Do not emit negative lag because of stale offsets (#14292)

The latest topic offsets are polled frequently and used to determine the lag based on the current offsets. However, when the offsets are stale (which can happen due to connection issues commonly), we may see a negative lag .

This PR prevents emission of metrics when the offsets are stale and at least one of the partitions has a negative lag.
This commit is contained in:
AmatyaAvadhanula 2023-07-05 14:44:23 +05:30 committed by GitHub
parent cc159f4317
commit 609833c97b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 40 additions and 0 deletions

View File

@ -4220,6 +4220,21 @@ public abstract class SeekableStreamSupervisor<PartitionIdType, SequenceOffsetTy
return; return;
} }
// Try emitting lag even with stale metrics provided that none of the partitions has negative lag
final long staleMillis = sequenceLastUpdated == null
? 0
: DateTimes.nowUtc().getMillis()
- (tuningConfig.getOffsetFetchPeriod().getMillis() + sequenceLastUpdated.getMillis());
if (staleMillis > 0 && partitionLags.values().stream().anyMatch(x -> x < 0)) {
// Log at most once every twenty supervisor runs to reduce noise in the logs
if ((staleMillis / getIoConfig().getPeriod().getMillis()) % 20 == 0) {
log.warn("Lag is negative and will not be emitted because topic offsets have become stale. "
+ "This will not impact data processing. "
+ "Offsets may become stale because of connectivity issues.");
}
return;
}
LagStats lagStats = computeLags(partitionLags); LagStats lagStats = computeLags(partitionLags);
Map<String, Object> metricTags = spec.getContextValue(DruidMetrics.TAGS); Map<String, Object> metricTags = spec.getContextValue(DruidMetrics.TAGS);
for (Map.Entry<PartitionIdType, Long> entry : partitionLags.entrySet()) { for (Map.Entry<PartitionIdType, Long> entry : partitionLags.entrySet()) {

View File

@ -62,6 +62,7 @@ import org.apache.druid.indexing.seekablestream.common.StreamPartition;
import org.apache.druid.indexing.seekablestream.supervisor.SeekableStreamSupervisorStateManager.SeekableStreamExceptionEvent; import org.apache.druid.indexing.seekablestream.supervisor.SeekableStreamSupervisorStateManager.SeekableStreamExceptionEvent;
import org.apache.druid.indexing.seekablestream.supervisor.SeekableStreamSupervisorStateManager.SeekableStreamState; import org.apache.druid.indexing.seekablestream.supervisor.SeekableStreamSupervisorStateManager.SeekableStreamState;
import org.apache.druid.indexing.seekablestream.supervisor.autoscaler.AutoScalerConfig; import org.apache.druid.indexing.seekablestream.supervisor.autoscaler.AutoScalerConfig;
import org.apache.druid.java.util.common.DateTimes;
import org.apache.druid.java.util.common.ISE; import org.apache.druid.java.util.common.ISE;
import org.apache.druid.java.util.common.StringUtils; import org.apache.druid.java.util.common.StringUtils;
import org.apache.druid.java.util.common.granularity.Granularities; import org.apache.druid.java.util.common.granularity.Granularities;
@ -1035,6 +1036,30 @@ public class SeekableStreamSupervisorStateTest extends EasyMockSupport
); );
} }
@Test
public void testStaleOffsetsNegativeLagNotEmitted() throws Exception
{
expectEmitterSupervisor(false);
CountDownLatch latch = new CountDownLatch(1);
final TestEmittingTestSeekableStreamSupervisor supervisor = new TestEmittingTestSeekableStreamSupervisor(
latch,
TestEmittingTestSeekableStreamSupervisor.LAG,
// Record lag must not be emitted
ImmutableMap.of("0", 10L, "1", -100L),
null
);
supervisor.start();
// Forcibly set the offsets to be stale
supervisor.sequenceLastUpdated = DateTimes.nowUtc().minus(Integer.MAX_VALUE);
latch.await();
supervisor.emitLag();
Assert.assertEquals(0, emitter.getEvents().size());
}
private List<Event> filterMetrics(List<Event> events, List<String> whitelist) private List<Event> filterMetrics(List<Event> events, List<String> whitelist)
{ {
List<Event> result = events.stream() List<Event> result = events.stream()