redhat-performance · openshift-merge-bot · Sep 27, 2024 · Aug 20, 2024
diff --git a/ci-scripts/rhdh-setup/deploy.sh b/ci-scripts/rhdh-setup/deploy.sh
@@ -57,6 +57,7 @@ export ENABLE_PROFILING="${ENABLE_PROFILING:-false}"
 
 export PSQL_LOG="${PSQL_LOG:-true}"
 export RHDH_METRIC="${RHDH_METRIC:-true}"
+export PSQL_EXPORT="${PSQL_EXPORT:-false}"
 export LOG_MIN_DURATION_STATEMENT="${LOG_MIN_DURATION_STATEMENT:-65}"
 export LOG_MIN_DURATION_SAMPLE="${LOG_MIN_DURATION_SAMPLE:-50}"
 export LOG_STATEMENT_SAMPLE_RATE="${LOG_STATEMENT_SAMPLE_RATE:-0.7}"
@@ -346,13 +347,49 @@ psql_debug() {
         $clin exec "${psql_db}" -- sh -c "sed -i "s/^\s*#log_min_duration_sample.*/log_min_duration_sample=${LOG_MIN_DURATION_SAMPLE}/" /var/lib/pgsql/data/userdata/postgresql.conf "
         $clin exec "${psql_db}" -- sh -c "sed -i "s/^\s*#log_statement_sample_rate.*/log_statement_sample_rate=${LOG_STATEMENT_SAMPLE_RATE}/" /var/lib/pgsql/data/userdata/postgresql.conf "
     fi
+    if ${PSQL_EXPORT}; then
+        $clin exec "${psql_db}" -- sh -c 'sed -i "s/^\s*#track_io_timing.*/track_io_timing = on/" /var/lib/pgsql/data/userdata/postgresql.conf'
+        $clin exec "${psql_db}" -- sh -c 'sed -i "s/^\s*#track_wal_io_timing.*/track_wal_io_timing = on/" /var/lib/pgsql/data/userdata/postgresql.conf'
+        $clin exec "${psql_db}" -- sh -c 'sed -i "s/^\s*#track_functions.*/track_functions = all/" /var/lib/pgsql/data/userdata/postgresql.conf'
+        $clin exec "${psql_db}" -- sh -c 'sed -i "s/^\s*#stats_fetch_consistency.*/stats_fetch_consistency = cache/" /var/lib/pgsql/data/userdata/postgresql.conf'
+        $clin exec "${psql_db}" -- sh -c "echo shared_preload_libraries = \'pgaudit,auto_explain,pg_stat_statements\' >> /var/lib/pgsql/data/userdata/postgresql.conf"
+    fi
     echo "Restarting RHDH DB..."
     $clin rollout restart statefulset/"$psql_db_ss"
     wait_to_start statefulset "$psql_db_ss" 300 300
 
+    if ${PSQL_EXPORT}; then
+        $clin exec "${psql_db}" -- sh -c 'psql -c "CREATE EXTENSION pg_stat_statements;"'
+        uid=$(oc get namespace "${RHDH_NAMESPACE}" -o go-template='{{ index .metadata.annotations "openshift.io/sa.scc.supplemental-groups" }}'| cut -d '/' -f 1)
+        pg_pass=$(${clin} get secret rhdh-postgresql -o jsonpath='{.data.postgres-password}'|base64 -d)
+        plugins=("backstage_plugin_permission" "backstage_plugin_auth" "backstage_plugin_catalog" "backstage_plugin_scaffolder" "backstage_plugin_search" "backstage_plugin_app")
+        helm repo add prometheus-community https://prometheus-community.github.io/helm-charts
+        cp template/postgres-exporter/chart-values.yaml  "$TMP_DIR/pg-exporter.yaml"
+        sed -i "s/uid/$uid/g"  "$TMP_DIR/pg-exporter.yaml"
+        sed -i "s/pg_password/'$pg_pass'/g"  "$TMP_DIR/pg-exporter.yaml"
+        helm install pg-exporter prometheus-community/prometheus-postgres-exporter -n "${RHDH_NAMESPACE}" -f "$TMP_DIR/pg-exporter.yaml"
+        for plugin in "${plugins[@]}"; do
+            cp template/postgres-exporter/values-template.yaml "${TMP_DIR}/${plugin}.yaml"
+            sed -i "s/'dbname'/'$plugin'/" "${TMP_DIR}/${plugin}.yaml"
+            sed -i "s/uid/$uid/g"  "${TMP_DIR}/${plugin}.yaml"
+            sed -i "s/pg_password/'$pg_pass'/g"  "${TMP_DIR}/${plugin}.yaml"
+            helm_name=${plugin//_/-}
+            helm install "${helm_name}" prometheus-community/prometheus-postgres-exporter -n "${RHDH_NAMESPACE}" -f "${TMP_DIR}/${plugin}.yaml"
+        done
+    fi
+
     echo "Restarting RHDH..."
