This is a helm
release for Project Eirini. In a nutshell Eirini is a Kubernetes backend for
Cloud Foundry, made in the effort to decouple Cloud Foundry from Diego, the only current option of a scheduler. It deploys CF apps
to a kube backend, using OCI images and Kube deployments.
The following CFAR (Cloud Foundry Application Runtime) distributions deploy CF on top of Kubernetes and bundle Eirini with it:
To build the pure yaml files included in our release please run:
./scripts/render-templates.sh <system-namespace> <output-directory>
This will produce the yamls for all eirini components in separate directories. The components needed for cf-for-k8s are core
, events
and workloads
.
Of an overview of how secure Eirini is compared to other popular container runtimes please look at this table
To learn about how you can use Kubernetes security primitives to make your deployment more secure, please take a look at our Security Guidelines.
As of v1.5.0 a single instance of the eirini deployment can take a sustained load of 90 parallel desire LRP operations. A desire operation takes about 300ms on average when under load.
In order to better understand this result we have to state some condtitions that we assumed when performing the tests:
For details about high availability see this doc.
We are working hard towards feature parity with Diego, but still there are some differences in behaviour
It is not possible to set environment variables containing :
to your apps containers because of Kubernetes restrictions.
By default Eirini does not allow docker images running with the root user. Diego allows this because the application runs in a separate user namespace, which is not supported in Kubernetes as of now. However, you can configure Eirini to allow such docker images - see Security Guidelines for more information.
Tasks in Diego are run at most once and once completed you can determine whether they failed or not. In Eirini we run tasks as Jobs in Kubernetes with both completions
and parallelism
set to 1. However, as per the Kubernetes documentation, there is no guarantee that the task won't be ran more than once.
If all the CF apps are running, it is safe to delete all files in /var/vcap/store/shared/cc-droplets/sh/a2/
directory on the blobstore-0
pod.
To do so, you can run this command:
kubectl exec -n <cf-system-namespace> blobstore-0 -c blobstore -- \
/bin/sh -c 'rm -rf /var/vcap/store/shared/cc-droplets/sh/a2/sha256:*'