percent_usage_connections

Connection utilization

Dimensions:None

Technical Annotations (21)

Configuration Parameters (6)

pool_moderecommended: transaction

Returns connection after each transaction, recommended for most use cases

default_pool_sizerecommended: 20

Actual database connections per database/user pair

max_client_connrecommended: 1000

Maximum application connections to pooler

query_timeoutrecommended: 30

Kill queries running longer than 30 seconds to prevent connection starvation

max_connections

defines maximum simultaneous connections allowed

superuser_reserved_connectionsrecommended: 3

Connection slots reserved for superusers to maintain admin access during exhaustion

Error Signatures (1)

FATAL: sorry, too many clients alreadylog pattern

CLI Commands (6)

SELECT query, calls, total_exec_time, mean_exec_time, rows FROM pg_stat_statements ORDER BY total_exec_time DESC LIMIT 20;

diagnostic

SELECT * FROM pg_stat_activitydiagnostic

SELECT state, COUNT(*) AS connection_count, MAX(EXTRACT(EPOCH FROM (now() - state_change))) AS max_age_seconds FROM pg_stat_activity WHERE pid <> pg_backend_pid() GROUP BY state;

diagnostic

SELECT max_conn, used, max_conn - used AS available, ROUND((used::float / max_conn) * 100, 2) AS usage_percent FROM ( SELECT (SELECT setting::int FROM pg_settings WHERE name = 'max_connections') AS max_conn, (SELECT COUNT(*) FROM pg_stat_activity) AS used ) AS conn_stats;

diagnostic

gcloud sql instances describe INSTANCE_NAME --format='get(settings.tier)'Get machine type to calculate vCPU count and actual connection limit

SELECT COUNT(*) as current_connections, setting::int as max_configured FROM pg_stat_activity, pg_settings WHERE pg_settings.name = 'max_connections' GROUP BY setting;

Compare current connections to configured max connections (may not reflect Cloud SQL vCPU limit)

Technical References (8)

PgBouncercomponent/etc/pgbouncer/pgbouncer.inifile pathcheck_postgrescomponentmax_connectionscomponentpgBouncercomponentpg_stat_activitycomponentpg_terminate_backend()componentpgpool-IIcomponent

Related Insights (7)

Connection limit exhaustion causes memory pressure and context-switching overheadcritical

▸

Connection count approaching max_connections causes connection failurescritical

▸

High connection count approaching limits degrades monitoring performancewarning

▸

Connection exhaustion prevents new client connectionscritical

Each PostgreSQL connection consumes 5-10MB RAM via fork model. At 200 connections: 1-2GB RAM overhead before queries run. At 500+ connections: excessive context-switching dominates query execution time. Database becomes unresponsive under concurrent load.

▸

Connection limit approaching maximum causes application timeoutscritical

▸

Connection exhaustion prevents new client connectionscritical

▸

Cloud SQL per-vCPU connection limits silently override PostgreSQL max_connectionscritical

Cloud SQL enforces hard connection limits based on instance vCPU count (25 connections per vCPU) that override PostgreSQL's configured max_connections parameter. Applications hitting this limit receive 'FATAL: sorry, too many clients already' errors even when PostgreSQL metrics show connections well below max_connections.

▸