     $clin rollout restart deployment/"$rhdh_deployment"
     wait_to_start deployment "$rhdh_deployment" 300 300
+    if ${PSQL_EXPORT}; then
+        plugins=("pg-exporter" "backstage-plugin-permission" "backstage-plugin-auth" "backstage-plugin-catalog" "backstage-plugin-scaffolder" "backstage-plugin-search" "backstage-plugin-app")
+        for plugin in "${plugins[@]}"; do
+            cp template/postgres-exporter/service-monitor-template.yaml "${TMP_DIR}/${plugin}-monitor.yaml"
+            sed -i "s/pglabel/$plugin/" "${TMP_DIR}/${plugin}-monitor.yaml"
+            sed -i "s/pgnamespace/$RHDH_NAMESPACE/g"  "${TMP_DIR}/${plugin}-monitor.yaml"
+            $clin create -f "${TMP_DIR}/${plugin}-monitor.yaml"
+        done
+    fi
 }
 setup_monitoring() {
     echo "Enabling user workload monitoring"

diff --git a/ci-scripts/rhdh-setup/template/postgres-exporter/chart-values.yaml b/ci-scripts/rhdh-setup/template/postgres-exporter/chart-values.yaml
@@ -0,0 +1,25 @@
+podSecurityContext:
+  runAsGroup: uid
+  runAsUser: uid
+  runAsNonRoot: true
+  seccompProfile:
+    type: RuntimeDefault
+securityContext: {}
+config:
+  logLevel: 'debug'
+  extraArgs:
+    - --collector.long_running_transactions
+    - --collector.process_idle
+    - --collector.stat_activity_autovacuum
+    - --collector.stat_user_tables
+    - --collector.statio_user_indexes
+    - --collector.statio_user_tables
+    - --collector.postmaster
+    - --collector.stat_statements
+    - --auto-discover-databases
+  datasource:
+    host: 'rhdh-postgresql-primary'
+    user: 'postgres'
+    password: pg_password
+    port: "5432"
+    database: ''
diff --git a/ci-scripts/rhdh-setup/template/postgres-exporter/service-monitor-template.yaml b/ci-scripts/rhdh-setup/template/postgres-exporter/service-monitor-template.yaml
@@ -0,0 +1,18 @@
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  labels:
+    release: pglabel
+  name: prometheus-pglabel
+spec:
+  endpoints:
+  - interval: 30s
+    port: http
+    scheme: http
+    path: /metrics
+  namespaceSelector:
+    matchNames:
+      - pgnamespace
+  selector:
+    matchLabels:
+      release: pglabel
diff --git a/ci-scripts/rhdh-setup/template/postgres-exporter/values-template.yaml b/ci-scripts/rhdh-setup/template/postgres-exporter/values-template.yaml
@@ -0,0 +1,34 @@
+podSecurityContext:
+  runAsGroup: uid
+  runAsUser: uid
+  runAsNonRoot: true
+  seccompProfile:
+    type: RuntimeDefault
+securityContext: {}
+config:
+  logLevel: 'debug'
+  extraArgs:
+    - --no-collector.database
+    - --no-collector.database_wraparound
+    - --no-collector.locks
+    - --no-collector.long_running_transactions
+    - --no-collector.postmaster
+    - --no-collector.process_idle
+    - --no-collector.replication
+    - --no-collector.replication_slot
+    - --no-collector.stat_activity_autovacuum
+    - --no-collector.stat_bgwriter
+    - --no-collector.stat_database
+    - --no-collector.stat_statements
+    - --no-collector.stat_wal_receiver
+    - --no-collector.wal
+    - --no-collector.xlog_location
+    - --collector.stat_user_tables
+    - --collector.statio_user_indexes
+    - --collector.statio_user_tables
+  datasource:
+    host: 'rhdh-postgresql-primary'
+    user: 'postgres'
+    password: pg_password
+    port: "5432"
+    database: 'dbname'
diff --git a/config/cluster_read_config.populate.yaml b/config/cluster_read_config.populate.yaml
@@ -89,3 +89,226 @@ value }}
 - name: measurements.nodejs.populate.nodejs_gc_duration_seconds_major_average
   monitoring_query: sum(rate(nodejs_gc_duration_seconds_sum{kind="major",job="rhdh-metrics"}[5m]))/sum(rate(nodejs_gc_duration_seconds_count{kind="major",job="rhdh-metrics"}[5m]))
