feat: update hms metrics namespace

CHANGELOG.md

@@ -3,6 +3,16 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/) and this project adheres to [Semantic Versioning](http://semver.org/spec/v2.0.0.html).
 
+## [7.7.0] - 2024-11-19
+### Changed
+- Updated the hms namespaces for metrics for both readwrite and readonly.
+### Added
+- Added following variables:
+  - `hms_ecs_metrics_readwrite_namespace`
+  - `hms_ecs_metrics_readonly_namespace`
+  - `hms_k8s_metrics_readwrite_namespace`
+  - `hms_k8s_metrics_readonly_namespace`
+
 ## [7.6.1] - 2024-10-30
 ### Changed
 - Changed Openmetric to Prometheus in DockerLabels in ECS.

common.tf

@@ -9,8 +9,7 @@ locals {
   apiary_bucket_prefix             = "${local.instance_alias}-${data.aws_caller_identity.current.account_id}-${data.aws_region.current.name}"
   apiary_assume_role_bucket_prefix = [for assumerole in var.apiary_assume_roles : "${local.instance_alias}-${data.aws_caller_identity.current.account_id}-${lookup(assumerole, "allow_cross_region_access", false) ? "*" : data.aws_region.current.name}"]
   enable_route53_records           = var.apiary_domain_name == "" ? false : true
-
-  datadog_tags = join(" ", formatlist("%s:%s", keys(var.apiary_tags), values(var.apiary_tags)))
+  datadog_tags                     = join(" ", formatlist("%s:%s", keys(var.apiary_tags), values(var.apiary_tags)))
   #
   # Create a new list of maps with some extra attributes needed later
   #
@@ -63,6 +62,8 @@ locals {
 
   ro_ingress_cidr = var.ingress_cidr
   rw_ingress_cidr = length(var.rw_ingress_cidr) == 0 ? var.ingress_cidr : var.rw_ingress_cidr
+  hms_metrics                = join("\\\",\\\"", var.datadog_metrics_hms_readwrite_readonly)
+  hms_metrics_type_overrides = join("\\\": \\\"gauge\\\",\\\"", var.datadog_metrics_hms_readwrite_readonly)
 }
 
 data "aws_iam_account_alias" "current" {}

k8s-readonly.tf

@@ -38,7 +38,7 @@ resource "kubernetes_deployment_v1" "apiary_hms_readonly" {
         annotations = {
           "ad.datadoghq.com/${local.hms_alias}-readonly.check_names" = var.datadog_metrics_enabled ? "[\"prometheus\"]" : null
           "ad.datadoghq.com/${local.hms_alias}-readonly.init_configs" = var.datadog_metrics_enabled ? "[{}]" : null
-          "ad.datadoghq.com/${local.hms_alias}-readonly.instances" = var.datadog_metrics_enabled ? "[{ \"prometheus_url\": \"http://%%host%%:${var.datadog_metrics_port}/actuator/prometheus\", \"namespace\": \"hms_readonly\", \"metrics\": [ \"${join("\",\"", var.datadog_metrics_hms_readwrite_readonly)}\" ] , \"type_overrides\": { \"${join("\": \"gauge\",\"", var.datadog_metrics_hms_readwrite_readonly)}\": \"gauge\"} }]"  : null
+          "ad.datadoghq.com/${local.hms_alias}-readonly.instances" = var.datadog_metrics_enabled ? "[{ \"prometheus_url\": \"http://%%host%%:${var.datadog_metrics_port}/actuator/prometheus\", \"namespace\": \"${var.hms_k8s_metrics_readonly_namespace}\", \"metrics\": [ \"${join("\",\"", var.datadog_metrics_hms_readwrite_readonly)}\" ] , \"type_overrides\": { \"${join("\": \"gauge\",\"", var.datadog_metrics_hms_readwrite_readonly)}\": \"gauge\"} }]"  : null
           "iam.amazonaws.com/role" = var.oidc_provider == "" ? aws_iam_role.apiary_hms_readonly.name : null
           "prometheus.io/path"     = "/metrics"
           "prometheus.io/port"     = "8080"

