selector: { name: 'weave-net' },
- clusterIP: 'None',
- },
- },
- serviceMonitorWeaveNet: {
- apiVersion: 'monitoring.coreos.com/v1',
- kind: 'ServiceMonitor',
- metadata: {
- name: 'weave-net',
- labels: {
- 'app.kubernetes.io/name': 'weave-net',
- },
- namespace: 'monitoring',
- },
- spec: {
- jobLabel: 'app.kubernetes.io/name',
- endpoints: [
- {
- port: 'weave-net-metrics',
- path: '/metrics',
- interval: '15s',
- },
- ],
- namespaceSelector: {
- matchNames: [
- 'kube-system',
- ],
- },
- selector: {
- matchLabels: {
- 'app.kubernetes.io/name': 'weave-net',
- },
- },
- },
- },
- },
- prometheusRules+: {
- groups+: [
- {
- name: 'weave-net',
- rules: [
- {
- alert: 'WeaveNetIPAMSplitBrain',
- expr: 'max(weave_ipam_unreachable_percentage) - min(weave_ipam_unreachable_percentage) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'Percentage of all IP addresses owned by unreachable peers is not same for every node.',
- description: 'actionable: Weave Net network has a split brain problem. Please find the problem and fix it.',
- },
- },
- {
- alert: 'WeaveNetIPAMUnreachable',
- expr: 'weave_ipam_unreachable_percentage > 25',
- 'for': '10m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'Percentage of all IP addresses owned by unreachable peers is above threshold.',
- description: 'actionable: Please find the problem and fix it.',
- },
- },
- {
- alert: 'WeaveNetIPAMPendingAllocates',
- expr: 'sum(weave_ipam_pending_allocates) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'Number of pending allocates is above the threshold.',
- description: 'actionable: Please find the problem and fix it.',
- },
- },
- {
- alert: 'WeaveNetIPAMPendingClaims',
- expr: 'sum(weave_ipam_pending_claims) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'Number of pending claims is above the threshold.',
- description: 'actionable: Please find the problem and fix it.',
- },
- },
- {
- alert: 'WeaveNetFastDPFlowsLow',
- expr: 'sum(weave_flows) < 15000',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'Number of FastDP flows is below the threshold.',
- description: 'actionable: Please find the reason for FastDP flows to go below the threshold and fix it.',
- },
- },
- {
- alert: 'WeaveNetFastDPFlowsOff',
- expr: 'sum(weave_flows == bool 0) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'FastDP flows is zero.',
- description: 'actionable: Please find the reason for FastDP flows to be off and fix it.',
- },
- },
- {
- alert: 'WeaveNetHighConnectionTerminationRate',
- expr: 'rate(weave_connection_terminations_total[5m]) > 0.1',
- 'for': '5m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'A lot of connections are getting terminated.',
- description: 'actionable: Please find the reason for the high connection termination rate and fix it.',
- },
- },
- {
- alert: 'WeaveNetConnectionsConnecting',
- expr: 'sum(weave_connections{state="connecting"}) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'A lot of connections are in connecting state.',
- description: 'actionable: Please find the reason for this and fix it.',
- },
- },
- {
- alert: 'WeaveNetConnectionsRetying',
- expr: 'sum(weave_connections{state="retrying"}) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'A lot of connections are in retrying state.',
- description: 'actionable: Please find the reason for this and fix it.',
- },
- },
- {
- alert: 'WeaveNetConnectionsPending',
- expr: 'sum(weave_connections{state="pending"}) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'A lot of connections are in pending state.',
- description: 'actionable: Please find the reason for this and fix it.',
- },
- },
- {
- alert: 'WeaveNetConnectionsFailed',
- expr: 'sum(weave_connections{state="failed"}) > 0',
- 'for': '3m',
- labels: {
- severity: 'critical',
- },
- annotations: {
- summary: 'A lot of connections are in failed state.',
- description: 'actionable: Please find the reason and fix it.',
- },
- },
- ],
- },
- ],
- },
- grafanaDashboards+:: {
- 'weave-net.json': (import './grafana-weave-net.json'),
- 'weave-net-cluster.json': (import './grafana-weave-net-cluster.json'),
- },
-}
diff --git a/jsonnet/kube-prometheus/addons/weave-net/alerts.libsonnet b/jsonnet/kube-prometheus/addons/weave-net/alerts.libsonnet
new file mode 100644
index 0000000000000000000000000000000000000000..c0ca940a9ca7260f99c0e2cacbbdec3ce1d5b71a
--- /dev/null
+++ b/jsonnet/kube-prometheus/addons/weave-net/alerts.libsonnet
@@ -0,0 +1,134 @@
+[
+ {
+ alert: 'WeaveNetIPAMSplitBrain',
+ expr: 'max(weave_ipam_unreachable_percentage) - min(weave_ipam_unreachable_percentage) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'Percentage of all IP addresses owned by unreachable peers is not same for every node.',
+ description: 'actionable: Weave Net network has a split brain problem. Please find the problem and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetIPAMUnreachable',
+ expr: 'weave_ipam_unreachable_percentage > 25',
+ 'for': '10m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'Percentage of all IP addresses owned by unreachable peers is above threshold.',
