DigitalOcean Managed Databases include metrics visualizations so you can monitor performance and health of your database cluster. There are two kinds of metrics:
Cluster metrics monitor the performance of the nodes in a database cluster. Cluster metrics cover primary and standby nodes; metrics for each read-only node are displayed independently. This data can help guide capacity planning and optimization. You can also set up alerting on cluster metrics.
Database metrics monitor the performance of the database itself. This data can help assess the health of the database, pinpoint performance bottlenecks, and identify unusual use patterns that may indicate an application bug or security breach.
To view performance metrics for a PostgreSQL database cluster, click the name of the database to go to its Overview page, then click the Insights tab.
The Select object drop-down menu lists the cluster itself and all of the databases in the cluster. Choose the database to view its metrics.
In the Select Period drop-down menu, you can choose a time frame for the x-axis of the graphs, ranging from 1 hour to 30 days. Each line in the graphs will display about 300 data points.
By default, the summary to the right shows the most recent metrics values. If you hover over a different time in a graph, the summary will display the values from that time instead.
If you recently provisioned the cluster or changed its configuration, it may take a few minutes for the metrics data to finish processing before you see it on the Insights page.
PostgreSQL databases have the following metrics:
The connections plot displays the number of client connections to the database and the connection limit as defined by the amount of memory associated with your primary node. Once the limit is reached, new client connections will be blocked until existing ones are closed.
You can use a connection pooling utility to avoid connection contention. Learn more about managing connection pools.
The cache hit ratio plot tracks read efficiency as measured by the proportion of reads from cache versus the total reads from both disk and cache. With the exception of data warehouse use cases, an ideal cache hit ratio is 99% or higher, meaning that at least 99% of reads are from cache and no more than 1% are from disk.
If your cache hit ratio is consistently lower than 99%, consider upgrading to a plan with additional memory to increase your cache size.
The index vs sequential scans plot displays the percentage of index scans as proportion of all scans, index and sequential, across all user table in the database. Indexes are pointers to table data and make data retrieval more efficient than when using row-based sequential scans. Ideally for read-heavy use cases, the proportion of index scans should be 99% for larger tables.
If you are using larger tables and the proportion of index scans is consistently much lower than 99%, ensure that your larger tables are indexed.
The throughput plot records the rate of row fetches, row inserts, row updates, and row deletes across all user tables in the database. Monitoring the overall usage pattern is useful for making tuning decisions and identifying potential problems. For example, unexpected changes in usage patterns could indicate a newly introduced bug or security breach.
This data is also useful for understanding how efficiently your database handles each type of operation and identifying opportunities to improve performance through tuning, design modifications, and scaling. For example, while indexes are helpful for improving performance in read-heavy use cases, they can slow down insert, update and delete (DML) operations.
The deadlocks plot shows the rate of deadlock creation in the database. Deadlocks occur when two or more transactions have simultaneous, conflicting locks on the same database object. PostgreSQL will abort at least one of the deadlocked transactions.
To identify the transactions involved in a deadlock, refer to the deadlock error details in your PostgreSQL logs. Look for log entries with
process 12345 detected deadlock. It may also be helpful to correlate the PostgreSQL error timestamp with the same time point in your application logs to understand under what conditions the deadlock was triggered.
You can prevent deadlocks by ensuring that all applications that use the database acquire locks on multiple objects in a consistent order.
If you have a standby node configured, the replication status plot records the lag in replicating data from primary to standby node(s), as expressed in bytes.
Significant replication lags could indicate a network connectivity problem or insufficient CPU resources.