k8s-readwrite.tf

@@ -38,7 +38,7 @@ resource "kubernetes_deployment_v1" "apiary_hms_readwrite" {
         annotations = {
           "ad.datadoghq.com/${local.hms_alias}-readwrite.check_names" = var.datadog_metrics_enabled ? "[\"prometheus\"]" : null
           "ad.datadoghq.com/${local.hms_alias}-readwrite.init_configs" = var.datadog_metrics_enabled ? "[{}]" : null
-          "ad.datadoghq.com/${local.hms_alias}-readwrite.instances" = var.datadog_metrics_enabled ? "[{ \"prometheus_url\": \"http://%%host%%:${var.datadog_metrics_port}/actuator/prometheus\", \"namespace\": \"hms_readwrite\", \"metrics\": [ \"${join("\",\"", var.datadog_metrics_hms_readwrite_readonly)}\" ] , \"type_overrides\": { \"${join("\": \"gauge\",\"", var.datadog_metrics_hms_readwrite_readonly)}\": \"gauge\"} }]" : null
+          "ad.datadoghq.com/${local.hms_alias}-readwrite.instances" = var.datadog_metrics_enabled ? "[{ \"prometheus_url\": \"http://%%host%%:${var.datadog_metrics_port}/actuator/prometheus\", \"namespace\": \"${var.hms_k8s_metrics_readwrite_namespace}\", \"metrics\": [ \"${join("\",\"", var.datadog_metrics_hms_readwrite_readonly)}\" ] , \"type_overrides\": { \"${join("\": \"gauge\",\"", var.datadog_metrics_hms_readwrite_readonly)}\": \"gauge\"} }]" : null
           "iam.amazonaws.com/role" = var.oidc_provider == "" ? aws_iam_role.apiary_hms_readwrite.name : null
           "prometheus.io/path"     = "/metrics"
           "prometheus.io/port"     = "8080"

templates.tf

@@ -27,7 +27,7 @@ locals{
     table_param_filter         = var.enable_metadata_events ? var.table_param_filter : ""
     enable_gluesync            = var.enable_gluesync ? "1" : ""
     gluedb_prefix              = "${local.gluedb_prefix}"
-
+    hms_metrics_namespace      = "${var.hms_ecs_metrics_readwrite_namespace}"
     ranger_service_name                    = "${local.instance_alias}-metastore"
     ranger_policy_manager_url              = "${var.ranger_policy_manager_url}"
     ranger_audit_solr_url                  = "${var.ranger_audit_solr_url}"
@@ -69,9 +69,11 @@ locals{
     datadog_agent_version = var.datadog_agent_version
     datadog_agent_enabled = var.datadog_agent_enabled
     datadog_tags          = local.datadog_tags
-    tcp_keepalive_time     = var.tcp_keepalive_time
-    tcp_keepalive_intvl    = var.tcp_keepalive_intvl
-    tcp_keepalive_probes   = var.tcp_keepalive_probes
+    tcp_keepalive_time    = var.tcp_keepalive_time
+    tcp_keepalive_intvl   = var.tcp_keepalive_intvl
+    tcp_keepalive_probes  = var.tcp_keepalive_probes
+    hms_metrics                = local.hms_metrics
+    hms_metrics_type_overrides = local.hms_metrics_type_overrides
   })
 
   hms_readonly_template = templatefile("${path.module}/templates/apiary-hms-readonly.json", {
@@ -117,14 +119,17 @@ locals{
     mysql_user_cred_arn    = data.aws_secretsmanager_secret.db_ro_user.arn
 
     # Datadog variables
-    datadog_agent_enabled = var.datadog_agent_enabled
-    datadog_secret_key    = length(var.datadog_key_secret_name) > 0 ? chomp(data.external.datadog_key[0].result["api_key"]) : ""
-    wd_instance_type      = var.hms_instance_type
-    metrics_port          = var.datadog_metrics_port
-    datadog_agent_version = var.datadog_agent_version
-    datadog_tags          = local.datadog_tags
-    tcp_keepalive_time     = var.tcp_keepalive_time
-    tcp_keepalive_intvl    = var.tcp_keepalive_intvl
-    tcp_keepalive_probes   = var.tcp_keepalive_probes
+    datadog_agent_enabled   = var.datadog_agent_enabled
+    datadog_secret_key      = length(var.datadog_key_secret_name) > 0 ? chomp(data.external.datadog_key[0].result["api_key"]) : ""
+    wd_instance_type        = var.hms_instance_type
+    metrics_port            = var.datadog_metrics_port
+    datadog_agent_version   = var.datadog_agent_version
+    datadog_tags            = local.datadog_tags
+    tcp_keepalive_time      = var.tcp_keepalive_time
+    tcp_keepalive_intvl     = var.tcp_keepalive_intvl
+    tcp_keepalive_probes    = var.tcp_keepalive_probes
+    hms_metrics_namespace   = "${var.hms_ecs_metrics_readonly_namespace}"
+    hms_metrics                = local.hms_metrics
+    hms_metrics_type_overrides = local.hms_metrics_type_overrides
   })
 }