   monitoring_step: 15
+
+
+{% macro pg_query_sum(alias, query) -%}
+# Gather monitoring data about the db {{ alias }}
+- name: measurements.postgresql.populate.{{ alias }}.{{ query }}
+  monitoring_query: sum({{ query }}{service='{{ alias }}-prometheus-postgres-exporter'})
+  monitoring_step: 15
+{%- endmacro %}
+
+{% for query in [
+'pg_statio_user_indexes_idx_blks_hit_total',
+'pg_statio_user_indexes_idx_blks_read_total',
+'pg_statio_user_tables_heap_blocks_hit',
+'pg_statio_user_tables_heap_blocks_read',
+'pg_statio_user_tables_idx_blocks_hit',
+'pg_statio_user_tables_idx_blocks_read',
+'pg_statio_user_tables_tidx_blocks_hit',
+'pg_statio_user_tables_tidx_blocks_read',
+'pg_statio_user_tables_toast_blocks_hit',
+'pg_statio_user_tables_toast_blocks_read',
+'pg_stat_user_tables_vacuum_count',
+'pg_stat_user_tables_size_bytes',
+'pg_stat_user_tables_seq_tup_read',
+'pg_stat_user_tables_seq_scan',
+'pg_stat_user_tables_n_tup_upd',
+'pg_stat_user_tables_n_tup_ins',
+'pg_stat_user_tables_n_tup_hot_upd',
+'pg_stat_user_tables_n_tup_del',
+'pg_stat_user_tables_n_mod_since_analyze',
+'pg_stat_user_tables_n_live_tup',
+'pg_stat_user_tables_n_dead_tup',
+'pg_stat_user_tables_last_vacuum',
+'pg_stat_user_tables_last_autovacuum',
+'pg_stat_user_tables_last_autoanalyze',
+'pg_stat_user_tables_last_analyze',
+'pg_stat_user_tables_idx_tup_fetch',
+'pg_stat_user_tables_idx_scan',
+'pg_stat_user_tables_autovacuum_count',
+'pg_stat_user_tables_autoanalyze_count',
+'pg_stat_user_tables_analyze_count'
+] %}
+{% for db in [
+'backstage-plugin-permission',
+'backstage-plugin-auth',
+'backstage-plugin-catalog',
+'backstage-plugin-scaffolder',
+'backstage-plugin-search',
+'backstage-plugin-app'
+] %}
+{{ pg_query_sum(db, query ) }}
+{% endfor %}
+{% endfor %}
+
+{% macro pg_query(alias, query) -%}
+# Gather monitoring data about the db {{ alias }}
+- name: measurements.postgresql.populate.{{ alias }}.{{ query }}
+  monitoring_query: {{ query }}{datname="{{ alias }}"}
+  monitoring_step: 15
+{%- endmacro %}
+
+{% for query in [
+'pg_stat_database_blk_read_time',
+'pg_stat_database_blk_write_time',
+'pg_stat_database_blks_hit',
+'pg_stat_database_blks_read',
+'pg_stat_database_conflicts',
+'pg_stat_database_conflicts_confl_bufferpin',
+'pg_stat_database_conflicts_confl_deadlock',
+'pg_stat_database_conflicts_confl_lock',
+'pg_stat_database_conflicts_confl_snapshot',
+'pg_stat_database_conflicts_confl_tablespace',
+'pg_stat_database_deadlocks',
+'pg_stat_database_numbackends',
+'pg_stat_database_temp_bytes',
+'pg_stat_database_temp_files',
+'pg_stat_database_tup_deleted',
+'pg_stat_database_tup_fetched',
+'pg_stat_database_tup_inserted',
+'pg_stat_database_tup_returned',
+'pg_stat_database_tup_updated',
+'pg_stat_database_xact_commit',
+'pg_stat_database_xact_rollback',
+'pg_database_size_bytes'
+] %}
+{% for db in [
+'backstage_plugin_permission',
+'backstage_plugin_auth',
+'backstage_plugin_catalog',
+'backstage_plugin_scaffolder',
+'backstage_plugin_search',
+'backstage_plugin_app'
+] %}
+{{ pg_query(db, query ) }}
+{% endfor %}
+{% endfor %}
+
+{% macro pg_stat_statements_sum(alias, query) -%}
+# Gather monitoring data about the db {{ alias }}
+- name: measurements.postgresql.populate.{{ alias }}.{{ query }}
+  monitoring_query: sum({{ query }}{datname='{{ alias }}'})
+  monitoring_step: 15
+{%- endmacro %}
+
+{% for query in [
+'pg_stat_statements_block_read_seconds_total',
+'pg_stat_statements_block_write_seconds_total',
+'pg_stat_statements_calls_total',
+'pg_stat_statements_rows_total',
+'pg_stat_statements_seconds_total',
