use of org.apache.beam.sdk.transforms.windowing.SlidingWindows in project beam by apache.
the class BeamAggregationRel method explainTerms.
@Override
public RelWriter explainTerms(RelWriter pw) {
super.explainTerms(pw);
if (this.windowFn != null) {
WindowFn windowFn = this.windowFn;
String window = windowFn.getClass().getSimpleName() + "($" + String.valueOf(windowFieldIndex);
if (windowFn instanceof FixedWindows) {
FixedWindows fn = (FixedWindows) windowFn;
window = window + ", " + fn.getSize().toString() + ", " + fn.getOffset().toString();
} else if (windowFn instanceof SlidingWindows) {
SlidingWindows fn = (SlidingWindows) windowFn;
window = window + ", " + fn.getPeriod().toString() + ", " + fn.getSize().toString() + ", " + fn.getOffset().toString();
} else if (windowFn instanceof Sessions) {
Sessions fn = (Sessions) windowFn;
window = window + ", " + fn.getGapDuration().toString();
} else {
throw new UnsupportedOperationException("Unknown window function " + windowFn.getClass().getSimpleName());
}
window = window + ")";
pw.item("window", window);
}
return pw;
}
use of org.apache.beam.sdk.transforms.windowing.SlidingWindows in project beam by apache.
the class SplittableDoFnTest method testPairWithIndexWindowedTimestamped.
private void testPairWithIndexWindowedTimestamped(IsBounded bounded) {
// Tests that Splittable DoFn correctly propagates windowing strategy, windows and timestamps
// of elements in the input collection.
MutableDateTime mutableNow = Instant.now().toMutableDateTime();
mutableNow.setMillisOfSecond(0);
Instant now = mutableNow.toInstant();
Instant nowP1 = now.plus(Duration.standardSeconds(1));
Instant nowP2 = now.plus(Duration.standardSeconds(2));
SlidingWindows windowFn = SlidingWindows.of(Duration.standardSeconds(5)).every(Duration.standardSeconds(1));
PCollection<KV<String, Integer>> res = p.apply(Create.timestamped(TimestampedValue.of("a", now), TimestampedValue.of("bb", nowP1), TimestampedValue.of("ccccc", nowP2))).apply(Window.into(windowFn)).apply(ParDo.of(pairStringWithIndexToLengthFn(bounded))).setCoder(KvCoder.of(StringUtf8Coder.of(), BigEndianIntegerCoder.of()));
assertEquals(windowFn, res.getWindowingStrategy().getWindowFn());
PCollection<TimestampedValue<KV<String, Integer>>> timestamped = res.apply(Reify.timestamps());
for (int i = 0; i < 4; ++i) {
Instant base = now.minus(Duration.standardSeconds(i));
IntervalWindow window = new IntervalWindow(base, base.plus(Duration.standardSeconds(5)));
List<TimestampedValue<KV<String, Integer>>> expectedUnfiltered = Arrays.asList(TimestampedValue.of(KV.of("a", 0), now), TimestampedValue.of(KV.of("bb", 0), nowP1), TimestampedValue.of(KV.of("bb", 1), nowP1), TimestampedValue.of(KV.of("ccccc", 0), nowP2), TimestampedValue.of(KV.of("ccccc", 1), nowP2), TimestampedValue.of(KV.of("ccccc", 2), nowP2), TimestampedValue.of(KV.of("ccccc", 3), nowP2), TimestampedValue.of(KV.of("ccccc", 4), nowP2));
List<TimestampedValue<KV<String, Integer>>> expected = new ArrayList<>();
for (TimestampedValue<KV<String, Integer>> tv : expectedUnfiltered) {
if (!window.start().isAfter(tv.getTimestamp()) && !tv.getTimestamp().isAfter(window.maxTimestamp())) {
expected.add(tv);
}
}
assertFalse(expected.isEmpty());
PAssert.that(timestamped).inWindow(window).containsInAnyOrder(expected);
}
p.run();
}
use of org.apache.beam.sdk.transforms.windowing.SlidingWindows in project beam by apache.
the class MultiStepCombineTest method testMultiStepCombineWindowed.
