[BEAM-3806] DirectRunner hangs if multiple timers set in the same bundle

• Type: Bug
• Status: Resolved
• Priority: P2
• Resolution: Fixed
• Fix Version/s: 2.5.0
• Component/s: runner-direct
• Assignee: Thomas Groh
• Reporter: Ben Chambers
• Created: 08/Mar/18 01:11
• Updated: 16/May/20 13:55
• Resolved: 10/Mar/18 01:09

Description

See the repro below:

package com.simbly.beam.cassandra;

import org.apache.beam.sdk.coders.KvCoder;
import org.apache.beam.sdk.coders.StringUtf8Coder;
import org.apache.beam.sdk.state.TimeDomain;
import org.apache.beam.sdk.state.Timer;
import org.apache.beam.sdk.state.TimerSpec;
import org.apache.beam.sdk.state.TimerSpecs;
import org.apache.beam.sdk.testing.PAssert;
import org.apache.beam.sdk.testing.TestPipeline;
import org.apache.beam.sdk.testing.TestStream;
import org.apache.beam.sdk.transforms.DoFn;
import org.apache.beam.sdk.transforms.ParDo;
import org.apache.beam.sdk.values.KV;
import org.apache.beam.sdk.values.PCollection;
import org.joda.time.Duration;
import org.junit.Rule;
import org.junit.Test;

public class DirectRunnerTest {

  @Rule
  public TestPipeline pipeline = TestPipeline.create();

  @Test
  public void badTimerBehavior() {
    TestStream<KV<String, String>> stream = TestStream
        .create(KvCoder.of(StringUtf8Coder.of(), StringUtf8Coder.of()))
        .addElements(KV.of("key1", "v1"))
        .advanceWatermarkToInfinity();

    PCollection<String> result = pipeline
        .apply(stream)
        .apply(ParDo.of(new TestDoFn()));
    PAssert.that(result).containsInAnyOrder("It works");

    pipeline.run().waitUntilFinish();
  }

  private static class TestDoFn extends DoFn<KV<String, String>, String> {
    @TimerId("timer")
    private final TimerSpec timer = TimerSpecs.timer(TimeDomain.EVENT_TIME);

    @ProcessElement
    public void process(ProcessContext c,
        @TimerId("timer") Timer timer) {
      timer.offset(Duration.standardMinutes(10)).setRelative();
      timer.offset(Duration.standardMinutes(30)).setRelative();
    }

    @OnTimer("timer")
    public void onTimer(OnTimerContext c, @TimerId("timer") Timer timer) {
      c.output("It works");
    }
  }
}

From inspection, this seems to be caused by the logic in WatermarkManager, which does the following if there are multiple timers for akey:

1. Adds the first timer to the `pendingTimers`, `keyTimers`, and `existingTimersForKey`.
2. Removes the first timer from `keyTimers`
3. Adds the second timer to `keyTimers` and `existingTimersForKey`.

This leads to inconsistencies since pendingTimers has only the first timer, keyTimers only the second, and existingTimers has both. This becomes more problematic since one of these lists is used for firing (and thus releasing holds) and the other is used for holds.

Issue Links

• links to: GitHub Pull Request #4829