HikariCP should be bulletproof with regards to broken connections.

作者说, HikariCP 是对死掉的connection防弹。

Generally speaking, HikariCP should be bulletproof with regards to broken connections.  In the case where the broken connection is detected by the com.zaxxer.hikari.proxy.ConnectionProxy there should absolutely be no way that that connection ever comes out of the pool again.

However, there is a possibility that the application (or framework in the stack) acquires a Connection, performs some action (which succeeds at commit()), but hangs onto the Connection rather than releasing it back quickly.  Subsequently, upon attempting to use the Connection, the application may find that it is dead.  In this case, HikariCP never got the connection back, and therefore could not detect that it was broken.  I have definitely seen Hibernate do this in the past -- based on configuration of its acquisition_mode or release_mode.

I suggest reading this issue to see if it helps, especially the last few comments.

Google forum:
!topic/hikari-cp/IpUjIJLA3H8