Skip to main content

Recording modes

Continuous​

Every frame from the camera is written to disk the whole time it's active. This is the well-understood default, and the recommended starting point for a new camera while its motion detector hasn't been tuned to its specific scene yet.

Motion​

Motion-mode cameras buffer segments in a RAM cache (tmpfs, sized by MOTION_CACHE_TMPFS_BYTES) and persist to disk only when motion is actually detected: the buffered pre-roll leading up to the event, the event itself, and a configured post-roll afterward. Idle time between events never touches disk at all, that's the entire point of the mode.

Two safety rails make this reasonable to leave running unattended:

  • Fail-open. The instant a camera's motion detector becomes unhealthy, a stalled sub-stream, a dead decoder, anything that means Crumb can no longer form a trustworthy keep/discard decision, that camera immediately starts persisting everything to disk, exactly like Continuous mode, until detection is verified healthy again. A health alert fires for the duration. "I can't tell if this is interesting" always resolves to "record it all," never to "record nothing."
  • Spill. If the RAM cache nears its configured size, because of many cameras, a burst of concurrent motion, or a slow disk, the oldest buffered segments are persisted to disk rather than being dropped from the cache unwritten. Cache pressure can change when something gets written, never whether it survives.

Mode lives on the policy​

Recording mode is a property of the recording policy, not of the camera directly. Switching one camera on a shared policy to Motion does not flip its neighbors: Crumb splits that camera onto its own policy (auto-named after it) and changes the mode there. That is expected, not a bug, and it is why you will see a new policy appear after the change.

Shadow mode​

Shadow mode is a server-wide diagnostic, not a per-camera toggle. It is off by default and turned on with the MOTION_RECORDING_SHADOW environment variable in your compose file, which affects every Motion-mode camera at once. While it is on, those cameras keep recording and indexing every segment exactly like Continuous mode (nothing is discarded), but Crumb also runs the motion buffer's keep/discard decision in parallel and stamps the verdict on each segment. That lets you see what Motion mode would have thrown away before you trust it to actually throw anything away. There is no console switch for this in v0.1.0; it is an env-level opt-in you set once for the whole server.

Recommendation for a new camera​

Start a new camera on Continuous, or run the server with shadow mode on while your detectors are still being tuned against real footage from each camera's actual scene. Only switch a camera to live Motion mode after checking what would have been discarded. An untuned detector missing a real event is a worse outcome than the disk space Motion mode saves.

See Motion & Detection for tuning the detector itself.