diff --git a/src/content/docs/infrastructure/host-integrations/host-integrations-list/temporal-cloud-integration.mdx b/src/content/docs/infrastructure/host-integrations/host-integrations-list/temporal-cloud-integration.mdx index 293334dac63..79a3879d80b 100644 --- a/src/content/docs/infrastructure/host-integrations/host-integrations-list/temporal-cloud-integration.mdx +++ b/src/content/docs/infrastructure/host-integrations/host-integrations-list/temporal-cloud-integration.mdx @@ -55,241 +55,120 @@ Our Temporal Cloud integration monitors the performance of your Temporal Cloud d ```yml integrations: - - name: nri-flex - config: - name: temporalCloudIntegration - apis: - - event_type: temporalCloudFrontendServiceError - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_frontend_service_error_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudFrontendServiceRequest - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_frontend_service_request_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudPollSuccess - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_poll_success_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudPollSuccessSync - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_poll_success_sync_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudPollTimeout - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_poll_timeout_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudResourceExhaustedErrors - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_resource_exhausted_error_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudScheduleActionSuccess - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_schedule_action_success_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudScheduleBufferOverruns - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_schedule_buffer_overruns_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudScheduleMissedCatchupWindow - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_schedule_missed_catchup_window_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudScheduleRateLimited - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_schedule_rate_limited_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudServiceLatencyBucket - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_service_latency_bucket - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudServiceLatencyCount - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_service_latency_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudServiceLatencySum - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_service_latency_sum - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudStateTransition - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_state_transition_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudTotalAction - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_total_action_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudWorkflowCancel - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_workflow_cancel_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudWorkflowContinued - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_workflow_continued_as_new_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudWorkflowFailed - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_workflow_failed_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudWorkflowSuccess - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_workflow_success_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudWorkflowTerminate - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_workflow_terminate_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem - - event_type: temporalCloudWorkflowTimeout - url: https://.tmprl.cloud/prometheus/api/v1/query?query=temporal_cloud_v0_workflow_timeout_count - headers: - accept: application/json - split_array: true - leaf_array: true - tls_config: - enable: true - ca: /ca-certificates.crt - key: /ca.key - cert: /ca.pem + - name: nri-flex + config: + name: temporalCloudIntegration + global: + base_url: https://.tmprl.cloud/prometheus/api/v1/ + headers: + accept: application/json + tls_config: + enable: true + key: /certs/client.key + cert: /certs/client.pem + interval: 60s + timeout: 55s + apis: + - event_type: temporalCloudFrontendServiceError + # sum(increase(temporal_cloud_v0_frontend_service_error_count[1m])) by (temporal_namespace,operation) + url: query?query=sum%28increase%28temporal_cloud_v0_frontend_service_error_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2Coperation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudFrontendServiceRequest + # sum(increase(temporal_cloud_v0_frontend_service_request_count[1m])) by (temporal_namespace,operation) + url: query?query=sum%28increase%28temporal_cloud_v0_frontend_service_request_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2Coperation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudPollSuccess + # sum(increase(temporal_cloud_v0_poll_success_count[1m])) by (temporal_namespace,operation,task_type) + url: query?query=sum%28increase%28temporal_cloud_v0_poll_success_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2Coperation%2Ctask_type%29 + split_array: true + leaf_array: true + - event_type: temporalCloudPollSuccessSync + # sum(increase(temporal_cloud_v0_poll_success_sync_count[1m])) by (temporal_namespace,operation,task_type) + url: query?query=sum%28increase%28temporal_cloud_v0_poll_success_sync_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2Coperation%2Ctask_type%29 + split_array: true + leaf_array: true + - event_type: temporalCloudPollTimeout + # sum(increase(temporal_cloud_v0_poll_timeout_count[1m])) by (temporal_namespace,operation,task_type) + url: query?query=sum%28increase%28temporal_cloud_v0_poll_timeout_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2Coperation%2Ctask_type%29 + split_array: true + leaf_array: true + - event_type: temporalCloudResourceExhaustedErrors + # sum(increase(temporal_cloud_v0_resource_exhausted_error_count[1m])) by (temporal_namespace,resource_exhausted_cause) + url: query?query=sum%28increase%28temporal_cloud_v0_resource_exhausted_error_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2Cresource_exhausted_cause%29 + split_array: true + leaf_array: true + - event_type: temporalCloudScheduleActionSuccess + # sum(increase(temporal_cloud_v0_schedule_action_success_count[1m])) by (temporal_namespace) + url: query?query=sum%28increase%28temporal_cloud_v0_schedule_action_success_count%5B1m%5D%29%29%20by%20%28temporal_namespace%29 + split_array: true + leaf_array: true + - event_type: temporalCloudScheduleBufferOverruns + # sum(increase(temporal_cloud_v0_schedule_buffer_overruns_count[1m])) by (temporal_namespace) + url: query?query=sum%28increase%28temporal_cloud_v0_schedule_buffer_overruns_count%5B1m%5D%29%29%20by%20%28temporal_namespace%29 + split_array: true + leaf_array: true + - event_type: temporalCloudScheduleMissedCatchupWindow + # sum(increase(temporal_cloud_v0_schedule_missed_catchup_window_count[1m])) by (temporal_namespace) + url: query?query=sum%28increase%28temporal_cloud_v0_schedule_missed_catchup_window_count%5B1m%5D%29%29%20by%20%28temporal_namespace%29 + split_array: true + leaf_array: true + - event_type: temporalCloudScheduleRateLimited + # sum(increase(temporal_cloud_v0_schedule_rate_limited_count[1m])) by (temporal_namespace) + url: query?query=sum%28increase%28temporal_cloud_v0_schedule_rate_limited_count%5B1m%5D%29%29%20by%20%28temporal_namespace%29 + split_array: true + leaf_array: true + - event_type: temporalCloudStateTransition + # sum(increase(temporal_cloud_v0_state_transition_count[1m])) by (temporal_namespace) + url: query?query=sum%28increase%28temporal_cloud_v0_state_transition_count%5B1m%5D%29%29%20by%20%28temporal_namespace%29 + split_array: true + leaf_array: true + - event_type: temporalCloudTotalAction + # sum(increase(temporal_cloud_v0_total_action_count[1m])) by (temporal_namespace, namespace_mode) + url: query?query=sum%28increase%28temporal_cloud_v0_total_action_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2C%20namespace_mode%29 + split_array: true + leaf_array: true + - event_type: temporalCloudWorkflowCancel + # sum(increase(temporal_cloud_v0_workflow_cancel_count[1m])) by (temporal_namespace, operation) + url: query?query=sum%28increase%28temporal_cloud_v0_workflow_cancel_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2C%20operation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudWorkflowContinued + # sum(increase(temporal_cloud_v0_workflow_continued_as_new_count[1m])) by (temporal_namespace, operation) + url: query?query=sum%28increase%28temporal_cloud_v0_workflow_continued_as_new_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2C%20operation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudWorkflowFailed + # sum(increase(temporal_cloud_v0_workflow_failed_count[1m])) by (temporal_namespace, operation) + url: query?query=sum%28increase%28temporal_cloud_v0_workflow_failed_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2C%20operation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudWorkflowSuccess + # sum(increase(temporal_cloud_v0_workflow_success_count[1m])) by (temporal_namespace, operation) + url: query?query=sum%28increase%28temporal_cloud_v0_workflow_success_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2C%20operation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudWorkflowTerminate + # sum(increase(temporal_cloud_v0_workflow_terminate_count[1m])) by (temporal_namespace, operation) + url: query?query=sum%28increase%28temporal_cloud_v0_workflow_terminate_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2C%20operation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudWorkflowTimeout + # sum(increase(temporal_cloud_v0_workflow_timeout_count[1m])) by (temporal_namespace, operation) + url: query?query=sum%28increase%28temporal_cloud_v0_workflow_timeout_count%5B1m%5D%29%29%20by%20%28temporal_namespace%2C%20operation%29 + split_array: true + leaf_array: true + - event_type: temporalCloudWorkflowExecutionLatencySecP95 + # histogram_quantile(0.95, sum(rate(temporal_cloud_v0_service_latency_bucket[1m] offset 1m)) by (temporal_namespace, operation, le)) + url: query?query=histogram_quantile%280.95%2C%20sum%28rate%28temporal_cloud_v0_service_latency_bucket%5B1m%5D%20offset%201m%29%29%20by%20%28temporal_namespace%2C%20operation%2C%20le%29%29 + split_array: true + leaf_array: true + - event_type: temporalCloudReplicationLagSecP95 + # histogram_quantile(0.95, sum(rate(temporal_cloud_v0_replication_lag_bucket[1m] offset 1m)) by (temporal_namespace, le)) + url: query?query=histogram_quantile%280.95%2C%20sum%28rate%28temporal_cloud_v0_replication_lag_bucket%5B1m%5D%20offset%201m%29%29%20by%20%28temporal_namespace%2C%20le%29%29 + split_array: true + leaf_array: true ``` 3. Use our instructions [to restart your infrastructure agent](/docs/infrastructure/install-infrastructure-agent/manage-your-agent/start-stop-restart-infrastructure-agent/):