François Scala arcenik

Popular posts

1. Hacking K0S Part 2 : Start patching and hacking

   Create the test image

   In the previous article, we’ve built an image using the image builder. So let’s create a new branch and a new version of the image builder.

   $ cd image-builder
   $ git checkout -b test
   $ cd images/apiserver-network-proxy-agent/
   $ v0.36.0-k0s.1 test
   

   Clone and branch the softwares

   Let’s clone the Kubernetes-sigs repo and create a tgz archive for the next step

   $ git clone https://github.com/kubernetes-sigs/apiserver-network-proxy.git k8ssigs-apiserver-network-proxy
   $ cd k8ssigs-apiserver-network-proxy
   $ tar cfz ../image-builder/images/apiserver-network-proxy-agent/test/k8ssigs-apiserver-network-proxy.tgz .
   

   Replace release with local test branch

   First, replace the wget from Github to use locally generated archive instead.

   WORKDIR /go/src/kubernetes-sigs/apiserver-network-proxy
   
   ARG VERSION HASH
   RUN --mount=type=cache,id=konnectivity-go-modcache,target=/go/pkg/mod \
     --mount=type=tmpfs,target=/tmp \
     set -e \
     && wget -qO /tmp/src.tar.gz https://github.com/kubernetes-sigs/apiserver-network-proxy/archive/refs/tags/v$VERSION.tar.gz \
     && { echo $HASH /tmp/src.tar.gz | sha256sum -c -; } || { sha256sum /tmp/src.tar.gz; exit 1; } \
     && tar xf /tmp/src.tar.gz --strip-components=1
   

   Add the COPY step and use local file instead of wget

   COPY k8ssigs-apiserver-network-proxy.tgz /root
   
   WORKDIR /go/src/kubernetes-sigs/apiserver-network-proxy
   
   RUN --mount=type=cache,id=konnectivity-go-modcache,target=/go/pkg/mod \
     --mount=type=tmpfs,target=/tmp \
     set -e \
     && tar xfz /root/k8ssigs-apiserver-network-proxy.tgz && \
     rm -vf /root/k8ssigs-apiserver-network-proxy.tgz
   

2. Hacking K0S Part 1 : Setup build environment

   Lately, I’ve encounter some issues with my K0S Kubernetes distro running on my Raspberry Pi cluster. The Konnectivity agent is generating a lot of logs and error. This problem has already been reported (like here) but it has been closed without any real response. So let start digging this.

   In this article, we will discuss about the build environment and dependencies. Namely, a container registry to store the artifacts to be tested, a multi architecture builder as the test environment runs on Raspberry Pi and the build environment is AMD64. Finally, validate all this by building and deploying an image using the code without any change.

   Deploy a local registry server with Docker

   Running a registry is simple as running a single container, using the registry image

   Run an additional container for the user interface, using registry-ui for easier management.

   Adds some environment so the web interface can query the registry without being rejected and you get the following services for Docker Compose.

   services:
     registry:
       image: docker.io/library/registry:3.1.2
       ports:
         - 5000:5000
       hostname: registry
       restart: always
       userns_mode: host
       environment:
         REGISTRY_HTTP_HEADERS_Access-Control-Allow-Origin: '[http://myserver.home.local:5080]'
         REGISTRY_HTTP_HEADERS_Access-Control-Allow-Methods: '[HEAD,GET,OPTIONS,DELETE]'
         REGISTRY_HTTP_HEADERS_Access-Control-Allow-Credentials: '[true]'
         REGISTRY_HTTP_HEADERS_Access-Control-Allow-Headers: '[Authorization,Accept,Cache-Control]'
         REGISTRY_HTTP_HEADERS_Access-Control-Expose-Headers: '[Docker-Content-Digest]'
         REGISTRY_STORAGE_DELETE_ENABLED: 'true'
       volumes:
         - /srv/registry:/var/lib/registry
   
     registry-ui:
       image: joxit/docker-registry-ui:2.6.0
       container_name: docker-registry-ui
       ports:
         - 5080:80
       environment:
         - REGISTRY_TITLE=myserver.home.local
         - REGISTRY_URL=http://myserver.home.local:5000
         - SINGLE_REGISTRY=true
         - SHOW_CATALOG_NB_TAGS=true
       depends_on:
         - registry
       restart: always
   

   Create the Docker Buildx multi arch builder

   To use the local unsecure registry you must create a custom Docker Buildx builder. Here are the steps.