+ description: 'actionable: Please find the problem and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetIPAMPendingAllocates',
+ expr: 'sum(weave_ipam_pending_allocates) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'Number of pending allocates is above the threshold.',
+ description: 'actionable: Please find the problem and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetIPAMPendingClaims',
+ expr: 'sum(weave_ipam_pending_claims) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'Number of pending claims is above the threshold.',
+ description: 'actionable: Please find the problem and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetFastDPFlowsLow',
+ expr: 'sum(weave_flows) < 15000',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'Number of FastDP flows is below the threshold.',
+ description: 'actionable: Please find the reason for FastDP flows to go below the threshold and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetFastDPFlowsOff',
+ expr: 'sum(weave_flows == bool 0) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'FastDP flows is zero.',
+ description: 'actionable: Please find the reason for FastDP flows to be off and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetHighConnectionTerminationRate',
+ expr: 'rate(weave_connection_terminations_total[5m]) > 0.1',
+ 'for': '5m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'A lot of connections are getting terminated.',
+ description: 'actionable: Please find the reason for the high connection termination rate and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetConnectionsConnecting',
+ expr: 'sum(weave_connections{state="connecting"}) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'A lot of connections are in connecting state.',
+ description: 'actionable: Please find the reason for this and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetConnectionsRetying',
+ expr: 'sum(weave_connections{state="retrying"}) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'A lot of connections are in retrying state.',
+ description: 'actionable: Please find the reason for this and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetConnectionsPending',
+ expr: 'sum(weave_connections{state="pending"}) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'A lot of connections are in pending state.',
+ description: 'actionable: Please find the reason for this and fix it.',
+ },
+ },
+ {
+ alert: 'WeaveNetConnectionsFailed',
+ expr: 'sum(weave_connections{state="failed"}) > 0',
+ 'for': '3m',
+ labels: {
+ severity: 'critical',
+ },
+ annotations: {
+ summary: 'A lot of connections are in failed state.',
+ description: 'actionable: Please find the reason and fix it.',
+ },
+ },
+]
diff --git a/jsonnet/kube-prometheus/addons/grafana-weave-net-cluster.json b/jsonnet/kube-prometheus/addons/weave-net/grafana-weave-net-cluster.json
similarity index 100%
rename from jsonnet/kube-prometheus/addons/grafana-weave-net-cluster.json
rename to jsonnet/kube-prometheus/addons/weave-net/grafana-weave-net-cluster.json
diff --git a/jsonnet/kube-prometheus/addons/grafana-weave-net.json b/jsonnet/kube-prometheus/addons/weave-net/grafana-weave-net.json
similarity index 100%
rename from jsonnet/kube-prometheus/addons/grafana-weave-net.json
rename to jsonnet/kube-prometheus/addons/weave-net/grafana-weave-net.json
diff --git a/jsonnet/kube-prometheus/addons/weave-net/weave-net.libsonnet b/jsonnet/kube-prometheus/addons/weave-net/weave-net.libsonnet
new file mode 100644
index 0000000000000000000000000000000000000000..71596f4568aac7fcd46869dd3f84c8406ef9fd2d
--- /dev/null
+++ b/jsonnet/kube-prometheus/addons/weave-net/weave-net.libsonnet
@@ -0,0 +1,73 @@
+{
+ prometheus+: {
+ local p = self,
+ serviceWeaveNet: {
+ apiVersion: 'v1',
+ kind: 'Service',
+ metadata: {
+ name: 'weave-net',
+ namespace: 'kube-system',
+ labels: { 'app.kubernetes.io/name': 'weave-net' },
+ },
+ spec: {
+ ports: [
+ { name: 'weave-net-metrics', targetPort: 6782, port: 6782 },
+ ],
+ selector: { name: 'weave-net' },
+ clusterIP: 'None',
+ },
+ },
+ serviceMonitorWeaveNet: {
+ apiVersion: 'monitoring.coreos.com/v1',
+ kind: 'ServiceMonitor',
+ metadata: {
+ name: 'weave-net',
+ labels: {
+ 'app.kubernetes.io/name': 'weave-net',
+ },
+ namespace: 'monitoring',
+ },
+ spec: {
+ jobLabel: 'app.kubernetes.io/name',
+ endpoints: [
+ {
+ port: 'weave-net-metrics',
+ path: '/metrics',
+ interval: '15s',
+ },
+ ],
+ namespaceSelector: {
+ matchNames: [
+ 'kube-system',
+ ],
+ },
+ selector: {
+ matchLabels: {
+ 'app.kubernetes.io/name': 'weave-net',
+ },
+ },
+ },
+ },
+ prometheusRuleWeaveNet: {
+ apiVersion: 'monitoring.coreos.com/v1',
+ kind: 'PrometheusRule',
+ metadata: {
+ labels: p.config.mixin.ruleLabels,
+ name: 'weave-net-rules',
+ namespace: p.config.namespace,
+ },
+ spec: {
+ groups: [{
+ name: 'weave-net',
+ rules: (import './alerts.libsonnet'),
+ }],
+ },
+ },
+ mixin+:: {
+ grafanaDashboards+:: {
+ 'weave-net.json': (import './grafana-weave-net.json'),
+ 'weave-net-cluster.json': (import './grafana-weave-net-cluster.json'),
+ },
+ },
+ },
+}