Dozzle K8s is here! See here for instructions and provide feedback.

[amir20]

Hi all,

While looking at the survey results, I noticed a few people rely on K8s support heavily. I am currently working on k8s support and it is almost complete. I still need to add documentation. But I am hoping I could get a couple volunteers who would be interested in testing Dozzle in their k8s cluster.

Implementation for Dozzle in k8s is a lot more performant since stats for containers is returned across all containers. Additionally, k8s services are stable so agents are not required.

I will be adding docs soon. For those who are interested, I will provide directions. In the future, I hope to release a helm package.

[troke12]

I have some cluster k8s, can't wait for that and go test it!

[amir20]

@troke12 could you try it? I put the config below.

[troke12]

Ah sure, i'll test it for tomorow

[amir20]

Hello,

I have been excited about adding k8s support. In the survey, a few people expressed, but not a lot. I thought it would be a good addition and apparently, judging from the ❤️ it seems like people might like.

Here is what you need to test Dozzle with k8s. Dozzle master branch has k8s support with DOZZLE_MODE=k8s.

Installing

This is the file I did to spin up Dozzle

# rbac.yaml
apiVersion: v1
kind: ServiceAccount
metadata:
  name: pod-viewer
---
# clusterrole.yaml
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: pod-viewer-role
rules:
  - apiGroups: [""]
    resources: ["pods", "pods/log", "nodes"]
    verbs: ["get", "list", "watch"]
  - apiGroups: ["metrics.k8s.io"]
    resources: ["pods"]
    verbs: ["get", "list"]
---
# clusterrolebinding.yaml
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: pod-viewer-binding
subjects:
  - kind: ServiceAccount
    name: pod-viewer
    namespace: default
roleRef:
  kind: ClusterRole
  name: pod-viewer-role
  apiGroup: rbac.authorization.k8s.io
---
# deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: dozzle
spec:
  replicas: 1
  selector:
    matchLabels:
      app: dozzle
  template:
    metadata:
      labels:
        app: dozzle
    spec:
      serviceAccountName: pod-viewer
      containers:
        - name: dozzle
          image: amir20/dozzle:master
          ports:
            - containerPort: 8080
          env:
            - name: DOZZLE_MODE
              value: "k8s"
            - name: DOZZLE_LEVEL
              value: "debug"
---
# service.yaml
apiVersion: v1
kind: Service
metadata:
  name: dozzle-service
spec:
  type: ClusterIP
  selector:
    app: dozzle
  ports:
    - port: 8080
      targetPort: 8080
      protocol: TCP

Install using kubectl apply -f dozzle.yml

Dozzle for K8s has turned out to be very performant because the API for k8s are more efficient.

Requirements

1. To collect metrics, metrics server API is required. In the future, I plan to make this optional where if it is not installed, it would just fallback to no stats. Metrics Server can be installed with kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml
2. rbac permission to pods and metrics is required to fetch and view logs. See above for example.

What doesn't work?

1. K8s doesn't currently support filters or user filters. I plan to come back to this later.
2. There maybe bugs as my k8s cluster is very simple.
3. The UI might still say Docker compose or services. I haven't updated a lot of the UI to reflect k8s

Please test this and provide feedback on this discussion. I am looking for a few folks who can install and learn from their experience. I also want to make sure Dozzle remains light so if you see high memory or CPU usage then let me know.

Thanks 🎉🚀

[amir20]

I have released the latest version and docs are at https://dozzle.dev/guide/k8s#kubernetes-setup

[JeuFore]

Hello,

I've just had a quick try on my k8s cluster. It's a success, the logs are up correctly!
I noticed that there was only the namespace default whitelist, is this a deliberate choice?

Thanks for the support ;)

[JeuFore]

Yes, I think it should remain configurable as needed, and if not defined, then all namespaces are watch.

On the other hand, we may have to look for a short- or medium-term solution involving categorization by namespace or something else.
On K8S, there are a lot of pods and replication, so visualization can be difficult to use.

[anthony-tarantini]

I agree, I think if not specified we should default to 'all namespaces'.

I have Dozzle up and running in my cluster and it grabs all the nodes nicely, but since I don't have anything in my 'default' namespace, I can't see any pods or logs at the moment.

[amir20]

On the other hand, we may have to look for a short- or medium-term solution involving categorization by namespace or something else.

I am still considering this. I haven't fully implemented labels yet. However, that wouldn't solve your problem because labels are meant to show or hide pods.

