Pool implementation for one-shot queries does not use all available resources

[tsegismont]

When executing one-shot queries, the pool implementation will often schedule the SQL commands on the few available connections instead of creating more if maximum size has not been reached.

Consider the following test:

  @Test
  public void testUseAvailableResources(TestContext ctx) {
    int poolSize = 10;
    Async async = ctx.async();
    PgPool pool = PgPool.pool(options, new PoolOptions().setMaxSize(poolSize));
    AtomicReference<PgConnection> ctrlConnRef = new AtomicReference<>();
    PgConnection.connect(vertx, options, ctx.asyncAssertSuccess(ctrlConn -> {
      ctrlConnRef.set(ctrlConn);
      for (int i = 0; i < poolSize; i++) {
        vertx.setTimer(10 * (i + 1), l -> {
          pool.query("select pg_sleep(5)").execute(ar -> {});
        });
      }
      vertx.setTimer(10 * (poolSize + 1), event -> {
        ctrlConn.query("select count(*) as cnt from pg_stat_activity where application_name like '%vertx%'").execute(ctx.asyncAssertSuccess(rows -> {
          Integer count = rows.iterator().next().getInteger("cnt");
          ctx.assertEquals(poolSize + 1, count);
          async.complete();
        }));
      });
    }));
    try {
      async.await();
    } finally {
      PgConnection ctrlConn = ctrlConnRef.get();
      if (ctrlConn != null) {
        ctrlConn.close();
      }
      pool.close();
    }
  }

It fails like this:

java.lang.AssertionError: Not equals : 11 != 2

This test targets Postgres but the problem is in PoolBase implementation so I believe it affects all SQL clients.