The connection pool size decreases from 20 (maximum pool size) to zero [Azure SQL Managed Instance]

[lucaoliano]

Hi
we are experiencing issues similar to #1474 where the total connections gradually decrease from the maximum pool size, 20, down to zero, see evidence from the debug log info:

2024-09-17 09:14:38.000 DEBUG 603 --- [l-1 housekeeper] com.zaxxer.hikari.pool.HikariPool : HikariPool-1 - Pool stats (total=19, active=0, idle=19, waiting=0)
…
2024-09-17 09:15:08.000 DEBUG 603 --- [l-1 housekeeper] com.zaxxer.hikari.pool.HikariPool : HikariPool-1 - Pool stats (total=18, active=0, idle=18, waiting=0)
…
2024-09-17 09:43:38.092 DEBUG 603 --- [l-1 housekeeper] com.zaxxer.hikari.pool.HikariPool : HikariPool-1 - Pool stats (total=0, active=0, idle=0, waiting=1)

When the total connections reach 0, all requests to the connection pool fail due to connection not available and the following exception is thrown:

org.springframework.jdbc.CannotGetJdbcConnectionException: Failed to obtain JDBC Connection; nested exception is java.sql.SQLTransientConnectionException: HikariPool-1 - Connection is not available, request timed out after 30002ms.

Here's the details for our application:

• DB is Azure SQL Managed Instance, with Proxy connection policy
• Hikari connection pool version is 4.0.3 we only set max pool size. all others settings are left to default
• JDBC driver is MS SQL 12.4.0

We reached Microsoft support, who confirmed that Azure SQL Managed Instance (SQL MI) utilizes an idle connection timeout mechanism, where connections that remain idle for approximately 30 minutes are automatically closed. This behavior is managed by the SQL MI proxy layer to free up resources and maintain optimal performance. To prevent connections from timing out, they recommend to implement a keep-alive mechanism or send periodic queries to keep the connection active.

However, they suggested opening a ticket on GitHub to HikariCP since they don’t have enough expertise or a dedicated team to provide specific settings.

Can you advice on what to do to resolve?

1. is there any known bug with a fix version?
2. or should we try with any specific settings e.g. maxLifeTime and keepAliveTime settings as below?

MaxLifetime: Setting this to 15 minutes (or 900,000 milliseconds) ensures that the lifetime of connections in the pool is shorter than the database or infrastructure-imposed connection timeout, such as the 30-minute idle timeout for Azure SQL Managed Instance. This helps avoid situations where the database closes idle connections unexpectedly.
KeepaliveTime: Setting this to 5 minutes (or 300,000 milliseconds) is a good approach to ensure that HikariCP sends periodic keep-alive queries. This should be less than the maxLifetime to avoid conflicts and ensure idle connections stay active.

[quaff]

I think you should set idleTimeout less than 30 minutes.