socialwifi / kubepy

Python wrapper on kubectl that makes deploying easy.

kubepy

Python wrapper on kubectl that makes deploying easy.

Installation

Requires python 3.5 and configured kubectl. To install run: pip3 install kubepy

Usage

You can use this package to install all yml definitions from given directory. Just run kubepy-apply-all from a directory where all of you Kubernetes definition yml files are.

Supported Kubernetes resources:

• CronJob
• Deployment
• StatefulSet
• Job
• Pod (used to run a one-off command)
• Service
• Ingress
• Secret
• StorageClass
• PersistentVolume
• PersistentVolumeClaim
• PodDisruptionBudget

Options:

• --directory <path> - uses path instead of local directory. Can be used multiple times to add new and partially override existing definitions.
• --build-tag <tag> - sets tag to all images without specified tag in your definition files
• --label <key>=<value> - adds label to definition. Can be used multiple times.
• --label-pod <key>=<value> - adds label to each pod definition. Can be used multiple times.
• --annotate <key>=<value> - adds annotation to definition. Can be used multiple times.
• --annotate-pod <key>=<value> - adds annotation to each pod definition. Can be used multiple times.
• --replace - if present, replaces deployments instead of updating them. Default: false.
• --host-volume <name>=<path> Adds host volume to each pod definition. Can be used multiple times.
• --env <VAR>=value Sets environment variable on every container.
• --max-job-retries <n> While waiting for job to finish if it fails n times than delete job and fail. Job sometimes can still be executed more than n times.

There is also kubepy-apply-one command which is called as kubepy-apply-one name1 [name2 ...] It applies only files selected files. Names should be without ".yml". It accepts all options from kubepy-apply-all. Additionally you can pass option:

• --show-definition - shows definition instead of applying them.

Applying jobs.

Applying usualy means that underlying kubectl apply or kubectl replace is called. However applying job is treated differently. To ensure that job finished and succeeded kubectl waits for job to finish and fails if job failed.

Applying pods.

Usually you don't need to apply pods manually, but if you want to run some kind of check and you need to know if it succeeded without retries then you can use pod with restartPolicy: Never. Only pods with this policy are currently supported. They are treated as jobs, so applying waits for them to finish and fails if they fail.

kubepy's Issues

Replace secrets instead of applying them

It is workaround for some of issues.
kubernetes/kubernetes#37216
kubernetes/kubernetes#23564

Print logs from failed jobs

Make ApiErrors more verbose.

Probably passing them stderr would be enough.

Remove --max-job-retries after it is done by kubernetes itself.

kubernetes/kubernetes#30243

Add option --dont-replicate

In development I would like to set all "replicas" to 1. It is easier to debug and requires less resources.