1979 lines
86 KiB
YAML
1979 lines
86 KiB
YAML
groups:
|
|
- name: Kafka-lag-Alerts
|
|
rules:
|
|
- alert: KafkaCatalogORViewConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"KafkaCatalogORViewConsumer"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: supplierTaxInvoiceConsumer01_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"supplierTaxInvoiceConsumer01"}) > 2000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
|
|
- alert: deactivationReasonTrashConsumer_greater_than_300
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~"taxonomy.deactivation.reason.trash", groupId=~"deactivationReasonTrashConsumer"}) > 300
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
bu: supply
|
|
service: taxonomy
|
|
team: cataloging
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
|
|
- alert: inStockListener_greater_than_40000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"inStockListener"}) > 40000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: resellerCollectionLastSeensHbaseConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"resellerCollectionLastSeensHbaseConsumer"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userSharedCatalogHbaseConsumer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userSharedCatalogHbaseConsumer"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userSubscriptionHbaseConsumer_greater_than_200
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userSubscriptionHbaseConsumer"}) > 200
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userWishListHbaseConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userWishListHbaseConsumer"}) > 1000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: collectionNotificationConsumer_greater_than_2000000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"collectionNotificationConsumer"}) > 2000000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: returnTrackingConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"returnTrackingConsumer"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: addSupplierSubOrderConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"addSupplierSubOrderConsumer"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
team: fulfilment
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkUpdateSupplierProductConsumer_greater_than_200000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkUpdateSupplierProductConsumer"}) > 200000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: exchangeTrackingConsumer_greater_than_4000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"exchangeTrackingConsumer"}) > 4000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: forwardTrackingConsumer_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"forwardTrackingConsumer"}) > 20000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: addSupplierProductConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"addSupplierProductConsumer"}) > 2000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkUpdateSupplierSubOrderConsumer_greater_than_90000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkUpdateSupplierSubOrderConsumer"}) > 90000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: vskuInventoryUpdateListener_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"vskuInventoryUpdateListener"}) > 5000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userSharedCatalogEtlConsumer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userSharedCatalogEtlConsumer"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userWishListEtlConsumer_greater_than_3500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userWishListEtlConsumer"}) > 3500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: inventory.order_state_change_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"inventory.order_state_change"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: KafkaUserHistoryConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"KafkaUserHistoryConsumer"}) > 1000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: addUserSubOrderConsumer_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"addUserSubOrderConsumer"}) > 20000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkUpdateUserSubOrderConsumer_greater_than_40000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkUpdateUserSubOrderConsumer"}) > 40000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: taxonomyVideoCreateConsumer_greater_than_20
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"taxonomyVideoCreateConsumer"}) > 20
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: taxonomyVideoDeleteConsumer_greater_than_20
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"taxonomyVideoDeleteConsumer"}) > 20
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: taxonomyVideoUpdateConsumer_greater_than_20
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"taxonomyVideoUpdateConsumer"}) > 20
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: videoSubOrderUpdateInternal_greater_than_20
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"videoSubOrderUpdateInternal"}) > 20
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: order_consumer_gamification_greater_than_3000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"order_consumer_gamification"}) > 3000
|
|
for: 2m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalog_consumer_gamification_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalog_consumer_gamification"}) > 1000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogCollageListener_greater_than_100000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogCollageListener"}) > 100000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: trackingVideoCreateListener_greater_than_20
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"trackingVideoCreateListener"}) > 20
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: trackingVideoDeleteListener_greater_than_20
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"trackingVideoDeleteListener"}) > 20
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkCatalogActivateConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkCatalogActivateConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkPostProcessingConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkPostProcessingConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkFirstViewConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkFirstViewConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkViewsDepleteConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkViewsDepleteConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkDeleteUserCatalogConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkDeleteUserCatalogConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkDecrementViewConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkDecrementViewConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulkCatalogValidityConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulkCatalogValidityConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: review_supplier_rating_scheduler_greater_than_80000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"review_supplier_rating_scheduler"}) > 80000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: wallet_schedule_processing_retry_g1_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"wallet_schedule_processing_retry_g1"}) > 2000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: wallet_retry_tx_processing_g1_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"wallet_retry_tx_processing_g1"}) > 2000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: wallet_schedule_processing_g1_greater_than_6000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"wallet_schedule_processing_g1"}) > 6000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: wallet_tx_processing_g1_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"wallet_tx_processing_g1"}) > 5000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: qwest.text_search_consumer_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"qwest.text_search_consumer"}) > 20000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: productSupplierMapCreateUpdateListener_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"productSupplierMapCreateUpdateListener"}) > 10000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_order_listener_greater_than_200
