janekbaraniewski / kubeserial

k8s operator for managing serial devices.

kubeserial's Issues

Fix hardcoded values

Some values are hardcoded, they should be stored in some common place for const values:

• template file paths

DaemonSet apiVersion should be updated

On Kubernetes 1.17, I get the below crashloop when running this:

{"level":"error","ts":1585579148.2771335,"logger":"cmd","msg":"Manager exited non-zero","error":"no matches for kind \"DaemonSet\" in version \"apps/v1beta2\"","stacktrace":"github.com/go-logr/zapr.(*zapLogger).Error\n\t/home/janek/go/pkg/mod/github.com/go-logr/[email protected]/zapr.go:128\nmain.main\n\tkubeserial/cmd/manager/main.go:150\nruntime.main\n\t/usr/lib/go/src/runtime/proc.go:203"}

apps/v1 is the stable apiVersion since at least K8S v1.14

interested in branching this to more serial devices

I have been working on alot of home automation stuff in kubernetes.

Was wondering if there was any way of reaching you to ask some questions regarding how to possible work with this to port some stuff like my serial zwave device over so i could use it in k8s.

thanks

Namespace not created as part of the kubeserial.yaml

kubectl create -f kubeserial.yaml

clusterrole.rbac.authorization.k8s.io/kubeserial created
clusterrolebinding.rbac.authorization.k8s.io/kubeserial created
customresourcedefinition.apiextensions.k8s.io/kubeserials.app.kubeserial.com created
Error from server (NotFound): error when creating "kubeserial.yaml": namespaces "kubeserial" not found
Error from server (NotFound): error when creating "kubeserial.yaml": namespaces "kubeserial" not found

Document helm chart values

Improved metrics + default dashboard

Dummy metrics are in place, once everything is working they should be improved to provide some actual data.

It would also be cool to include some default dashboard in chart

Use CRDs for devices

Instead of keeping all configuration in one crd, devices should be defined in separate objects, one per device

Release CI should update version in project files

ATM version in project files is always APP_VERSION or 0.0.1 and then it's substituted using hack/update-chart-version.sh

This results in local dev images always build with tag 0.0.1-<git-sha>, even though they are modifying different version

Would be good to commit bump of this version as part of release workflow

Merge kubeserial-device-monitor into this repo

https://github.com/janekbaraniewski/kubeserial-device-monitor

Fix labels/selectors of objects created by controller, pass them from helm

Make manager optional, support injecting devices to other workloads

Need to support scenario where user wants to manage software that's using serial device on their own. Example can be home automation using HomeAssistant:

• we want to use device like Zigbee bridge
• we don't want to remove HomeAssistant deployment when this device will be disconnected
• we want to manage HomeAssistant config separately
• ...

in such cases KubeSerial would schedule device gateway which then would be mounted by dedicated sidecar container running in management software pod on shared volume

Q:

• can this be done automatically? I give KubeSerial name of deployment I want to expose my device to and operator adds sidecar, volumes and everything else. How would it behave when deployment is upgraded by helm?
• maybe mutating webhook is the way to go?

Update images of other components when deploying new version

Currently when running helm upgrade with new version, managers and other components that are created by operator are not updated.

➜  VERSION=0.1.5 make deploy-dev
controller-gen... Done!
helm upgrade --install kubeserial-crds ./deploy/chart/kubeserial-crds
Release "kubeserial-crds" has been upgraded. Happy Helming!
NAME: kubeserial-crds
LAST DEPLOYED: Tue Apr 26 14:36:56 2022
NAMESPACE: default
STATUS: deployed
REVISION: 11
TEST SUITE: None
helm upgrade --install kubeserial ./deploy/chart/kubeserial -f ./deploy/chart/kubeserial/values-local.yaml
Release "kubeserial" has been upgraded. Happy Helming!
NAME: kubeserial
LAST DEPLOYED: Tue Apr 26 14:37:00 2022
NAMESPACE: default
STATUS: deployed
REVISION: 11
TEST SUITE: None

➜  k get deploy kubeserial -o=jsonpath='{.spec.template.spec.containers[0].image}'
janekbaraniewski/kubeserial:0.1.5

➜  k get daemonset kubeserial-monitor -o=jsonpath='{.spec.template.spec.containers[0].image}'
janekbaraniewski/kubeserial-device-monitor:0.0.1-de8647a

Use CRDs for managers

Current implementation requires managers to be hardcoded. This makes it difficult to test and extend and impossible for end users to add their own managers according to their needs.

Would be much easier to add separate CRD to define new managers as k8s objects

Inject socat command to container run command

since sidecar approach doesn't work that's the only path left

Add project docs

Would be good to have small docs site (gh pages) describing

• components
• CRDs
• configuration
• installation
• examples

More control over where device monitors are scheduled

ATM device monitor will be scheduled to all cluster nodes. It's possible that we don't really want that - some nodes are running in cloud, some are running physically and we want only physical ones to monitor for devices.

Support watching all namespaces

Is there anyway to generalize this so that this can be used with any project?

First, this is a really cool project. Thank you for doing all the work.
I think this project can be used with lot of home automation products. Such as Zwave or zigbee dongles. Is there anyway to achieve this without rebuilding the official images/charts?

Add integration tests with emulated USB device

• figure out how to simulate USB device to trigger udev rules
• figure out how to connect USB device to minikube cluster (which driver?)
• figure out how to run this in github CI

Expose log level setting in chart

When device is still connected while kubeserial is uninstalled link created by udev rules remains on node

no stable image available

kubeserial/deploy/kubeserial.yaml

Line 192 in 156abef

image: janekbaraniewski/kubeserial:stable

changing this to latest and it downloaded the image as expected.

Stick to 1 way of providing app config

replace things like os.Getenv("NODE_NAME") with flags passed to binary