147 lines
5.9 KiB
YAML
147 lines
5.9 KiB
YAML
groups:
|
|
- name: Kafka-lag-Alerts
|
|
rules:
|
|
- alert: displayAdsCampaignReactivate_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"display_ads.campaign_reactivate", groupId=~"displayAdsCampaignReactivate"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu : supply
|
|
team : display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: displayAdsBudgetExhausted_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"display_ads.budget_exhausted", groupId=~"displayAdsBudgetExhausted"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu : supply
|
|
team : display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
|
|
- alert: displayAdsRedisRetry_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"display_ads.redis_retry", groupId=~"displayAdsRedisRetry"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
|
|
- alert: displayAdsWidgetView_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"test_topic", groupId=~"displayAdsWidgetView"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: displayAdsDSFeed_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"display_ads.ds_feed", groupId=~"displayAdsDSFeed"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: displayAdsBannerView_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"test_topic", groupId=~"displayAdsBannerView"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adWidgetViewEventConsumerGroup_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"rt_ad_widget_views", groupId=~"adWidgetViewEventConsumerGroup"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adBannerViewEventConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"rt_banner_viewed", groupId=~"adBannerViewEventConsumer"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adWidgetClickEventConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"rt_widget_clicked", groupId=~"adWidgetClickEventConsumer"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adBannerClickEventConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-display-ads-kafka.*", name="OffsetLag", topic=~"rt_banner_clicked", groupId=~"adBannerClickEventConsumer"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
team: display-ads
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
|