Crash-consistent snapshots of storage volumes are a required feature in today’s storage systems. In Kubernetes, the CSI protocol introduced the VolumeGroupSnapshot concept which requires crash-consistent snapshots be guaranteed within the context of volumes, which may be block devices or mounts on a distributed POSIX file system such as CephFS. This talk will cover the challenges and mechanisms to achieve this design goal. In particular, we will show how CephFS provides consistency by exposing a new API to pause write IOs to a set of volumes for the duration of the snapshot process. We will explore the new API, the mechanisms used to achieve this IO pause and discuss broadly the software changes made to Ceph.