[amir20]

@JeuFore I pushed amir20/dozzle:pr-3631. Can you give it a try? It should be usable with DOZZLE_NAMESPACE or --namespace. I'll update the doc if it works. It defaults to "".

[amir20]

If someone could provide a screenshot, I would better understand how many nodes or pods there are. Then, I can consider how to improve my grouping.

[dhop90]

Hey @amir20 Thanks for developing k8s version. I tried installing and I'm getting some errors.

Running v8.11.1 - it only finds pods in default namespace and the dozzle pod

Running pr-3631 - It finds all the containers (appox. 529) from all namespaces but it crashes. I've included the logs from running pr-3631. Looks like it maybe a memory leak.

The last log messages are:

`{"level":"debug","event":"create","id":"db-cb9c6f99d-qr96h:db","time":"2025-02-14T20:57:26Z","message":"container event from store"}
panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x0 pc=0x9a9c19]

goroutine 14 [running]:
github.com/amir20/dozzle/internal/utils.(*RingBuffer[...]).Push(0x25e7420, {{0x0, 0x0}, 0x3fb999999999999a, 0x0, 0x4187dc8000000000})
/dozzle/internal/utils/ring_buffer.go:38 +0x59
github.com/amir20/dozzle/internal/container.(*ContainerStore).init(0xc000214150)
/dozzle/internal/container/container_store.go:397 +0x232
created by github.com/amir20/dozzle/internal/container.NewContainerStore in goroutine 1
/dozzle/internal/container/container_store.go:54 +0x2db`
dozzle-58c6bb99cc-mwbsj_dozzle (1).log

[amir20]

If you mean agents then no. It's completely different mode. I would think one instance for k8s and another for docker is good enough. Probably not worth the complexity for just a few people.

[dhop90]

I do. Went ahead and configured two separate instances, one for k8s and another for docker

[amir20]

Nice. Yea that's the way to do it.

Are using any thing else with k8s? Like auth?

[dhop90]

yep, got a whole setup. using both authentik and pocketid, managing it with rancher. Here is my tech stack. Not all applications are included.

[anthony-tarantini]

@anthony-tarantini I see you have more than 100 containers. Hmmm 🤔 everything should be performant. But I do wonder how the UI should be better. Search is probably the best choice right now to navigate between pods. Do you use k8s labels a lot? I wonder if Dozzle can be smart and group by namespaces.

Sorry for the late reply, but the performance doesn't seem too shabby honestly with the amount of containers. I haven't moved around too much in there yet, but some filtering tools would be pretty sweet. There are a bunch of containers in there that I really don't care about very often (a bunch of Cilium containers, and some of the operator containers), so being able to default filter those out would be nice.

[scuba75]

I must say I am glad I looked this project up again for another project I was working on. I haven't used it much since switching to k8s.
As for suggestions, I would like to be able to use an annotation or label to show, hide and or group pods together.
With multi pod workers I don't really know which one might have taken the job that errored and currently have to look thru each pod to see. Being able to see all pods from a deployment, or statefulset or daemonset at once would be nice.

[amir20]

https://dozzle.dev/guide/k8s#namespaces-and-filters. Filters should now work as expected.

[scuba75]

getting Container has no logs yet when using filters

actually I am getting this with or without filter applied

v8.11.3 i get logs
V8.11.4 no logs

[amir20]

Hmm I'll check. I might have broken something. Can you create a bug so I can track it? Also provide how you are running Dozzle.

[amir20]

I wasn't able to reproduce this @scuba75. I am using k3s to deploy a local cluster. One thing that changed in latest version is that I pass stdout, stderr or all accordingly to K8s. It doesn't seem like it matters but I wonder if that broke it.

If you can investigate a little and open a new issue, I can try disabling the stream and see if that is what broke it.

[scuba75]

I wasn't able to reproduce this @scuba75. I am using k3s to deploy a local cluster. One thing that changed in latest version is that I pass stdout, stderr or all accordingly to K8s. It doesn't seem like it matters but I wonder if that broke it.

If you can investigate a little and open a new issue, I can try disabling the stream and see if that is what broke it.

done
#3646

[christiaangoossens]

Very cool and easy to install. Only future goal I would like to see is namespace grouping support instead of only per host, just like with docker swarm/docker compose stacks. Love this, thank you!