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_order_listener"}) > 200
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_catalog_shared_listener_greater_than_600
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_catalog_shared_listener"}) > 600
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_product_review_listener_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_product_review_listener"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_follow_user_listener_greater_than_50
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_follow_user_listener"}) > 50
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_profile_updates_listener_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_profile_updates_listener"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_referral_listener_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_referral_listener"}) > 5000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_level_transition_topic_listener_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_level_transition_topic_listener"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: aggregateFileProcessingRelayConsumer_greater_than_200
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"aggregateFileProcessingRelayConsumer"}) > 200
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: forwardInvoiceCreationConsumer_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"forwardInvoiceCreationConsumer"}) > 10000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: creditInvoiceCreationConsumer_greater_than_2500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"creditInvoiceCreationConsumer"}) > 2500
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: poInvoiceCreationConsumerId_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"poInvoiceCreationConsumerId"}) > 2000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: retryConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"retryConsumer"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: journey_point_credit_dead_letter_greater_than_50
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"journey_point_credit_dead_letter"}) > 50
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: supply-user-create_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"supply-user-create"}) > 500
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: supplyUserPaymentContainerId_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"supplyUserPaymentContainerId"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: add_referral_group_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"add_referral_group"}) > 500
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: referralBadgeUpgradeContainerId_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"referralBadgeUpgradeContainerId"}) > 1000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: referralPaymentAggregatorContainerId_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"referralPaymentAggregatorContainerId"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: supplyOrderEventConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"supplyOrderEventConsumer"}) > 5000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: paPaymentUpdateConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"paPaymentUpdateConsumer"}) > 1000
|
|
for: 20m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: test_meesho_greater_than_50
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"test_meesho"}) > 50
|
|
for: 2m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: test_meesho_1_greater_than_50
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"test_meesho_1"}) > 50
|
|
for: 2m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adServiceViewEventsProcessor_greater_than_150000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"adServiceViewEventsProcessor"}) > 150000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: manualRefundRecoveryContainer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"manualRefundRecoveryContainer"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogIngestionServiceListener2_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogIngestionServiceListener2"}) > 500
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: singleCatalogUploadServiceListener_greater_than_225
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"singleCatalogUploadServiceListener"}) > 225
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: taxonomyInventoryCudEventsListener_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"taxonomyInventoryCudEventsListener"}) > 10000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: fyNotificationConsumer_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"fyNotificationConsumer"}) > 20000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulletin.delayListener_greater_than_2500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulletin.delayListener"}) > 2500
|
|
for: 60m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulletin.productPriceChangeRecordSuccessListener_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulletin.productPriceChangeRecordSuccessListener"}) > 5000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulletin.productPriceChangeListener_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulletin.productPriceChangeListener"}) > 20000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulletin.wishlistEventListenerForPriceDrop_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulletin.wishlistEventListenerForPriceDrop"}) > 5000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulletin.shareEventListenerForPriceDrop_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulletin.shareEventListenerForPriceDrop"}) > 10000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: bulletin.priceDropScheduleListener_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"bulletin.priceDropScheduleListener"}) > 20000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: qwest_order_consumer_v3_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"qwest_order_consumer_v3"}) > 2000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_notification_listener_greater_than_120000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_notification_listener"}) > 120000
|
|
for: 60m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adServiceCatalogAttributeProcessor_greater_than_600
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"adServiceCatalogAttributeProcessor"}) > 600
