Docker logs ate 2.4 GB on my server. Here's how to cap them

While checking my server, I measured how much space container logs were using:

docker run --rm -v /var/lib/docker/containers:/c:ro alpine du -sh /c
2.4G    /c

One container alone had 1.1 GB of logs. That was the Zigbee bridge that had been stuck in a crash loop, writing its startup messages every 14 seconds for three days.

(The alpine trick is so I can read the root-owned Docker folder without sudo. Mounting it read-only is harmless.)

Why it happens

Docker’s default json-file log driver has no size limit. Every line a container prints goes into a file under /var/lib/docker/containers/<id>/ and stays there until the container is removed. A chatty container or a crash loop can fill a disk.

The fix: global log options

Edit (or create) /etc/docker/daemon.json and add log-opts. Keep whatever is already there. Mine already had the NVIDIA runtime and a shutdown timeout:

{
    "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
        }
    },
    "shutdown-timeout": 60,
    "log-driver": "json-file",
    "log-opts": {
        "max-size": "20m",
        "max-file": "3"
    }
}

That’s at most 3 files of 20 MB per container. Docker rotates them automatically. With ~30 containers, the worst case is about 2 GB, and in practice it’s far less.

Validate the JSON before restarting. A typo stops Docker from starting at all:

python3 -c 'import json; json.load(open("/etc/docker/daemon.json")); print("ok")'
sudo systemctl restart docker

The catch: existing containers keep their old settings

Log options are fixed when a container is created. Restarting Docker isn’t enough. You have to recreate the containers. For all running Compose stacks:

for d in $(docker ps --format '{{.Label "com.docker.compose.project.working_dir"}}' | sort -u); do
  (cd "$d" && docker compose up -d --force-recreate)
done

This only touches stacks that are currently running, so stopped projects stay stopped. Data in volumes and bind mounts survives, and the old log files are deleted along with the old containers.

Check it worked

docker inspect <container> --format '{{.HostConfig.LogConfig}}'
# {json-file map[max-file:3 max-size:20m]}

After recreating everything, my log folder went from 2.4 GB to under 2 MB.

Per-service override

If one container needs more history, set it in its Compose file:

services:
  homeassistant:
    logging:
      driver: json-file
      options:
        max-size: "50m"
        max-file: "5"

Do this on day one of any Docker host. It’s two lines, and it saves a full disk later.

FAQ

Will I lose logs I need? You keep the newest ~60 MB per container. For a home server that’s days or weeks of normal output. If you need long-term history (for auditing, say), send logs to a proper log system instead of keeping them in Docker’s files.

Does docker compose logs still work? Yes. Rotation is invisible to docker logs; you just can’t scroll back past the oldest kept file.

What about the local log driver? Docker’s local driver compresses logs and rotates by default, which is another good option. I stayed with json-file because some monitoring tools read those files directly.

Do I need to repeat this after updating images? No. Pulling a new image and recreating a container (docker compose up -d) creates it with the daemon defaults, so the limit sticks.