feature: Observability: Prometheus Metrics and Grafana
This commit is contained in:
@@ -0,0 +1,91 @@
|
||||
using Confluent.Kafka;
|
||||
using Confluent.Kafka.Admin;
|
||||
using Microsoft.Extensions.Options;
|
||||
|
||||
public sealed class KafkaConsumerLagCollector : BackgroundService
|
||||
{
|
||||
private static readonly string[] Groups =
|
||||
{
|
||||
"es-indexer",
|
||||
"sepsis-engine",
|
||||
"notification-publisher",
|
||||
"data-lake-writer",
|
||||
};
|
||||
|
||||
private readonly KafkaOptions _kafkaOptions;
|
||||
private readonly ClinicalMetrics _metrics;
|
||||
private readonly ILogger<KafkaConsumerLagCollector> _logger;
|
||||
|
||||
public KafkaConsumerLagCollector(
|
||||
IOptions<KafkaOptions> kafkaOptions,
|
||||
ClinicalMetrics metrics,
|
||||
ILogger<KafkaConsumerLagCollector> logger)
|
||||
{
|
||||
_kafkaOptions = kafkaOptions.Value;
|
||||
_metrics = metrics;
|
||||
_logger = logger;
|
||||
}
|
||||
|
||||
protected override async Task ExecuteAsync(CancellationToken ct)
|
||||
{
|
||||
// Initial delay so Kafka is reachable before the first collection.
|
||||
await Task.Delay(TimeSpan.FromSeconds(20), ct);
|
||||
|
||||
using var timer = new PeriodicTimer(TimeSpan.FromSeconds(30));
|
||||
while (await timer.WaitForNextTickAsync(ct))
|
||||
{
|
||||
foreach (var group in Groups)
|
||||
{
|
||||
try { await CollectGroupLagAsync(group, ct); }
|
||||
catch (Exception ex)
|
||||
{
|
||||
_logger.LogDebug(ex, "KafkaConsumerLagCollector: failed for group {Group}", group);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private async Task CollectGroupLagAsync(string groupId, CancellationToken ct)
|
||||
{
|
||||
var adminConfig = new AdminClientConfig
|
||||
{ BootstrapServers = _kafkaOptions.BootstrapServers };
|
||||
|
||||
using var admin = new AdminClientBuilder(adminConfig).Build();
|
||||
|
||||
// List the committed offsets for this consumer group across all its partitions.
|
||||
var result = await admin.ListConsumerGroupOffsetsAsync(
|
||||
new[] { new ConsumerGroupTopicPartitions(groupId, null) },
|
||||
new ListConsumerGroupOffsetsOptions { RequireStableOffsets = false });
|
||||
|
||||
var partitions = result.FirstOrDefault()?.Partitions ?? [];
|
||||
if (!partitions.Any())
|
||||
{
|
||||
_metrics.KafkaConsumerLag.WithLabels(groupId).Set(0);
|
||||
return;
|
||||
}
|
||||
|
||||
// Query high watermarks using a temporary consumer (does not join the group).
|
||||
using var tempConsumer = new ConsumerBuilder<Ignore, Ignore>(new ConsumerConfig
|
||||
{
|
||||
BootstrapServers = _kafkaOptions.BootstrapServers,
|
||||
GroupId = $"__lag-probe",
|
||||
}).Build();
|
||||
|
||||
long totalLag = 0;
|
||||
foreach (var tpo in partitions)
|
||||
{
|
||||
if (tpo.Error.IsError || tpo.Offset == Offset.Unset) continue;
|
||||
|
||||
var watermarks = tempConsumer.QueryWatermarkOffsets(
|
||||
tpo.TopicPartition, TimeSpan.FromSeconds(5));
|
||||
|
||||
// Committed offset is the next offset the consumer will read.
|
||||
// High watermark is the latest available offset + 1.
|
||||
// Lag = messages the consumer has not yet read.
|
||||
var lag = watermarks.High.Value - tpo.Offset.Value;
|
||||
totalLag += Math.Max(0L, lag);
|
||||
}
|
||||
|
||||
_metrics.KafkaConsumerLag.WithLabels(groupId).Set(totalLag);
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user