Google Cloud SQL insights

Sustained high query rates combined with increased latency indicate the instance is approaching throughput limits. This pattern often precedes resource exhaustion and requires proactive scaling.

Long-running or sub-optimal queries consuming excessive resources can be identified and terminated through Query Insights Enterprise Plus to free resources, unblock critical operations, and prevent resource exhaustion.

CPU-intensive queries on read replicas (sorting, regex, complex functions) can cause replication lag by consuming CPU needed for replication apply workers, especially when replica vCPUs are insufficient.

High network throughput combined with replication lag indicates network bandwidth saturation between primary and replica, especially during initial replica creation or after primary disk space recovery.

When client_backend_create_time and client_backend_wait_time are elevated, new connections are queuing because the instance is saturated with connection requests. This often precedes connection pool exhaustion and instance instability.

The most common cause of Cloud SQL Auth Proxy connection failures is the Cloud SQL Admin API not being enabled in the project. The proxy requires this API to create ephemeral certificates and retrieve instance metadata.

With Query Insights Enterprise Plus 30-day retention, comparing query plans over time reveals when optimizer choices change and cause performance degradation without any application code changes. This indicates statistics staleness or plan instability.

When CPU utilization is elevated alongside high disk I/O operations, queries are experiencing I/O contention rather than pure compute load. This pattern indicates disk throughput bottleneck, not CPU shortage.

When primary instance disk becomes full during replica creation or normal operations, replica lag increases dramatically and replication may fail. Disk space issues on primary directly impact replication health and can cause WAL segment accumulation.

Monitoring cloudsql_googleapis_com/database/up and cloudsql_googleapis_com/database/uptime provides critical availability signals for SLA compliance and detecting unplanned restarts or outages.

When CPU utilization approaches reserved cores capacity, query performance degrades and connection handling slows. Understanding reserved vs. utilization metrics is critical for capacity planning.

Network sent/received bytes approaching instance network bandwidth limits can cause query timeouts, replication lag, and connection failures. Cloud SQL instances have network bandwidth limits based on machine type.

High client_backend_wait_time combined with low client_backend_usage indicates application connection pool misconfiguration or excessive connection churn, even when database-side cloudsql_googleapis_com/database/network/connections appears normal.

Query Insights Enterprise Plus stores detailed metrics data on the instance disk (approximately 180GB for 30-day retention), requiring automatic storage increases to be enabled to prevent disk exhaustion and data loss.

Query Insights for Enterprise Plus reveals specific wait events (disk I/O, locks, etc.) causing query slowdowns that aggregate metrics alone cannot diagnose. Wait event analysis provides granular root cause identification for query performance issues.

High memory utilization combined with increased disk read operations indicates insufficient buffer cache, causing excessive disk I/O and query performance degradation. For PostgreSQL, buffer cache hit ratio below 99% signals memory pressure.

Long-running import/export operations block other critical operations including automated daily backups, creating backup gaps and recovery risks. Only one import/export/backup can run at a time per instance.

When disk space reaches 100% utilization (cloudsql_googleapis_com/database/disk/bytes_used >= cloudsql_googleapis_com/database/disk/quota), the instance cannot write new data or recover from restart, even after deleting data, because disk space is not immediately reclaimed.

Sustained high write operation rates on the primary instance can cause replication lag on read replicas, especially when disk I/O becomes saturated. This creates cascading performance degradation affecting read workloads.

When Cloud SQL instances use private IP only, VMs without external IPs cannot connect unless Private Google Access is enabled on their subnet. This causes proxy dial failures, direct connection timeouts, and prevents API access required for authentication.

Excessive table counts degrade instance performance, risk SLA violations, and can cause availability issues. Cloud SQL recommends keeping table counts reasonable (< 10,000 tables) to maintain performance.

When memory utilization is high and disk read operations increase significantly, the database buffer cache is undersized, forcing queries to read from disk instead of memory. This degrades read performance and increases query latency.