SqlConnectionPool.execute() does not cancel the connection-timeout timer when acquire fails → IllegalStateException: Result is already complete #1706

Version

4.5.34

Context

When the database is unreachable and a query is executed through the pool (Pool#query(...).execute() / preparedQuery), the caller correctly receives the connection error. However, once the pool connection timeout elapses, Vert.x reports an uncaught exception:

java.lang.IllegalStateException: Result is already complete
    at io.vertx.core.Promise.fail(Promise.java:90)
    at io.vertx.sqlclient.impl.pool.SqlConnectionPool.lambda$execute$3(SqlConnectionPool.java:216)
    at io.vertx.core.impl.VertxImpl$InternalTimerHandler.handle(...)
    ...

In SqlConnectionPool.execute(ContextInternal, CommandBase<R>, long timeout):

1. If timeout > 0, a timer is registered that calls res.fail(POOL_QUERY_TIMEOUT_EXCEPTION).
2. The timer is only cancelled in the success branch of p.future().compose(lease -> ...).
3. In the failure branch (t -> { dequeueAndReject(...); return Future.failedFuture(t); }) the timer is not cancelled; onComplete then calls res.fail(ar.cause()).
4. When the timer later fires, res.fail(...) is invoked on an already completed promise → IllegalStateException.

This is deterministic: every acquire failure that happens before the timeout expires produces the exception exactly timeout ms later.

Steps to reproduce

1. Create a Pool with PoolOptions#setConnectionTimeout(2) (seconds) pointing to an unreachable host/port (or a stopped PostgreSQL container).
2. Execute a query via the pool; it fails with the connection error (expected).
3. Wait ~3 seconds without further requests.
4. Observe the uncaught IllegalStateException from the timer handler.

Do you have a reproducer?

https://gist.github.com/Rene2000k/8e0e0a2aaec8fdd13e6ba8e567bdd301