Log rotation and retention limits

What it is

Rotation caps how much log data a layer keeps: the container runtime's json-file driver takes max-size and max-file, system journals have size caps, proxies rotate daily or by size, and log collectors apply retention policies. Retention is the deliberate choice of how long records are kept.

Why it matters

An unbounded log fills the disk and takes the service down with it; an over-long retention keeps personal data (addresses, URLs) longer than justified. Both are common incident causes.

How to apply

• Set size and count limits on every container's logging driver (for example 5 MB × 3) and on the system journal.
• Choose retention per log type: application logs a few days to weeks, access logs per policy, audit logs longer and separately protected.
• Ship logs you need to keep to a collector with its own retention; do not use local rotation as an archive.
• Document the policy where operators and privacy notices can reference it.
• Test what happens when the disk is full anyway: the application must keep serving or fail cleanly.

Pitfalls

Rotation configured on the host but not inside containers, or the reverse. Collectors that duplicate retention. Logs that contain secrets are a retention problem no matter how short.