System insights metrics reference

The following table describes the metrics charts that can be viewed on the AlloyDB System insights dashboard and in the Metrics Explorer. The metric type strings must be prefixed with alloydb.googleapis.com/. For example, alloydb.googleapis.com/Instance.

Metrics are either available in BETA or GA Launch stages. For more information about the launch stage of the metric, see AlloyDB metrics.

Display name and metric name	Description	Cluster metric	Instance metric	Read pool metric
Rows in database by state database/postgresql/tuples	The number of live and dead rows in the instance. Only available for instances with the number of databases less than 50.	Yes	Yes	No
Query latency database/postgresql/insights/aggregate/latencies	The aggregated P99 query latency distribution for each instance. Only available for instances with Query Insights enabled.	Yes	Yes	Yes
Database load database/postgresql/insights/aggregate/execution_time	The accumulated query execution time for each instance. This is the sum of CPU time, I/O wait time, lock wait time, process context switch, and scheduling for all processes involved in query execution. Only available for instances with Query Insights enabled.	Yes	Yes	Yes
Max CPU utilization instance/cpu/maximum_utilization	The maximum CPU utilization calculated across all serving nodes of the instance, represented as a percentage for each instance.	Yes	Yes	Yes
Mean CPU utilization instance/cpu/average_utilization	The mean CPU utilization calculated as an average across all serving nodes of the instance, represented as a percentage for each instance.	Yes	No	Yes
Minimum available memory instance/memory/min_available_memory	The minimum available memory across all serving nodes of the instance. The available memory is an estimate of memory in bytes available for allocation on the VM including memory that is used, but can be potentially freed.	Yes	Yes	Yes
Cluster storage cluster/storage/usage	The billable storage that the AlloyDB cluster uses. It does not include the storage that automatic and on-demand backups use.	Yes	No	No
Queries IO wait time database/postgresql/insights/aggregate/io_time	The aggregated queries I/O wait time for each instance. Only available for instances with Query Insights enabled.	Yes	Yes	No
Number of connections instance/postgresql/backends_by_state	The number of connections to each instance across all serving nodes of the instance. This includes connections in all states such as idle, active, idle_in_transaction, idle_in_transaction_aborted, disabled, and fastpath_function_call.	Yes	Yes	Yes
Rows fetched vs. rows returned vs. rows written instance/postgresql/fetched_tuples_count instance/postgresql/returned_tuples_count instance/postgresql/written_tuples_count	Rows fetched: The aggregated number of rows fetched as a result of queries across the cluster or instance. Rows returned: The aggregated number of rows scanned while processing the queries across the cluster or instance. Rows written: The aggregated number of rows written while performing insert, update, and delete operations across the cluster or instance. Only available for instances with the number of databases less than 5000.	Yes	Yes	No
Rows fetched vs. rows returned instance/postgresql/fetched_tuples_count instance/postgresql/returned_tuples_count	Rows fetched: The number of rows fetched as a result of queries across all serving nodes of the read pool instance. Rows returned: The number of rows scanned while processing the queries across all serving nodes of the read pool instance. If the difference between rows returned and rows fetched is too large that their values aren't shown in the same scale, then the value of rows fetched is shown as 0 because it's negligible compared to the value of rows returned.	No	No	Yes
Rows processed by operation database/postgresql/ inserted_tuples_count_for_top_databases database/postgresql/ updated_tuples_count_for_top_databases database/postgresql/ deleted_tuples_count_for_top_databases	The number of rows processed in the primary instance per operation such as insert, update, or delete. Only available for instances with the number of databases less than 5000.	Yes	No	No
Oldest transaction age postgresql/vacuum/oldest_transaction_age	The current age of the oldest uncommitted transaction for each instance that is blocking the vacuum operation. It's measured in the number of transactions that started after the oldest transaction.	Yes	Yes	Yes
Transaction ID utilization postgresql/vacuum/ transaction_id_utilization	The current percentage of transaction ID space that's consumed by each instance. It records up to a maximum of two billion unvacuumed transactions as a percentage.	Yes	Yes	Yes
Wait time instance/postgresql/wait_time	The aggregated accumulated elapsed wait time per wait event name and wait event type across the cluster, instance, or across all serving nodes of the read pool instance.	Yes	Yes	Yes
Wait count instance/postgresql/wait_count	The number of times processes waited per wait event name and wait event type across the cluster, instance, or across all serving nodes of the read pool instance.	Yes	Yes	Yes
Temp data size used instance/postgresql/ temp_bytes_written_count	The amount of data (bytes) used in each instance for executing queries and performing algorithms such as join and sort. Only available for instances with the number of databases less than 5000.	Yes	No	No
Temp files used instance/postgresql/temp_files_written_count	The number of temporary files that are used in each instance for executing queries and performing algorithms such as join and sort. Only available for instances with the number of databases less than 5000.	Yes	No	No
