using vmsh in the vhive lambda environment
Vhive makes available web-endpoints by managing lambda-functions deployed with firecracker-containerd micro-VMs. Lambda-pirate listens for errors in those lambda-functions and then spawns a debug container with shell access into it. The debug container is attached by vmsh which supports multiple hypervisors and guest linux kernel versions. Furthermore vmsh is agnostic towards the micro-VMs userspace like networking and its running services (specifically ssh) which gives developers freedom to change their hypervisor and boot images in a race to the fastest and most lightweight micro-VM.
Supported OS: NixOS
Status: experimental
To reproduce the usecase as presented in the VMSH paper, follow this guide
$ git clone https://github.com/pogobanane/lambda-pirate.git
$ cd lambda-pirate
We use nix with nix flakes to build modules.
To list all defined components use:
$ nix flake show
Packages can be built into result/
with nix build .#$pkgname
i.e.:
$ nix build .#vhive
Enter the development shell to load and make available all command line dependencies and variables:
$ nix develop
If you are working on evaluation machines provided by us, please skip this step as it is already completed.
In NixOS one can include the nixos modules in their configuration to deploy a single-node k3s, firecracker-containerd, containerd, knative and vhive.
To do so include the following configuration in your flake.nix
{
description = "NixOS configuration";
inputs.lambda-pirate.url = "github:pogobanane/lambda-pirate";
outputs = { nixpkgs, lambda-pirate }: {
bernie = nixpkgs.lib.nixosSystem {
system = "x86_64-linux"; # replace with your arch
modules = [
./configuration.nix # or whatever configuration you use...
lambda-pirate.nixosModules.knative
lambda-pirate.nixosModules.vhive
({ config, ... }: {
# for lambda pirate
services.vhive.dockerRegistryIp = 1.1.1.1; # the ipv4 of this machine
})
];
};
};
}
Checkout the nixos modules in nix/modules for further details. Finally apply the changes and start the services
$ nixos-rebuild switch
After adding the nixos modules, the kubernetes manifests have to be deployed:
$ just reset
$ sudo -E kubectl get pod --all-namespaces
NAMESPACE NAME READY STATUS RESTARTS AGE
kube-system metrics-server-86cbb8457f-89hls 1/1 Running 0 5m49s
kube-system local-path-provisioner-5ff76fc89d-k6qlg 1/1 Running 0 5m49s
kube-system coredns-7448499f4d-v67ps 1/1 Running 0 5m49s
metallb-system controller-8687cdc65-jk4bl 1/1 Running 0 5m45s
metallb-system speaker-q5vbb 1/1 Running 0 5m45s
kube-system calico-kube-controllers-8654f74bf8-rnw98 1/1 Running 0 5m44s
default minio-deployment-877b8596f-b8j29 1/1 Running 0 5m45s
istio-system istiod-796c467-grbsl 1/1 Running 0 5m39s
istio-system cluster-local-gateway-949654c8d-wn6fw 1/1 Running 0 3m41s
istio-system istio-ingressgateway-59c64f5f9c-hgk2z 1/1 Running 0 3m41s
knative-serving istio-webhook-56748b47-p9wbb 1/1 Running 0 3m9s
knative-eventing eventing-controller-55b6f79c99-6bfdg 1/1 Running 0 3m10s
knative-eventing eventing-webhook-67877858b4-llt8g 1/1 Running 0 3m10s
knative-eventing imc-controller-5f4bdf86cf-z54tl 1/1 Running 0 3m7s
knative-eventing mt-broker-ingress-546d6868c9-p2qn5 1/1 Running 0 3m7s
knative-eventing mt-broker-filter-685f7f46d8-bcf54 1/1 Running 0 3m7s
knative-serving networking-istio-5db557d5c4-zn2vw 1/1 Running 0 3m9s
knative-eventing imc-dispatcher-6c664678f9-9nzgt 1/1 Running 0 3m7s
knative-eventing mt-broker-controller-cffcc449c-v594x 1/1 Running 0 3m7s
knative-serving default-domain-zcr5v 0/1 Error 0 3m8s
knative-serving default-domain-9wztl 0/1 Error 0 91s
knative-serving webhook-89656b4c5-2m9h7 1/1 Running 0 3m12s
knative-serving autoscaler-569cfb8b96-8mlvn 1/1 Running 0 3m12s
knative-serving controller-74f8f6ccb8-b5q7k 1/1 Running 0 3m12s
knative-serving activator-6d7f96d7fc-6jlk8 1/1 Running 0 3m12s
knative-serving default-domain-t75n7 0/1 Completed 0 81s
kube-system canal-59rld 1/2 Running 0 5m44s
default hellooopsie-0-00001-deployment-94c4d9c74-2q4zf 2/2 Running 0 19s
The vhive deployer has now started a hellooopsie lambda which produces an error
every 15 seconds. Start lambda-pirate (just lambda-pirate
) and wait until an
error is detected and vmsh is attached to the micro VM.
To use your locally checked out lambda-pirate, replace inputs.lambda-pirate.url
with a string/path to it.
To use your own local vhive build, check the commented paths in:
- justfile
- nix/modules/vhive.nix
To clean deploy after you made your changes (deletes k3s/containerd completely everything!)
$ nixos-rebuild switch
$ just reset