[#IOPID-1925] Alert Profile poison (#1134)

pagopa · Aug 30, 2024 · 17af2fd · 17af2fd
1 parent c8c901e
commit 17af2fd
Show file tree

Hide file tree

Showing 3 changed files with 38 additions and 3 deletions.
diff --git a/src/domains/elt/_modules/function_apps/function_app_elt.tf b/src/domains/elt/_modules/function_apps/function_app_elt.tf
@@ -95,7 +95,7 @@ locals {
       MESSAGE_STATUS_FAILURE_QUEUE_NAME      = "pdnd-io-cosmosdb-message-status-failure"
       SERVICES_FAILURE_QUEUE_NAME            = "pdnd-io-cosmosdb-services-failure"
       SERVICE_PREFERENCES_FAILURE_QUEUE_NAME = local.service_preferences_failure_queue_name
-      PROFILES_FAILURE_QUEUE_NAME            = "pdnd-io-cosmosdb-profiles-failure"
+      PROFILES_FAILURE_QUEUE_NAME            = local.profiles_failure_queue_name
 
       INTERNAL_TEST_FISCAL_CODES = module.tests.test_users.all
     }
@@ -168,8 +168,8 @@ module "function_elt" {
       "${local.function_elt.app_settings.SERVICES_FAILURE_QUEUE_NAME}-poison",
       local.service_preferences_failure_queue_name,
       "${local.service_preferences_failure_queue_name}-poison",
-      local.function_elt.app_settings.PROFILES_FAILURE_QUEUE_NAME,
-      "${local.function_elt.app_settings.PROFILES_FAILURE_QUEUE_NAME}-poison"
+      local.profiles_failure_queue_name,
+      "${local.profiles_failure_queue_name}-poison"
     ],
     "containers"           = [],
     "blobs_retention_days" = 1,

diff --git a/src/domains/elt/_modules/function_apps/locals.tf b/src/domains/elt/_modules/function_apps/locals.tf
@@ -17,4 +17,5 @@ locals {
   pn_service_id = "01G40DWQGKY5GRWSNM4303VNRP"
 
   service_preferences_failure_queue_name = "pdnd-io-cosmosdb-service-preferences-failure"
+  profiles_failure_queue_name            = "pdnd-io-cosmosdb-profiles-failure"
 }
diff --git a/src/domains/elt/_modules/function_apps/monitor.tf b/src/domains/elt/_modules/function_apps/monitor.tf
@@ -54,5 +54,39 @@ resource "azurerm_monitor_scheduled_query_rules_alert_v2" "service_preferences_f
     ]
   }
 
+  tags = var.tags
+}
+
+resource "azurerm_monitor_scheduled_query_rules_alert_v2" "profiles_failure_alert_rule" {
+  enabled             = true
+  name                = "[CITIZEN-AUTH | iopfneltsdt] Failures on pdnd-io-cosmosdb-profiles-failure-poison"
+  resource_group_name = var.resource_group_name
+  location            = var.location
+
+  scopes                  = [data.azurerm_storage_account.function_elt_internal_storage.id]
+  description             = "Permanent failures processing Profiles export to PDND. REQUIRED MANUAL ACTION"
+  severity                = 1
+  auto_mitigation_enabled = false
+
+  window_duration      = "PT15M" # Select the interval that's used to group the data points by using the aggregation type function. Choose an Aggregation granularity (period) that's greater than the Frequency of evaluation to reduce the likelihood of missing the first evaluation period of an added time series.
+  evaluation_frequency = "PT15M" # Select how often the alert rule is to be run. Select a frequency that's smaller than the aggregation granularity to generate a sliding window for the evaluation.
+
+  criteria {
+    query                   = <<-QUERY
+      StorageQueueLogs
+        | where OperationName contains "PutMessage"
+        | where Uri contains "${local.profiles_failure_queue_name}-poison"
+      QUERY
+    operator                = "GreaterThan"
+    threshold               = 0
+    time_aggregation_method = "Count"
+  }
+
+  action {
+    action_groups = [
+      data.azurerm_monitor_action_group.quarantine_error_action_group.id,
+    ]
+  }
+
   tags = var.tags
 }