尽管将setBufferError设置为0，但我已经测量到keyBy操作的延迟在大多数情况下通常在0-1 ms之间。然而，随着程序继续运行，我偶尔会观察到延迟达到数千毫秒。这是正常行为吗？
在这段代码中，我记录了keyBy操作前后的时间，然后我注意到了前面提到的情况。

originalTrajectories = originalTrajectories.keyBy(t -> t.getGridID()).map(new MapFunction<Trajectory1, Trajectory1>() {
            @Override
            public Trajectory1 map(Trajectory1 value) {
                value.t2 = System.currentTimeMillis();
                return value;
            }
        });

queryTrajectories = queryTrajectories.keyBy(t -> t.getGridID()).map(new MapFunction<Trajectory1, Trajectory1>() {
            @Override
            public Trajectory1 map(Trajectory1 value) {
                value.t2 = System.currentTimeMillis();
                return value;
            }
        });

DataStream<Tuple4<Long, Long, Trajectory1, Trajectory1>> timedResultTrajectories = originalTrajectories
            .keyBy(t -> t.getGridID())
            .intervalJoin(queryTrajectories.keyBy(t -> t.getGridID()))
            .between(Time.milliseconds(-slideStep * 1000), Time.milliseconds(slideStep * 1000))
            .process(new ProcessJoinFunction<Trajectory1, Trajectory1, Tuple4<Trajectory1, Trajectory1, Long, Double>>() {
                @Override
                public void processElement(Trajectory1 t1, Trajectory1 t2, ProcessJoinFunction<Trajectory1, Trajectory1, Tuple4<Trajectory1, Trajectory1, Long, Double>>.Context ctx, Collector<Tuple4<Trajectory1, Trajectory1, Long, Double>> out) {
                    if (Math.abs(t1.getTimestamp() - t2.getTimestamp()) == 0) {
                        Double distance = calculateClosestPairDistance(t1, t2);
                        if (distance <= queryRadius) {
                            out.collect(Tuple4.of(t1, t2, System.currentTimeMillis(), distance));
                        }
                    }
                }
            })

字符串