docs: add usage guidance for downstream_sink_latency metric

Author: LeraKravts

src/current/observability/changefeeds/downstream_sink_latency.md

@@ -0,0 +1,32 @@
+---
+title: Monitor downstream_sink_latency in Changefeeds
+summary: How to interpret and monitor the downstream sink latency metric to detect bottlenecks in CockroachDB changefeeds.
+---
+
+#### Downstream sink acknowledgment latency
+
+**Metric**: `changefeed.stage.downstream_sink_latency`
+
+Measures the time it takes for a downstream sink (e.g., Kafka, cloud storage, webhook) to acknowledge messages. High values may indicate bottlenecks or backpressure at the end of the pipeline.
+
+**When it's high**:
+- Sink is slow to ACK; may cause backpressure or stall the changefeed.
+
+**Possible causes**:
+- Network latency between CockroachDB and the sink
+- Throttling or limits on the sink
+- Large payloads or inefficient encoding
+- Sink I/O capacity constraints (e.g., disk, Kafka throughput)
+
+**Where to monitor**:
+- CockroachDB Admin UI → *Metrics → Changefeeds*
+- Prometheus/Grafana if metrics are scraped
+
+**Suggested alert**:
+- For example: `changefeed.stage.downstream_sink_latency: p95 > 1s over 1m`, sustained
+
+**Mitigation strategies**:
+- Scale the sink or increase throughput
+- Enable batching or compression
+- Tune changefeed sink settings (partitioning, parallelism)
+- Diagnose with logs and sink system metrics