@Test
public void testMultiStepCombineWindowed() {
SlidingWindows windowFn = SlidingWindows.of(Duration.millis(6L)).every(Duration.millis(3L));
PCollection<KV<String, Long>> combined = pipeline.apply(Create.timestamped(TimestampedValue.of(KV.of("foo", 1L), new Instant(1L)), TimestampedValue.of(KV.of("bar", 2L), new Instant(2L)), TimestampedValue.of(KV.of("bizzle", 3L), new Instant(3L)), TimestampedValue.of(KV.of("bar", 4L), new Instant(4L)), TimestampedValue.of(KV.of("bizzle", 11L), new Instant(11L)))).apply(Window.into(windowFn)).apply(Combine.perKey(new MultiStepCombineFn()));
PAssert.that("Windows should combine only elements in their windows", combined).inWindow(new IntervalWindow(new Instant(0L), Duration.millis(6L))).containsInAnyOrder(KV.of("foo", 1L), KV.of("bar", 6L), KV.of("bizzle", 3L));
PAssert.that("Elements should appear in all the windows they are assigned to", combined).inWindow(new IntervalWindow(new Instant(-3L), Duration.millis(6L))).containsInAnyOrder(KV.of("foo", 1L), KV.of("bar", 2L));
PAssert.that(combined).inWindow(new IntervalWindow(new Instant(6L), Duration.millis(6L))).containsInAnyOrder(KV.of("bizzle", 11L));
PAssert.that(combined).containsInAnyOrder(KV.of("foo", 1L), KV.of("foo", 1L), KV.of("bar", 6L), KV.of("bar", 2L), KV.of("bar", 4L), KV.of("bizzle", 11L), KV.of("bizzle", 11L), KV.of("bizzle", 3L), KV.of("bizzle", 3L));
pipeline.run();
}
use of org.apache.beam.sdk.transforms.windowing.SlidingWindows in project beam by apache.
the class ParDoTest method testSideInputsWithMultipleWindows.
@Test
@Category(ValidatesRunner.class)
public void testSideInputsWithMultipleWindows() {
// Tests that the runner can safely run a DoFn that uses side inputs
// on an input where the element is in multiple windows. The complication is
// that side inputs are per-window, so the runner has to make sure
// to process each window individually.
MutableDateTime mutableNow = Instant.now().toMutableDateTime();
mutableNow.setMillisOfSecond(0);
Instant now = mutableNow.toInstant();
SlidingWindows windowFn = SlidingWindows.of(Duration.standardSeconds(5)).every(Duration.standardSeconds(1));
PCollectionView<Integer> view = pipeline.apply(Create.of(1)).apply(View.<Integer>asSingleton());
PCollection<String> res = pipeline.apply(Create.timestamped(TimestampedValue.of("a", now))).apply(Window.<String>into(windowFn)).apply(ParDo.of(new FnWithSideInputs(view)).withSideInputs(view));
for (int i = 0; i < 4; ++i) {
Instant base = now.minus(Duration.standardSeconds(i));
IntervalWindow window = new IntervalWindow(base, base.plus(Duration.standardSeconds(5)));
PAssert.that(res).inWindow(window).containsInAnyOrder("a:1");
}
pipeline.run();
}
use of org.apache.beam.sdk.transforms.windowing.SlidingWindows in project beam by apache.
the class ParDoTest method testTimerReceivedInOriginalWindow.
@Test
@Category({ ValidatesRunner.class, UsesTimersInParDo.class })
public void testTimerReceivedInOriginalWindow() throws Exception {
final String timerId = "foo";
DoFn<KV<String, Integer>, BoundedWindow> fn = new DoFn<KV<String, Integer>, BoundedWindow>() {
@TimerId(timerId)
private final TimerSpec spec = TimerSpecs.timer(TimeDomain.EVENT_TIME);
@ProcessElement
public void processElement(ProcessContext context, @TimerId(timerId) Timer timer) {
timer.offset(Duration.standardSeconds(1)).setRelative();
}
@OnTimer(timerId)
public void onTimer(OnTimerContext context, BoundedWindow window) {
context.output(context.window());
}
public TypeDescriptor<BoundedWindow> getOutputTypeDescriptor() {
return (TypeDescriptor) TypeDescriptor.of(IntervalWindow.class);
}
};
SlidingWindows windowing = SlidingWindows.of(Duration.standardMinutes(3)).every(Duration.standardMinutes(1));
PCollection<BoundedWindow> output = pipeline.apply(Create.timestamped(TimestampedValue.of(KV.of("hello", 24), new Instant(0L)))).apply(Window.<KV<String, Integer>>into(windowing)).apply(ParDo.of(fn));
PAssert.that(output).containsInAnyOrder(new IntervalWindow(new Instant(0), Duration.standardMinutes(3)), new IntervalWindow(new Instant(0).minus(Duration.standardMinutes(1)), Duration.standardMinutes(3)), new IntervalWindow(new Instant(0).minus(Duration.standardMinutes(2)), Duration.standardMinutes(3)));
pipeline.run();
}
Aggregations