Skip to content
Maple Docs
Open app
Browse · WarpStream
On this page

WarpStream

Monitor WarpStream clusters in Maple. Scrape Agent /metrics endpoints directly, or pull consumer lag, request latency, and object-store health from WarpStream's hosted Prometheus endpoint.

WarpStream exposes Prometheus metrics in two places. Every Agent serves a local /metrics endpoint, and the WarpStream control plane offers a hosted, authenticated Prometheus endpoint per virtual cluster. Both work with Maple’s Prometheus scraping. Pick the one that matches your network topology.

Prerequisites

  • A WarpStream virtual cluster, and access to create an Agent Key for it.
  • For Option B: a network path from the internet to the Agents’ metrics port, or an OpenTelemetry Collector next to them.

The control plane serves cluster health (consumer group lag, partition sizes, agent heartbeats) at a single internet-reachable URL. Maple can scrape it without network changes:

https://api.warpstream.com/api/v1/monitoring/prometheus/virtual_clusters/<id>

Replace <id> with your virtual cluster ID (vci_…).

  1. In WarpStream, create a read-only Agent Key for the cluster (least privilege; an account-level API key also works if you want one key for several clusters).

  2. In Maple, open Sources, click Connect source, pick WarpStream (or Prometheus), and configure:

    • Endpoint URL: the hosted endpoint above, with your virtual cluster ID
    • Authentication: Basic Auth, with Username prometheus and Password set to the API key
    • Scrape interval (seconds): 30 to 60 is enough for control-plane metrics

    To tag every series with a cluster name, set labels_json (for example {"cluster": "prod-kafka"}) through the scrape targets API.

See WarpStream’s Hosted Prometheus Endpoint docs for the metric set, which includes Tableflow and Schema Registry metrics on those cluster types.

Option B: Scrape the Agents directly

Each WarpStream Agent serves the full agent-level metric set (request latency histograms, produce/fetch byte counters, object-store operation latency) on its internal port. The endpoint is enabled by default and has no authentication:

http://$AGENT_IP:8080/metrics

Because Agents run inside your VPC under WarpStream’s BYOC model, Maple’s scraper can only reach them if you expose the endpoint (for example through an internal load balancer with auth, which you can pair with Maple’s Bearer Token or Basic Auth options). If the Agents are fully private, run an OpenTelemetry Collector next to them with a prometheus receiver scraping :8080/metrics and an OTLP exporter pointed at Maple’s ingest endpoint instead.

Add one Maple target per agent (or per load-balanced agent pool), with the agent host in Endpoint URL. Every series already carries an instance attribute with the target host. To add cluster or agent names, set labels_json such as {"cluster": "...", "agent": "..."} through the API.

Metrics worth alerting on

All WarpStream metrics carry the warpstream_ prefix. From WarpStream’s Important Metrics and Logs:

MetricWhy it matters
warpstream_consumer_group_lagConsumer lag in offsets. The main “are we keeping up” gauge.
warpstream_agent_kafka_request_latencyProduce/fetch latency histogram by request type.
warpstream_agent_kafka_request_outcomeSuccess vs. error counters per Kafka request type.
warpstream_blob_store_operation_latencyObject-store PUT/GET health. Check this first when latency spikes.
warpstream_agent_control_plane_operation_latencyAgent ↔ control-plane RPC health.
warpstream_topics_count / warpstream_partitions_count (+ _limit)Headroom against cluster limits.

Once samples arrive, build dashboards and alert rules on these like any other Maple metric. For example, alert when warpstream_consumer_group_lag stays above a threshold for 5 minutes.

Verify

  1. On the source page, click Sync now. Maple runs a scrape and reports success or the exact failure.
  2. Wait one scrape interval. The status changes from No checks to Up.
  3. Search for warpstream_ in the metrics explorer.

Troubleshooting

  • 401 from the hosted endpoint. The Basic Auth username must be exactly prometheus. The password is the API or Agent key.
  • Timeouts scraping Agents. Confirm the Agent’s internal port (8080 by default) and that the endpoint is reachable from the internet. curl $AGENT_IP:8080/v1/status should return OK.
  • Any other scrape error. The connect step checks the endpoint when you save and shows the exact upstream error. Later, Sync now on the source page runs the same check.

Next steps

  • Prometheus scraping: every scrape target field and how the data is stored.
  • Alert rules: alert on warpstream_consumer_group_lag and the other metrics above.
  • Metrics: chart the scraped series.