OpenStack Cinder Volume Detachment Delays with NetApp ONTAP Backends – Eventlet Reactor Blocked by Performance/Dedupe Stats Polling
Applies to
- OpenStack Cinder with NetApp ONTAP driver
- Multi-backend Cinder configurations (multiple SVMs/pools per cinder-volume process)
- Environments using Platform9 or upstream OpenStack Cinder
- All Flash FAS (AFF) platforms
Issue
- Volume detach operations in OpenStack Cinder take 7 minutes or longer, remaining stuck in “Detaching” state.
- Manual detach/attach operations complete quickly (3–4 seconds) outside of incident windows.
- Cinder scheduler logs show:
No capability reports from this host between [timestamp gap] — this is what stalled detach for ~6 minutes. Ignoring old capability report from ... at [timestamp] — three seconds before detach timeout.
- cinder-volume logs contain:
report_state ... outlasted interval warnings (e.g., overrun at incident time)
- All NetApp pools/backends served by a single cinder-volume process (single eventlet reactor).
