Storage notes

Give application logs a storage budget

Useful diagnostics need bounded retention and enough free space for the next recovery operation.

Decide which evidence earns its space

A camera application can generate images, per-frame events, model outputs and detailed runtime logs. Keeping everything indefinitely turns storage capacity into an accidental timer. Begin with the questions a maintainer must answer after failure, then assign retention to the evidence needed for those questions. Preserve compact identifiers, state changes and error context before choosing continuous verbose output. The goal is a useful account of behaviour, with a defined maximum footprint and a way to notice when evidence is being discarded. Keep a count of dropped records or expired files where practical. That lets a maintainer distinguish an uneventful interval from an interval whose detailed evidence no longer exists on the device.

Measure the competing workload

Logging shares storage with model loading, updates and sometimes application data. If those jobs interfere, reproduce their overlap before changing the storage device. fio provides explicit access-pattern and reporting controls that can help investigate a suspected I/O bottleneck, but the application workload remains the acceptance check. Keep diagnostic files separate from irreplaceable data during testing. Record log growth over a representative operating interval, including a failure burst; quiet idle behaviour alone is a poor estimate of the space a repeated error can consume.

Reserve space for recovery

An update may need both the active release and its candidate, while a diagnostic capture may need room precisely when the device is misbehaving. Put that reserve into the storage plan. Define what rotates first, what must be retained and how a low-space state changes application behaviour. The ext4 journal’s metadata protections do not create a backup of application records, so unique data still needs its own retention or transfer policy. Review the budget after adding a new model or diagnostic mode. A bounded system should remain maintainable at the end of its retention window, not merely fit on the day it is installed.

Sources & publication notes

Retrospective period: May 2026. Written and published November 2, 2026. Product announcements establish release timing; support pages are linked as current references.

Related reading

← More guides & analysisCompare hardware →