Grafana Alloy

Telemetry agents and gateway for tenant-aware logs and OpenTelemetry signals.

Agentic Friendly

Component Category

Monitoring and debugging / telemetry collection

Component Description

Grafana Alloy is a telemetry collector that can gather, process, and forward logs, metrics, traces, and related signals.

Why It Is Used

In BullSequana AI Foundation, Alloy collects pod logs, applies tenant routing, and receives OpenTelemetry signals before sending them to the observability backends.

Learn More

Deployment notes

Alloy v1.18.1 runs per node, in the grafana-alloy namespace at sync wave 10 in the common tier, to collect pod standard output and error and send it to the Loki distributor. These infrastructure logs go to the platform scope; a tenant_id in the log body does not change their authorization boundary. Pods annotated with telemetry.bsqai.io/logs-exporter: otlp are excluded to avoid duplicate collection of tenant-aware logs.

A separate mTLS-authenticated Alloy gateway, the same Alloy version, runs in the alloy-gateway namespace at sync wave 10. It receives application telemetry over OTLP/gRPC and OTLP/HTTP. It remote-writes metrics to the platform Prometheus, which forwards them to Mimir through the tenant router. It routes logs to Loki by the server-authorized log-record attribute tenant_id, and traces to Tempo by each span's tenant.id, setting the organization of each write. Missing or malformed tenant context goes to quarantine. Exporter certificates and authorized query scopes are described in Observability and audit. KServe controller and model-workload metrics are scraped by this gateway; KServe is not configured as a general OTLP trace or log source.

Interacts With

  • Grafana Loki, which receives tenant-partitioned log streams collected by Alloy.
  • Grafana Tempo, which receives traces routed through the Alloy gateway.
  • Prometheus, which receives the gateway's metrics through remote write and scrapes Alloy's own metrics.
  • Grafana Mimir, which stores the gateway's tenant-attributed metrics after Prometheus forwards them.
  • Kubernetes workloads across the platform, because Alloy runs as a daemonset and collects node and pod-level telemetry.

On this page