Skip to content

When kafka consumer metric is NaN, report previous value instead of invalid value #6741

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
graytaylor0 merged 1 commit into from
Apr 8, 2026
+119 −0
Merged

When kafka consumer metric is NaN, report previous value instead of invalid value #6741

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions ...rc/main/java/org/opensearch/dataprepper/plugins/kafka/util/KafkaTopicConsumerMetrics.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -221,6 +221,10 @@ public void update(final KafkaConsumer consumer) {
continue;
}
double newValue = (Double)value.metricValue();
if (Double.isNaN(newValue) || Double.isInfinite(newValue)) {
LOG.debug("Skipping non-finite metric value {} for {}", newValue, metricName);
continue;
}
if (metricName.equals("records-consumed-total")) {
synchronized(consumerMetrics) {
double prevValue = consumerMetrics.get(metricName);
Expand Down
115 changes: 115 additions & 0 deletions ...st/java/org/opensearch/dataprepper/plugins/kafka/util/KafkaTopicConsumerMetricsTests.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,6 +259,121 @@ public void KafkaTopicMetricTest_checkMetricUpdates(int numConsumers) {

}

@Test
public void update_withNaNMetricValue_preservesPreviousValue() {
doAnswer((i) -> {
bytesConsumedCount += (double)i.getArgument(0);
return null;
}).when(bytesConsumedCounter).increment(any(Double.class));
doAnswer((i) -> {
recordsConsumedCount += (double)i.getArgument(0);
return null;
}).when(recordsConsumedCounter).increment(any(Double.class));

topicMetrics = createObjectUnderTest();
KafkaConsumer kafkaConsumer = mock(KafkaConsumer.class);
topicMetrics.register(kafkaConsumer);

// First update with real values
Map<MetricName, KafkaTestMetric> metrics = new HashMap<>();
when(kafkaConsumer.metrics()).thenReturn(metrics);
double expectedLag = 500.0;
double expectedLead = 100.0;
Map<String, String> emptyTags = new HashMap<>();
Map<String, String> topicTags = new HashMap<>();
topicTags.put("topic", topicName);
metrics.put(getMetric("records-lag-max", expectedLag, emptyTags).metricName(),
getMetric("records-lag-max", expectedLag, emptyTags));
metrics.put(getMetric("records-lead-min", expectedLead, emptyTags).metricName(),
getMetric("records-lead-min", expectedLead, emptyTags));
metrics.put(getMetric("commit-rate", 5.0, emptyTags).metricName(),
getMetric("commit-rate", 5.0, emptyTags));
metrics.put(getMetric("bytes-consumed-total", 100.0, topicTags).metricName(),
getMetric("bytes-consumed-total", 100.0, topicTags));
metrics.put(getMetric("records-consumed-total", 10.0, topicTags).metricName(),
getMetric("records-consumed-total", 10.0, topicTags));
topicMetrics.update(kafkaConsumer);

// Second update with NaN values (simulating Kafka 3.x idle consumer)
Map<MetricName, KafkaTestMetric> nanMetrics = new HashMap<>();
when(kafkaConsumer.metrics()).thenReturn(nanMetrics);
nanMetrics.put(getMetric("records-lag-max", Double.NaN, emptyTags).metricName(),
getMetric("records-lag-max", Double.NaN, emptyTags));
nanMetrics.put(getMetric("records-lead-min", Double.NaN, emptyTags).metricName(),
getMetric("records-lead-min", Double.NaN, emptyTags));
nanMetrics.put(getMetric("commit-rate", Double.NaN, emptyTags).metricName(),
getMetric("commit-rate", Double.NaN, emptyTags));
nanMetrics.put(getMetric("bytes-consumed-total", Double.NaN, topicTags).metricName(),
getMetric("bytes-consumed-total", Double.NaN, topicTags));
nanMetrics.put(getMetric("records-consumed-total", Double.NaN, topicTags).metricName(),
getMetric("records-consumed-total", Double.NaN, topicTags));
topicMetrics.update(kafkaConsumer);

