Dozzle will tell you just about everything you want to know about the health of a container. Sadly, to my knowledge, it does not integrate with any notification platforms like nfty, even though there is a long standing request for that feature.

I just put a healthcheck in my compose files and then run an autoheal container that will automatically restart them if they are “unhealthy”.

1. I don’t, in general all of them are restarted automatically. I have monitoring configured for services, not containers themselves.
2. Yes, looking at the original Dockerfile/Containerfile if they have HEALTHCHECK keyword you can assume that they do something.
3. Person building the container, often it doesn’t make sense to create a healthcheck at all. Some times healthcheck feature is provided by application as well, it still needs to be part of the containerfile.
4. Better to monitor your service (application+db+proxy+queue+whatever), not containers in isolation.

docker inspect --format='{{json .State.Health}}' <container_name>

HEALTHCHECK is part of the Dockerfile syntax and ought to be supported by all your container runtimes

https://docs.docker.com/reference/dockerfile/#healthcheck

You could extend all the dockerfiles that don’t have a health check to implement this feature with whatever health check makes sense for the application, even if for now it’s just a curl of an endpoint.

You can read the Dockerfile for the HEALTHCHECK clause however not all have it as it’s been introduced in later docker versions.

You can also write your own using things like curl.

2, no, just check the docs.

3, yup

You can make your own health checks in docker compose, so for instance, I had etcd running provided by another company, and I just set up the check in compose using the etcdctl commands (etcdctl endpoint health).

https://docs.docker.com/reference/compose-file/services/#healthcheck

I use uptimekuma with notifications through home assistant. I get notifications on my phone and watch. I had notifications set up to go to a room on my matrix homeserver but recently migrated it and don’t feel like messing with the room.

A superb image will have a health check endpoint set up in the dockerfile.

A good image will have a health check endpoint on either the service or another port that you can set up manually.

Most images will require you to manually devise some convoluted health check procedure using automated auth tokens.

All of my images fall into that latter category. You’re welcome.

(Ok, ok, I’m sorry. But you did just remind me that I need to code a health check endpoint and put it in the dockerfile.)

1. Some kind of monitoring software, like the Grafana stack. I like email and Discord notifications.
2. The Dockerfile will have a HEALTHCHECK statement, but in my experience this is pretty rare. Most of the time I set up a health check in the docker compose file or I extended the Dockerfile and add my own. You sometimes need to add a tool (like curl) to do the health check anyway.
3. It’s a feature of the container, but the app needs to support some way of signaling “health”, such as through a web API.
4. It depends on your needs. You can do all of the above. You can do so-called black box monitoring where you’re just monitoring whether your webapp is up or down. Easy. However, for a business you may want to know about problems before they happen, so you add white box monitoring for sub-components (database, services), timing, error counts, etc.

To add to that: health checks in Docker containers are mostly for self-healing purposes. Think about a system where you have a web app running in many separate containers across some number of nodes. You want to know if one container has become too slow or non-responsive so you can restart it before the rest of the containers are overwhelmed, causing more serious downtime. So, a health check allows Docker to restart the container without manual intervention. You can configure it to give up if it restarts too many times, and then you would have other systems (like a load balancer) to direct traffic away from the failed subsystems.

It’s useful to remember that containers are “cattle not pets”, so a restart or shutdown of a container is a “business as usual” event and things should continue to run in a distributed system.

So I’m also using Beszel and Ntfy to track my systems because it’s lightweight and very very easy. Coming from having tried Grafana and Prometheus and different TSDBs I felt like I was way better off.

I’ve been following Beszels development closely because it was previously missing features like container monitoring and systemd monitoring which I’m very thankful for them having added recently and I use containers as my primary way of hosting all my applications. The “Healthy” or “Unhealthy” status is directly reported by Docker itself and not something Beszel monitors directly so it has to be configured, either by the configuration in the Dockerfile of the container image or afterwards using the healthcheck options when running a container.

As some other comments mentioned, some containers do come with a healthcheck built in which makes docker auto-configure and enable that healthcheck endpoint. Some containers don’t have a healthcheck built into the container build file and some have documentation for adding a healthcheck to the docker run command or compose file. Some examples are Beszel and Ntfy themselves.

For containers that do not have a healthcheck built into the build file it is either documented how to add it to the compose or you have to figure out a way to do it yourself. For docker images that are built using a more standard image like Alpine, Debian or others you usually have something like curl installed. If the service you are running has a webpage going you can use that. Some programs have a healthcheck command built into it that you can also use.

As an example, the postgresql program has a built in healthcheck command you can use of that’ll check if the database is ready. The easiest way to add it would be to do

    healthcheck:
      test: ["CMD", "pg_isready", "-U", "root",  "-d", "db_name"]
      interval: 30s
      retries: 5
      start_period: 60s

That’ll run the command inside the container pg_isready -U root -d db_name every 30 seconds but not before 60 seconds to get the container up and running. Options can be changed depending on the speed of the system.

Another example, for a container that has the curl program available inside it you can add something like

    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:3000/"]
      interval: 1m
      retries: 3

This will run curl -f http://localhost:3000/ every 1 minute. If either of the above examples would exit with an exit code higher than 0 Docker would report the container has unhealthy. Beszel will then read that data and report back that the container is not healthy. Some web apps have something along the line of a /health endpoint you can use the curl command with as well.

Unless the developer has spent some extra time on the healthchecks it is often just a basic way to see that the program inside the container is running. However, usually the container itself exits if the program it is running crashes or quits. So a healthcheck isn’t always necessary as the healthcheck will be that the container has abruptly stopped. This is why things like Uptime Kuma is something to consider running alongside Beszel because it can monitor when a web address or similar is down as well even if a container exits which as of now Beszel is still sadly lacking.

I would recommend you read up on the Docker Compose spec for healthchecks since with the other options you can also do things like timeouts and what not, combining that with whatever program you’re running with the healthcheck you can get very creative with it if you must.

My personal recommendation would be to sticking with Uptime Kuma regarding proper service availability healthchecks since it’ll be easier to configure and get an overview of things like slow load times of web pages and containers that have stopped while using Beszel to monitor performance and resource usage.

i just let kubernetes handle it for me. k3s specifically.

I rely on the developers putting in a health check, but few do.

I’ve also got uptime kuma setup, which is kinda like an external healthcheck.

What happened to grafana and Prometheus?

I have been putting off rebuilding my home cluster since moving but that used to be the default for much of this and I’m not hearing that in these responses.

If I go to its web interface (because everything is a web interface) and it’s down, then I know it has a problem.

I could set up monitoring, but I wouldn’t care enough to fix it until I had free time to use it either.

I decided that at my scale, NixOS is easier to maintain. So for me its just a `systemctl status <thing I host>ˋ