[maximemoreillon]

I would also be in favor of namespace grouping instead of host grouping

[myevit]

DOZZLE_NAMESPACE not sure how to filter multiple namespaces

• name: DOZZLE_NAMESPACE
  value: "production, development"

• name: DOZZLE_NAMESPACE
  value: "production development"

• name: DOZZLE_NAMESPACE
  value: "(production|development)"

  all not valid

[amir20]

That's not supported. See #3678.

[amir20]

I have added support for this in #3848

Can you test it out with amir20/dozzle:pr-3848?

[myevit]

I have added support for this in #3848

Can you test it out with amir20/dozzle:pr-3848?

So far, all good. working. Thanks!

[amir20]

Great. I released the feature yesterday. So you can use the latest release again.

[thanhhoan062]

I have the developer role in my K8s cluster, which means I only have permissions to create Role and RoleBinding resources. I do not have cluster-admin or broader privileges.
My requirement is to forward logs only from a few specific containers in namespaces that I explicitly specify, rather than collecting logs from the entire cluster.
Is there a way to configure Dozzle so that I can restrict log access and forwarding to only the containers or namespaces I choose, given my limited permissions?

[amir20]

You should be able to use https://dozzle.dev/guide/k8s#namespaces-and-filters. But I am not too familiar with dev role. Try it and see what works or not. You'll have to investigate since I don't have a similar env.

[adedamolalawal]

Hello @amir20, Thanks for the work on this. Quick observation, the documentation says

Dozzle supports multiple namespaces, you can set the "DOZZLE_NAMESPACES" environment variable to a comma separated list of namespaces.

However, what works is DOZZLE_NAMESPACE (without the S) for multiple namespaces.

[adedamola-lawal]

See PR to update documentation here: #4072

[vtngan]

Currently, I’ve tried deploying Dozzle on k8s for the dev and staging environments, and everything works as I expect. However, when I tried creating Custom Groups on k8s, I noticed that the Custom Groups option doesn’t appear in the k8s environment.

Meanwhile, in a Docker standalone test environment, I can easily create Custom Groups just by following the documentation:
https://dozzle.dev/guide/container-groups

Is Custom Groups currently only supported for Docker? Do you have any way to create Custom Groups in a k8s environment? I’m really looking forward to this.

[amir20]

Custom groups is not implemented for k8s.

[vtngan]

Do you have any plans to implement Custom Groups on k8s?
I think this would be a very useful feature on k8s. In a k8s cluster, when using policies related to Replicas or HPA, there are many pods handling requests concurrently. If we could group these pods, we’d get an overall, consolidated view of their logs.
Do you feel the same way?

[amir20]

I agree it would be useful. However, k8s is about 0.1% usage. Most people are using swarm or just docker. So unless there is more usage for k8s, it isn't worth a lot of my time.

[abnerrizzi]

I created a very simple helm chart at https://github.com/abnerrizzi/charts
and fell free to make any changes

[zuri-33]

is it possible to install dozzle for k8s in a machine and view pods from other remote ips? like dozzle for docker with agents.

[amir20]

Nope. Agents and remote host not possible in k8s.

[dhop90]

is it possible to install dozzle for k8s in a machine and view pods from other remote ips? like dozzle for docker with agents.

Sort of, I end up running two instances for dozzle, 1 that can see all nodes in the cluster and the other one can see non-cluster agents.

[scuba75]

To me where the main issue is for use of this with k8 is that i want to see log grouping by deployment and/or stateful set.
If i am having an issue i know it is with a specific deployment/statefulset and just have to chase the pod.
Grouping by Node is not helpful because the pods can be all over. Node metric info is nice, but i just want to be able to view all pod/container logs of a deployment at once.

[amir20]

I think that's a fair ask actually. I don't use k8s mode enough to improve it. I need someone to brainstorm with me though. Dozzle already groups conttainers in Docker mode using dev.dozzle.name label or compose stack name. I just don't know what to use for K8s.

@scuba75 can you create a new feature issue and help me come up with the right grouping name? I am curious to how you have deployed and how Dozzle can access the deployment name.

[scuba75]

2 ways I can think of doing this.

1. Group by label instead of host, can than use labels to view multiples at once.
2. Give a way to see all container logs at once (maybe you can do this), can than make a pod per filter label (multiple instances)

This must be a me problem though cause with all the tools for k8 out there no one really shows all pod logs of a set together.

[dhop90]

Group by namespace would be nice!

[MnLujan]

would be great!