templates/apiary-hms-readonly.json

@@ -75,7 +75,7 @@
         }
     },
     "dockerLabels": {
-        "com.datadoghq.ad.instances": "[{ \"prometheus_url\": \"http://%%host%%:8080/actuator/prometheus\", \"namespace\": \"hms_readonly_legacy\", \"metrics\": [\"metrics_classloading_loaded_value\", \"metrics_threads_count_value\", \"metrics_memory_heap_max_value\", \"metrics_init_total_count_tables_value\", \"metrics_init_total_count_dbs_value\", \"metrics_memory_heap_used_value\", \"metrics_init_total_count_partitions_value\", \"jvm_threads_current\", \"jvm_threads_started_total\", \"jvm_memory_bytes_used\", \"jvm_memory_bytes_init\", \"jvm_gc_collection_seconds_count\", \"jvm_gc_collection_seconds\", \"process_cpu_seconds_total\", \"java_lang_operatingsystem_processcpuload\", \"java_lang_operatingsystem_processcputime\", \"metrics_threads_runnable_count_value\", \"metrics_threads_waiting_count_value\", \"java_lang_memory_heapmemoryusage_used\", \"metrics_memory_heap_init_value\", \"metrics_api_get_partition_by_name_count\", \"metrics_api_get_partitions_by_names_count\", \"metrics_api_get_partition_names_count\", \"metrics_api_get_partitions_by_expr_count\", \"metrics_api_get_partitions_count\", \"metrics_api_get_partition_count\", \"metrics_api_get_partitions_by_filter_count\", \"metrics_api_add_partitions_count\", \"metrics_api_add_partitions_req_count\", \"metrics_api_drop_partition_by_name_count\", \"metrics_api_add_partition_count\", \"metrics_api_alter_partitions_count\", \"metrics_api_create_table_count\", \"metrics_api_alter_table_with_cascade_count\", \"metrics_api_get_table_meta_count\", \"metrics_api_get_table_metas_count\", \"metrics_api_get_table_count\", \"metrics_api_alter_table_count\", \"metrics_api_get_tables_count\", \"metrics_api_get_all_tables_count\", \"metrics_api_drop_table_count\", \"metrics_api_get_multi_table_count\", \"metrics_api_get_database_count\", \"metrics_api_get_all_databases_count\", \"metrics_api_get_databases_count\", \"metrics_api_create_function_count\", \"metrics_api_getmetaconf_count\", \"metrics_api_alter_table_with_environment_context_count\", \"metrics_api_delete_column_statistics_by_table_count\", \"metrics_api_get_functions_count\", \"metrics_api_get_function_count\", \"metrics_api_shutdown_count\", \"metrics_api_flushcache_count\", \"metrics_api_get_indexes_count\", \"metrics_api_get_config_value_count\", \"metrics_api_set_ugi_count\", \"metrics_api_get_all_functions_count\", \"metrics_api_get_table_req_95thpercentile\", \"metrics_api_get_table_req_50thpercentile\", \"metrics_api_get_table_req_count\", \"metrics_api_get_table_req_max\", \"metrics_api_get_databases_count\", \"metrics_api_get_databases_95thpercentile\", \"metrics_api_get_databases_50thpercentile\", \"metrics_api_get_databases_max\", \"metrics_api_get_partitions_95thpercentile\", \"metrics_api_get_partitions_50thpercentile\", \"metrics_api_get_partitions_count\", \"metrics_api_get_partitions_max\", \"metrics_api_get_partitions_50thpercentile\", \"metrics_api_get_table_req_50thpercentile\", \"metrics_api_get_database_95thpercentile\", \"metrics_api_get_database_50thpercentile\", \"metrics_kafka_listener_failures_count\", \"metrics_kafka_listener_successes_count\", \"metrics_api_get_table_objects_by_name_req_max\" ], \"type_overrides\": { \"metrics_classloading_loaded_value\": \"gauge\", \"metrics_threads_count_value\": \"gauge\", \"metrics_memory_heap_max_value\": \"gauge\", \"metrics_init_total_count_tables_value\": \"gauge\", \"metrics_init_total_count_dbs_value\": \"gauge\", \"metrics_memory_heap_used_value\": \"gauge\", \"metrics_init_total_count_partitions_value\": \"gauge\", \"jvm_threads_current\": \"gauge\", \"jvm_threads_started_total\": \"gauge\", \"jvm_memory_bytes_used\": \"gauge\", \"jvm_memory_bytes_init\": \"gauge\", \"jvm_gc_collection_seconds_count\": \"gauge\", \"jvm_gc_collection_seconds\": \"gauge\", \"process_cpu_seconds_total\": \"gauge\", \"java_lang_operatingsystem_processcpuload\": \"gauge\", \"java_lang_operatingsystem_processcputime\": \"gauge\", \"metrics_threads_runnable_count_value\": \"gauge\", \"metrics_threads_waiting_count_value\": \"gauge\", \"java_lang_memory_heapmemoryusage_used\": \"gauge\", \"metrics_memory_heap_init_value\": \"gauge\", \"metrics_api_get_partition_by_name_count\": \"gauge\", \"metrics_api_get_partitions_by_names_count\": \"gauge\", \"metrics_api_get_partition_names_count\": \"gauge\", \"metrics_api_get_partitions_by_expr_count\": \"gauge\", \"metrics_api_get_partitions_count\": \"gauge\", \"metrics_api_get_partition_count\": \"gauge\", \"metrics_api_get_partitions_by_filter_count\": \"gauge\", \"metrics_api_add_partitions_count\": \"gauge\", \"metrics_api_add_partitions_req_count\": \"gauge\", \"metrics_api_drop_partition_by_name_count\": \"gauge\", \"metrics_api_add_partition_count\": \"gauge\", \"metrics_api_alter_partitions_count\": \"gauge\", \"metrics_api_create_table_count\": \"gauge\", \"metrics_api_alter_table_with_cascade_count\": \"gauge\", \"metrics_api_get_table_meta_count\": \"gauge\", \"metrics_api_get_table_metas_count\": \"gauge\", \"metrics_api_get_table_count\": \"gauge\", \"metrics_api_alter_table_count\": \"gauge\", \"metrics_api_get_tables_count\": \"gauge\", \"metrics_api_get_all_tables_count\": \"gauge\", \"metrics_api_drop_table_count\": \"gauge\", \"metrics_api_get_multi_table_count\": \"gauge\", \"metrics_api_get_database_count\": \"gauge\", \"metrics_api_get_all_databases_count\": \"gauge\", \"metrics_api_get_databases_count\": \"gauge\", \"metrics_api_create_function_count\": \"gauge\", \"metrics_api_getmetaconf_count\": \"gauge\", \"metrics_api_alter_table_with_environment_context_count\": \"gauge\", \"metrics_api_delete_column_statistics_by_table_count\": \"gauge\", \"metrics_api_get_functions_count\": \"gauge\", \"metrics_api_get_function_count\": \"gauge\", \"metrics_api_shutdown_count\": \"gauge\", \"metrics_api_flushcache_count\": \"gauge\", \"metrics_api_get_indexes_count\": \"gauge\", \"metrics_api_get_config_value_count\": \"gauge\", \"metrics_api_set_ugi_count\": \"gauge\", \"metrics_api_get_all_functions_count\": \"gauge\", \"metrics_api_get_table_req_95thpercentile\": \"gauge\", \"metrics_api_get_table_req_50thpercentile\": \"gauge\", \"metrics_api_get_table_req_count\": \"gauge\", \"metrics_api_get_table_req_max\": \"gauge\", \"metrics_api_get_databases_count\": \"gauge\", \"metrics_api_get_databases_95thpercentile\": \"gauge\", \"metrics_api_get_databases_50thpercentile\": \"gauge\", \"metrics_api_get_databases_max\": \"gauge\", \"metrics_api_get_partitions_95thpercentile\": \"gauge\", \"metrics_api_get_partitions_50thpercentile\": \"gauge\", \"metrics_api_get_partitions_count\": \"gauge\", \"metrics_api_get_partitions_max\": \"gauge\", \"metrics_api_get_partitions_50thpercentile\": \"gauge\", \"metrics_api_get_table_req_50thpercentile\": \"gauge\",\"metrics_api_get_database_95thpercentile\": \"gauge\",\"metrics_api_get_database_50thpercentile\": \"gauge\", \"metrics_kafka_listener_failures_count\": \"gauge\", \"metrics_kafka_listener_successes_count\": \"gauge\", \"metrics_api_get_table_objects_by_name_req_max\": \"gauge\"} }]",
+        "com.datadoghq.ad.instances": "[{ \"prometheus_url\": \"http://%%host%%:8080/actuator/prometheus\", \"namespace\": \"${hms_metrics_namespace}\", \"metrics\": [\"${hms_metrics}\"],\"type_overrides\":{\"${hms_metrics_type_overrides}\": \"gauge\"} }]",
         "com.datadoghq.ad.check_names": "[\"prometheus\"]",
         "com.datadoghq.ad.init_configs": "[{}]"
     },