+'pg_locks_count'
+] %}
+{% for db in [
+'backstage_plugin_permission',
+'backstage_plugin_auth',
+'backstage_plugin_catalog',
+'backstage_plugin_scaffolder',
+'backstage_plugin_search',
+'backstage_plugin_app'
+] %}
+{{ pg_stat_statements_sum(db, query ) }}
+{% endfor %}
+{% endfor %}
+
+
+{% macro pg_settings(query) -%}
+# Gather monitoring data about the db {{ alias }}
+- name: measurements.postgresql.populate.{{ query }}
+  monitoring_query: {{ query }}{service="pg-exporter-prometheus-postgres-exporter"}
+  monitoring_step: 30
+{%- endmacro %}
+
+{% for query in [
+'pg_settings_max_connections',
+'pg_settings_superuser_reserved_connections',
+'pg_settings_shared_buffers_bytes',
+'pg_settings_work_mem_bytes',
+'pg_settings_maintenance_work_mem_bytes',
+'pg_settings_shared_memory_size_in_huge_pages',
+'pg_settings_effective_cache_size_bytes',
+'pg_settings_effective_io_concurrency',
+'pg_settings_random_page_cost',
+'pg_settings_track_io_timing',
+'pg_settings_max_wal_senders',
+'pg_settings_checkpoint_timeout_seconds',
+'pg_settings_checkpoint_completion_target',
+'pg_settings_max_wal_size_bytes',
+'pg_settings_min_wal_size_bytes',
+'pg_settings_wal_buffers_bytes',
+'pg_settings_wal_writer_delay_seconds',
+'pg_settings_wal_writer_flush_after_bytes',
+'pg_settings_bgwriter_delay_seconds',
+'pg_settings_bgwriter_lru_maxpages',
+'pg_settings_bgwriter_lru_multiplier',
+'pg_settings_bgwriter_flush_after_bytes',
+'pg_settings_max_worker_processes',
+'pg_settings_max_parallel_workers_per_gather',
+'pg_settings_max_parallel_maintenance_workers',
+'pg_settings_max_parallel_workers',
+'pg_settings_parallel_leader_participation',
+'pg_settings_enable_partitionwise_join',
+'pg_settings_enable_partitionwise_aggregate',
+'pg_settings_jit',
+'pg_settings_max_slot_wal_keep_size_bytes',
+'pg_settings_track_wal_io_timing',
+'pg_settings_maintenance_io_concurrency',
+'pg_settings_wal_recycle',
+'pg_process_idle_seconds_sum',
+'pg_process_idle_seconds_count',
+'pg_stat_bgwriter_buffers_alloc_total',
+'pg_stat_bgwriter_buffers_backend_fsync_total',
+'pg_stat_bgwriter_buffers_backend_total',
+'pg_stat_bgwriter_buffers_checkpoint_tota',
+'pg_stat_bgwriter_buffers_clean_total',
+'pg_stat_bgwriter_checkpoint_sync_time_total',
+'pg_stat_bgwriter_checkpoint_write_time_total',
+'pg_stat_bgwriter_checkpoints_req_total',
+'pg_stat_bgwriter_checkpoints_timed_total',
+'pg_stat_bgwriter_maxwritten_clean_total',
+'pg_stat_archiver_archived_count',
+'pg_stat_archiver_failed_count',
+'pg_long_running_transactions',
+'pg_long_running_transactions_oldest_timestamp_seconds',
+'pg_wal_segments',
+'pg_wal_size_bytes',
+'process_cpu_seconds_total',
+'process_max_fds',
+'process_open_fds',
+'process_resident_memory_bytes',
+'process_virtual_memory_bytes',
+'process_virtual_memory_max_bytes',
+] %}
+{{ pg_settings( query ) }}
+{% endfor %}
+
+{% macro pg_stat_activity(alias, query, state) -%}
+# Gather monitoring data about the db {{ alias }}
+- name: measurements.postgresql.populate.{{ alias }}.{{ query }}.{{ state }}
+  monitoring_query: sum({{ query }}{datname='{{ alias }}',state='{{ state }}',service="pg-exporter-prometheus-postgres-exporter"})
+  monitoring_step: 15
+{%- endmacro %}
+
+{% for query in [
+'pg_stat_activity_count',
+'pg_stat_activity_max_tx_duration'
+] %}
+{% for db in [
+'backstage_plugin_permission',
+'backstage_plugin_auth',
+'backstage_plugin_catalog',
+'backstage_plugin_scaffolder',
+'backstage_plugin_search',
+'backstage_plugin_app'
+] %}
+{% for state in [
+'active',
+'disabled',
+'fastpath',
+'idle'
+] %}
+{{ pg_stat_activity(db, query, state ) }}
+{% endfor %}
+{% endfor %}
+{% endfor %}