Skip to main content

CRE-2025-0033

OpenTelemetry Collector refuses to scrape due to memory pressureLow
Impact: 7/10
Mitigation: 4/10

CRE-2025-0033View on GitHub

Description

The OpenTelemetry Collector may refuse to ingest metrics during a Prometheus scrape if it exceeds its configured memory limits. When the `memory_limiter` processor is enabled, the Collector actively drops data to prevent out-of-memory errors, resulting in log messages indicating that data was refused due to high memory usage.


Cause

When the Collector''s memory usage surpasses configured soft or hard thresholds, the memory_limiter processor prevents ingestion of new telemetry data. This safeguards the process from crashing but leads to temporary loss of incoming metrics.


Mitigation

  • Tune the memory_limiter configuration using `limit_mib` and `check_interval`.
  • Increase resource limits for the Collector container (e.g., via Kubernetes).
  • Reduce incoming data volume or adjust scrape intervals to lower memory pressure.
  • Monitor Collector memory usage over time and adjust pipeline design accordingly.

References