Expose max timestamp for each topic and/or partition (#3)

* Expose max timestamp for each topic and/or partition


Kafka 3.0+ added support to fetch offset with the max timestamp.
This is useful in monitoring the steady progress of producing to a topic/partion
but also to detect very high timestamps in the topic that can stuck topic retention.

This patch adds two new metrics:
- kafka_topic_partition_max_timestamp: for each topic partition reports the max timestamp in that partition
- kafka_topic_max_timestamp: for each topic reports the max timestamp in that topic, computed as the maximum timestamp across partitions


See: 
- https://issues.apache.org/jira/browse/KAFKA-12541
- https://cwiki.apache.org/confluence/display/KAFKA/KIP-734%3A+Improve+AdminClient.listOffsets+to+return+timestamp+and+offset+for+the+record+with+the+largest+timestamp

* Update prometheus/collect_topic_partition_offsets.go

Author: amuraru

prometheus/collect_topic_partition_offsets.go

@@ -30,6 +30,15 @@ func (e *Exporter) collectTopicPartitionOffsets(ctx context.Context, ch chan<- p
 		return false
 	}
 
+	// Highest Timestamp Offsets
+	// NB: this requires Kafka Brokers 3.0+ (see https://issues.apache.org/jira/browse/KAFKA-12541)
+	// In older versions this is returning the timestamp of the low watermarks (earliest offset)
+	maxTimestampOffsets, err := e.minionSvc.ListOffsetsCached(ctx, -3)
+	if err != nil {
+		e.logger.Error("failed to fetch offsets for max timestamp", zap.Error(err))
+		return false
+	}
+
 	// Process Low Watermarks
 
 	for topicName, partitions := range lowWaterMarks {
@@ -105,5 +114,47 @@ func (e *Exporter) collectTopicPartitionOffsets(ctx context.Context, ch chan<- p
 		}
 	}
 
+	// Process Max Timestamps
+	for _, topic := range maxTimestampOffsets.Topics {
+		if !e.minionSvc.IsTopicAllowed(topic.Topic) {
+			continue
+		}
+		topicMaxTimestamp := int64(0)
+		hasErrors := false
+		for _, partition := range topic.Partitions {
+			err := kerr.ErrorForCode(partition.ErrorCode)
+			if err != nil {
+				hasErrors = true
+				isOk = false
+				continue
+			}
+			if topicMaxTimestamp < partition.Timestamp {
+				topicMaxTimestamp = partition.Timestamp
+			}
+			// Let's end here if partition metrics shall not be exposed
+			if e.minionSvc.Cfg.Topics.Granularity == minion.TopicGranularityTopic {
+				continue
+			}
+			if partition.Timestamp > 0 {
+				ch <- prometheus.MustNewConstMetric(
+					e.partitionMaxTimestamp,
+					prometheus.GaugeValue,
+					float64(partition.Timestamp),
+					topic.Topic,
+					strconv.Itoa(int(partition.Partition)),
+				)
+			}
+		}
+		// We only want to report the max of all partition max timestamps if we receive results from all partitions
+		// and the topic is not empty
+		if !hasErrors && topicMaxTimestamp > 0 {
+			ch <- prometheus.MustNewConstMetric(
+				e.topicMaxTimestamp,
+				prometheus.GaugeValue,
+				float64(topicMaxTimestamp),
+				topic.Topic,
+			)
+		}
+	}
 	return isOk
 }

prometheus/exporter.go

@@ -41,6 +41,8 @@ type Exporter struct {
 	partitionHighWaterMark     *prometheus.Desc
 	topicLowWaterMarkSum       *prometheus.Desc
 	partitionLowWaterMark      *prometheus.Desc
+	topicMaxTimestamp          *prometheus.Desc
+	partitionMaxTimestamp      *prometheus.Desc
 
 	// Consumer Groups
 	consumerGroupInfo                    *prometheus.Desc
@@ -172,6 +174,20 @@ func (e *Exporter) InitializeMetrics() {
 		[]string{"topic_name"},
 		nil,
 	)
+	// Partition Max Timestamp
+	e.partitionMaxTimestamp = prometheus.NewDesc(
+		prometheus.BuildFQName(e.cfg.Namespace, "kafka", "topic_partition_max_timestamp"),
+		"Partition Max Timestamp",
+		[]string{"topic_name", "partition_id"},
+		nil,
+	)
+	// Topic Max Timestamp
+	e.topicMaxTimestamp = prometheus.NewDesc(
+		prometheus.BuildFQName(e.cfg.Namespace, "kafka", "topic_max_timestamp"),
+		"Topic Max Timestamp",
+		[]string{"topic_name"},
+		nil,
+	)
 
 	// Consumer Group Metrics
 	// Group Info