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: campaignCatalogUpdateEventConsumer_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"campaignCatalogUpdateEventConsumer"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adServiceDelayedRetryConsumer_greater_than_200
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"adServiceDelayedRetryConsumer"}) > 200
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adServiceDelayedRetryConsumer_greater_than_150000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"adServiceDelayedRetryConsumer"}) > 150000
|
|
for: 10m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: gamification_journey_point_credit_listener_greater_than_50000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"gamification_journey_point_credit_listener"}) > 50000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: orderPaymentScheduledEventConsumerId_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"orderPaymentScheduledEventConsumerId"}) > 500
|
|
for: 30m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: orderPaymentCompletedEventConsumerId_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"orderPaymentCompletedEventConsumerId"}) > 500
|
|
for: 30m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userPaymentCompletedEventConsumerId_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userPaymentCompletedEventConsumerId"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: orderPaymentNoBankDetailEventConsumerId_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"orderPaymentNoBankDetailEventConsumerId"}) > 5000
|
|
for: 45m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userPaymentNoBankDetailEventConsumerId_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userPaymentNoBankDetailEventConsumerId"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userBankDetailStateUpdateConsumerId_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userBankDetailStateUpdateConsumerId"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: retryConsumerId_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"retryConsumerId"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: entityFilePayoutContainer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"entityFilePayoutContainer"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: fpPaymentsInfoUpdateContainer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"fpPaymentsInfoUpdateContainer"}) > 500
|
|
for: 30m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: apPaymentsInfoUpdateContainer_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"apPaymentsInfoUpdateContainer"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: epgInfoUpdateContainer_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"epgInfoUpdateContainer"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: instantPaymentsContainer_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"instantPaymentsContainer"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: nonInstantPaymentsContainer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"nonInstantPaymentsContainer"}) > 2000
|
|
for: 30m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: retryContainer_greater_than_25
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"retryContainer"}) > 25
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: payoutStateContainer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"payoutStateContainer"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: orderShippedContainer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"orderShippedContainer"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: anonymousUserSharedCatalogEtlConsumer_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"anonymousUserSharedCatalogEtlConsumer"}) > 500
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: userAppsConsumer_greater_than_100000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"userAppsConsumer"}) > 100000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: anonymousUserAppsEtlConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"anonymousUserAppsEtlConsumer"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: labelGenerationConsumer_greater_than_10
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"labelGenerationConsumer"}) > 10
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.supplyProductCreateUpdateGroup_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.supplyProductCreateUpdateGroup"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.supplyPSMCreateUpdateGroup_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.supplyPSMCreateUpdateGroup"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.inventoryChangeGroup_greater_than_2500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.inventoryChangeGroup"}) > 2500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.supplyCatalogCrudGroup_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.supplyCatalogCrudGroup"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.supplyCatalogCrudGroupV0_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.supplyCatalogCrudGroupV0"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.taxonomyCreateUpdateGroup_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.taxonomyCreateUpdateGroup"}) > 10000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.taxonomyProductSubSubCategoryGroup_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.taxonomyProductSubSubCategoryGroup"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.templateUpdateGroup_greater_than_10
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.templateUpdateGroup"}) > 10
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.catalogEventGroup_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.catalogEventGroup"}) > 10000
|
|
for: 3m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.productEventGroup_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.productEventGroup"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cms.sscatEventGroup_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cms.sscatEventGroup"}) > 500
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: clevertapNotificationsConsumer_greater_than_100000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"clevertapNotificationsConsumer"}) > 100000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: loginService_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"loginService"}) > 100
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adServiceInteractionEventsProcessor_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"adServiceInteractionEventsProcessor"}) > 10000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: holdCategoryUpdateConsumerOrderSearch_greater_than_5
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"holdCategoryUpdateConsumerOrderSearch"}) > 5
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: holdPinCodeUpdateConsumerOrderSearch_greater_than_5
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"holdPinCodeUpdateConsumerOrderSearch"}) > 5
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: ors_epg_refund_group_greater_than_50
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"ors_epg_refund_group"}) > 50
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: psmValidUpdateListener_greater_than_75000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"psmValidUpdateListener"}) > 75000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: communityPostInteractionNotificationConsumer_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"communityPostInteractionNotificationConsumer"}) > 100