3. Victoria Logs Grafana Dashboard

   Create the first dashboard to access Victoria Logs in Grafana

   Adding the datasource

   First you need to add the victoriametrics-logs-datasource plugin. Then you can add the datasource.

   

   Set the URL, then click on the Save & Test buttom at the bottom of the page.

   

4. Logging with Victoria Logs

   Victoria Logs is a scalable logging solution designed to be fast and highly efficient. It provides users and organizations with better observability of their systems and helpsthem identify issues as well as debug, understand and analyze product behaviour.

   

   Single server setup, using Docker Compose

   The setup to runs Victoria Logs on Docker Compose is very easy. One port, one data volume and all the configuration is done through command line parameters. For this simple example, only one parameter: the storage data path.

   services:
     vic-logs:
       image: docker.io/victoriametrics/victoria-logs:v1.52.0
       ports:
         - "9428:9428"
         - "5514:5514"
         - "5514:5514/udp"
       volumes:
         - "/srv/victoria:/victoria"
       command:
         - "-storageDataPath=/victoria/data/logs"
         - "-syslog.listenAddr.tcp=:5514"
         - "-syslog.listenAddr.udp=:5514"
   

   Adding syslog listener is easy as adding some ports and command line options

5. Testing Garage S3 Objects Storage

   After 11 years of development, MinIO has decided to move away from open source community, archiving it’s public Github repository.

   

   So, let’s try Garage from DeuxFleurs, an S3-compatible distributed object storage service designed for self-hosting at a small-to-medium scale.

   

   DeuxFleurs is a non profit association, founded un 2020, with the goal of building tools for a decentralized Internet.

   

   Running with Docker Compose

   For now, let’s stay on a single node, and add it to the docker compose stack.

6. 2026 07 15 Ollama Update to v0.30

   Ollama version 0.30 has been released some time ago, bringing a major change: it now use llama.cpp instead of the internal GGML fork (see v0.30.0)

   This has a big impact on my Strix Halo + Rocm setup

   Perfomance on ollama 0.24.0

   Using gemma4.31b, it runs 100% on the GPU

   NAME          ID              SIZE     PROCESSOR    CONTEXT    UNTIL
   gemma4:31b    6316f0629137    47 GB    100% GPU     262144     4 minutes from now
   

   And using a simple question `what is json ?’ I obtained an answer at ~10 token per second.

   total duration:       1m8.796615034s
   load duration:        164.085088ms
   prompt eval count:    17 token(s)
   prompt eval duration: 205.756478ms
   prompt eval rate:     82.62 tokens/s
   eval count:           702 token(s)
   eval duration:        1m8.1488854s
   eval rate:            10.30 tokens/s
   

   Performance on ollama v0.30.5

   After upgrading to v0.30.5, ollama-rocm now detect both the CPU and the GPU

   jui 15 19:18:57 phi25 ollama[1548]: device_info:
   jui 15 19:18:57 phi25 ollama[1548]:   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (31727 MiB, 31727 MiB free)
   jui 15 19:18:57 phi25 ollama[1548]:   - ROCm0   : AMD Radeon 8060S Graphics (98304 MiB, 29071 MiB free)
   

   And now, the modele is spread across the two devices

   NAME          ID              SIZE     PROCESSOR          CONTEXT    UNTIL
   gemma4:31b    6316f0629137    34 GB    57%/43% CPU/GPU    262144     4 minutes from now
   

   And this has a major impact on perfomance, as the token per second is almsot devided by two.

   total duration:       1m44.78409003s
   load duration:        309.925333ms
   prompt eval count:    17 token(s)
   prompt eval duration: 770.637ms
   prompt eval rate:     22.06 tokens/s
   eval count:           689 token(s)
   eval duration:        1m43.701536s
   eval rate:            6.64 tokens/s
   

   An issue about this is opened on github: #16462