New connection count instance/postgresql/new_connections_count	The number of new connections that are added to each instance. Only available for instances with the number of databases less than 5000.	Yes	No	No
Statements executed count database/postgresql/statements_executed_count	The number of statements that are executed in each instance. Only available for instances with Query Insights enabled.	Yes	Yes	No
Transaction count instance/postgres/transaction_count	The number of transactions that have occurred in each instance. Only available for instances with the number of databases less than 5000.	Yes	No	No
Block read count instance/postgresql/blks_hit instance/postgresql/blks_read	The number of blocks read from the disk and from the buffer cache across the cluster, instance, or across all serving nodes of the read pool instance. Only available for instances with the number of databases less than 5000.	Yes	Yes	Yes
Deadlock count instance/postgresql/deadlock_count	The number of deadlocks in each instance. Only available for instances with the number of databases less than 5000.	Yes	No	No
Read pool replication lag instance/postgres/replication/maximum_lag	The maximum time taken for replication from a primary instance to a read pool. The time lag is calculated across all serving read pool nodes per read pool instance in the cluster.	Yes	Yes	Yes
Active nodes instance/postgres/instances	The number of nodes that can serve user traffic.	Yes	No	Yes
Connections per database database/postgresql/backends_for_top_databases	The number of connections to the instance per database.	No	Yes	Yes
Connections per application instance/postgresql/backends_for_top_applications	The number of connections to the instance, grouped by applications. Connections without an application name are grouped as Unknown.	No	Yes	Yes
Rows processed by operation instance/postgresql/inserted_tuples_count instance/postgresql/updated_tuples_count instance/postgresql/deleted_tuples_count	The number of rows processed in the primary instance per operation such as insert, update, or delete.	No	Yes	No
Temp data size used per database database/postgresql/temp_bytes_written_for_top_databases	The amount of data (bytes) used in each database for executing queries and performing algorithms such as join and sort.	No	Yes	Yes
Temp files used per database database/postgresql/temp_files_written_for_top_databases	The number of temporary files that are used in each database for executing queries and performing algorithms such as join and sort.	No	Yes	Yes
New connection count per database database/postgresql/new_connections_ for_top_databases	The number of new connections that are added to the instance per database.	No	Yes	Yes
Transaction count database/postgresql/transaction_count	The number of committed and rolled-back transactions that have occurred in the instance.	No	Yes	Yes
Deadlock count per database database/postgresql/deadlock_count_ for_top_databases	The number of deadlocks in the instance per database.	No	Yes	Yes
Nodes and replication status instance/postgres/replication/replicas	The number of read nodes connected to the primary instance per read pool along with one of the following states: startup, catchup, streaming, backup, and stopping.	No	No	Yes
Ultra Fast Cache hit rate per instance instance/postgres/ultrafastcache_hitrate	The hit rate of AlloyDB ultra-fast cache across all the nodes in an instance. The chart might be empty if there are no hits in the ultra-fast cache and all reads are served from the storage layer. The chart might have gaps if there are no read operations on the database or if all reads are served from the buffer cache.	No	Yes	Yes
Ultra Fast Cache hit rate per node node/postgres/ultrafastcache_hitrate	The hit rate of AlloyDB ultra-fast cache per node. The chart might be empty if there are no hits in the ultra-fast cache and all reads are served from the storage layer. The chart might have gaps if there are no read operations on the database or if all reads are served from the buffer cache.	No	Yes	Yes
Node CPU usage node/cpu/usage_time	The rate of CPU utilization in the node, which is sampled every 60 seconds. After sampling, data is not visible for up to 180 seconds.	No	Yes	Yes
Node replication lag node/postgres/replay_lag	The lag in milliseconds for the individual node, derived from replay_lag in pg_stat_replication. This metric is sampled every 60 seconds. After sampling, data is not visible for up to 180 seconds.	No	Yes	Yes
Node database availability node/postgres/uptime	The rate of database availability in the node, which sampled every 60 seconds. After sampling, data is not visible for up to 180 seconds.	No	Yes	Yes
Number of connections to the node per connection state node/postgres/backends_by_state	The number of connections to the node grouped by the status: idle, active, idle_in_transaction, idle_in_transaction_aborted, disabled, and fastpath_function_call. For information about these statuses, see the state text row in the pg_stat_activity documentation.	No	Yes	Yes
Node Wait count node/postgres/wait_count	Total number of times processes waited for each wait event type and wait event name per node.	No	Yes	Yes
Node Wait time node/postgres/wait_time	Total elapsed wait time (microseconds) for each wait event type and wait event name per node.	No	Yes	Yes

Additionally, the Cloud Logging metric, Log entries by severity (logging.googleapis.com/log_entry_count), shows the total numbers of error and warning log entries per instance. These are extracted from postgres.log, which is the database log, and pgaudit.log, which contains data access information.

For more information, see AlloyDB metrics.