// Verify gauges still return valid previous values, not NaN
Map<KafkaConsumer, Map<String, Double>> metricValuesMap = topicMetrics.getMetricValues();
Map<String, Double> consumerMetrics = metricValuesMap.get(kafkaConsumer);
assertEquals(expectedLag, consumerMetrics.get("records-lag-max"), 0.01d);
assertEquals(expectedLead, consumerMetrics.get("records-lead-min"), 0.01d);
assertEquals(5.0, consumerMetrics.get("commit-rate"), 0.01d);

// Verify gauge callbacks produce valid numbers
ToDoubleFunction lagGauge = pluginMetricsMap.get("topic." + topicName + ".recordsLagMax");
double lagResult = lagGauge.applyAsDouble(metricValuesMap);
assertEquals(expectedLag, lagResult, 0.01d);

ToDoubleFunction leadGauge = pluginMetricsMap.get("topic." + topicName + ".recordsLeadMin");
double leadResult = leadGauge.applyAsDouble(metricValuesMap);
assertEquals(expectedLead, leadResult, 0.01d);
}

@Test
public void update_withNegativeInfinityMetricValue_preservesPreviousValue() {
topicMetrics = createObjectUnderTest();
KafkaConsumer kafkaConsumer = mock(KafkaConsumer.class);
topicMetrics.register(kafkaConsumer);

// First update with real values
Map<MetricName, KafkaTestMetric> metrics = new HashMap<>();
when(kafkaConsumer.metrics()).thenReturn(metrics);
double expectedLag = 250.0;
Map<String, String> emptyTags = new HashMap<>();
metrics.put(getMetric("records-lag-max", expectedLag, emptyTags).metricName(),
getMetric("records-lag-max", expectedLag, emptyTags));
topicMetrics.update(kafkaConsumer);

// Second update with -Infinity (Kafka Max SampledStat empty window)
Map<MetricName, KafkaTestMetric> infMetrics = new HashMap<>();
when(kafkaConsumer.metrics()).thenReturn(infMetrics);
infMetrics.put(getMetric("records-lag-max", Double.NEGATIVE_INFINITY, emptyTags).metricName(),
getMetric("records-lag-max", Double.NEGATIVE_INFINITY, emptyTags));
topicMetrics.update(kafkaConsumer);

Map<String, Double> consumerMetrics = topicMetrics.getMetricValues().get(kafkaConsumer);
assertEquals(expectedLag, consumerMetrics.get("records-lag-max"), 0.01d);
}

@Test
public void update_withNaNMetricValue_andNopriorUpdate_preservesInitialZero() {
topicMetrics = createObjectUnderTest();
KafkaConsumer kafkaConsumer = mock(KafkaConsumer.class);
topicMetrics.register(kafkaConsumer);

// First update is NaN (consumer never received any records)
Map<MetricName, KafkaTestMetric> nanMetrics = new HashMap<>();
when(kafkaConsumer.metrics()).thenReturn(nanMetrics);
Map<String, String> emptyTags = new HashMap<>();
nanMetrics.put(getMetric("records-lag-max", Double.NaN, emptyTags).metricName(),
getMetric("records-lag-max", Double.NaN, emptyTags));
topicMetrics.update(kafkaConsumer);

// Should preserve the initial 0.0 from register()
Map<String, Double> consumerMetrics = topicMetrics.getMetricValues().get(kafkaConsumer);
assertEquals(0.0, consumerMetrics.get("records-lag-max"), 0.01d);

ToDoubleFunction lagGauge = pluginMetricsMap.get("topic." + topicName + ".recordsLagMax");
double lagResult = lagGauge.applyAsDouble(topicMetrics.getMetricValues());
assertEquals(0.0, lagResult, 0.01d);
}

@Test
void recordTimeBetweenPolls_records_metric_correctly() throws InterruptedException {
Expand Down
Loading