SQL state [0A000] keeps recurring, seems unrecoverable

[martijnhaex]

Environment

HikariCP version: x.x.x
JDK version     : 1.8.0_144
Database        : PostgreSQL
Driver version  : 9.4.1212.jre7

---

When doing a rolling deploy of our application (over multiple application nodes).
The first node to be redeployed with the new version will apply database changes through Flyway.

Exception: SQL state [0A000]; error code [0]; ERROR: cached plan must not change result type; nested exception is org.postgresql.util.PSQLException: ERROR: cached plan must not change result type.

PostgreSQL errorcode 0A000 (FEATURE NOT SUPPORTED)
This seems to be an unrecoverable exception.
But isn't listed in the SQL_ERRORS list (ProxyConnection) so the connection will not be closed.
So the exception keeps recurring.

[brettwooldridge]

In your opinion, evicting the connection upon receipt of 0A000 would allow the application to continue?

[martijnhaex]

Indeed.

Tried this locally. I placed a custom implementation of ProxyConnection on the classpath to override the default ProxyConnection implementation. Upon receiving the 0A000 errorcode the connection will be closed.
If this doesn't happen then the connection stays in the pool in a corrupted state, because the cached plan is saved per connection. Seems to me that the connection is then corrupt or the cached plan should be removed from that connection upon receiving this errorcode.