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: productImageAddRetryListener_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"productImageAddRetryListener"}) > 1000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: taxonomyInventoryEventsListener_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"taxonomyInventoryEventsListener"}) > 5000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: taxonomySupplierUpsertListener_greater_than_300
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"taxonomySupplierUpsertListener"}) > 300
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogCollageInternalListener_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogCollageInternalListener"}) > 20000
|
|
for: 10m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogCollageInternalListener_greater_than_600000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogCollageInternalListener"}) > 600000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: startTransferConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"startTransferConsumer"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementUpsertConsumer_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementUpsertConsumer"}) > 10000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adminInventoryUpdateServiceListener_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"adminInventoryUpdateServiceListener"}) > 500
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogIngestionSkipQcServiceListener_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogIngestionSkipQcServiceListener"}) > 1000
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogLiveFlagServiceListener_greater_than_500
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogLiveFlagServiceListener"}) > 500
|
|
for: 15m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: fileIngestionServiceListener_greater_than_10
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"fileIngestionServiceListener"}) > 10
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: fileLevelIngestionServiceListener_greater_than_150
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"fileLevelIngestionServiceListener"}) > 150
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: psmPriceUpdateServiceListener_greater_than_20000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"psmPriceUpdateServiceListener"}) > 20000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: recordLevelIngestionServiceListener_greater_than_3000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"recordLevelIngestionServiceListener"}) > 3000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: supplierTemplateCreationListener_greater_than_100
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"supplierTemplateCreationListener"}) > 100
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: orderConfirmedConsumerOSM_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"orderConfirmedConsumerOSM"}) > 10000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: orderConfirmedRetryConsumerOSM_greater_than_20
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"orderConfirmedRetryConsumerOSM"}) > 20
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: user.profile.update.cue.group_greater_than_75
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"user.profile.update.cue.group"}) > 75
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementDuplicateProductConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementDuplicateProductConsumer"}) > 2000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementInventoryUpdateConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementInventoryUpdateConsumer"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementAddSupplierProductConsumer_greater_than_1200
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementAddSupplierProductConsumer"}) > 1200
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementProductUpdateConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementProductUpdateConsumer"}) > 1000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementCatalogCRUDEventConsumer_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementCatalogCRUDEventConsumer"}) > 2000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementPSMUpdateConsumer_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementPSMUpdateConsumer"}) > 5000
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogManagementPSVMMetadataUpdateConsumer_greater_than_3000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogManagementPSVMMetadataUpdateConsumer"}) > 3000
|
|
for: 60m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: catalogUpdateCumulativeFieldsListener_greater_than_5000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"catalogUpdateCumulativeFieldsListener"}) > 5000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: supplyOrderEventSPConsumer_greater_than_100000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"supplyOrderEventSPConsumer"}) > 100000
|
|
for: 30m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: penaltyContainerOrderCancelledNew_greater_than_10000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"penaltyContainerOrderCancelledNew"}) > 10000
|
|
for: 30m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: penaltyContainerOrderExchangePickedUpNew_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"penaltyContainerOrderExchangePickedUpNew"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: penaltyContainerOrderExchangeToReturnNew_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"penaltyContainerOrderExchangeToReturnNew"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: penaltyContainerOrderForwardRTONew_greater_than_2000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"penaltyContainerOrderForwardRTONew"}) > 2000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: penaltyContainerOrderExchangeRTONew_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"penaltyContainerOrderExchangeRTONew"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: penaltyContainerOrderReturnPickedUpNew_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"penaltyContainerOrderReturnPickedUpNew"}) > 1000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: paPaymentUpdateSPConsumer_greater_than_1000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"paPaymentUpdateSPConsumer"}) > 1000
|
|
for: 20m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cpsCampaignCatalogUpdateEventConsumer_greater_than_200
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cpsCampaignCatalogUpdateEventConsumer"}) > 200
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: cpsIngestionViewEventConsumer_greater_than_200
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"cpsIngestionViewEventConsumer"}) > 200
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: adsBudgetExhaustedEventConsumer_greater_than_5
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"adsBudgetExhaustedEventConsumer"}) > 5
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: update_xo_ums_greater_than_300000
|
|
expr: sum by (groupId,topic)(kafka_consumer_group_ConsumerLagMetrics_Value{instance=~".*bac-p-shared-.*", name="OffsetLag", topic=~".*", groupId=~"update_xo_ums"}) > 300000
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
env: prd
|
|
metric_type: consumer_lag_metric
|
|
annotations:
|
|
summary: "Consumer Lag on {{ $labels.groupId }}"
|
|
description: "Consumer Lag on {{ $labels.groupId }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
|