[HN Gopher] Show HN: Clace - Application Server with support for...
___________________________________________________________________
Show HN: Clace - Application Server with support for scaling down
to zero
I have been building the open source project
https://github.com/claceio/clace. Clace is an application server
that builds and deploys containers, allowing it to manage webapps
in any language/framework. Compared to application servers like
Nginx Unit, Clace has the advantage of being able to work with any
application, without requiring any dependency or packaging changes.
Clace provides a blue-green staged deployment model for apps. Not
just code changes, even configuration changes are staged and can be
verified before being made live. Clace is not a PaaS solution, it
does not support deploying databases and other auxiliary services.
It does share the fact that it manages containers with PaaS
solutions. Clace is different in that it builds its own reverse
proxy, instead of depending on Traefik/Nginx. This allows Clace to
implement features like shutting down idle apps and adding app
level OAuth authentication. Clace runs natively on Windows/OSX in
addition to Linux. Clace works with Docker/Podman/Orbstack. Clace
allows you to run hundreds of apps on a single machine. Since app
containers are shut down when not in use, there is no CPU/memory
resource usage when the apps are idle. It provides a Google Cloud
Run type interface on your own hardware. https://clace.io/ has a
demo video and docs. Do let me know any feedback.
Author : ajayvk
Score : 51 points
Date : 2024-09-11 13:50 UTC (9 hours ago)
(HTM) web link (github.com)
(TXT) w3m dump (github.com)
| ajayvk wrote:
| An example to create an app is clace app create
| --approve --spec image --param
| image=ghcr.io/gchq/cyberchef:latest --param port=80 -
| cyberchef.localhost:/
|
| This creates a Cyberchef app from its image at
| https://cyberchef.localhost:25223/, which can be bookmarked.
| Opening that link will create the container (first access can be
| slow since the image is downloaded). The Cyberchef container is
| shut down when not in use. Subsequent access to the url will
| instantly create the container.
|
| Using --spec container allows you to deploy code from any github
| repo which has a Containerfile. There are also language specific
| specs (mainly python currently). For example
| clace app create --approve --spec python-fasthtml --param
| APP_MODULE=basic_ws:app
| https://github.com/AnswerDotAI/fasthtml/examples
| fasthtmlapp.localhost:/
|
| creates a FastHTML based app, building the image and starting the
| container. Checking out the source code and using local path
| allows you to setup a dev environment by adding the --dev option.
| No dependencies have to installed on the dev machine for this.
| Docker/Podman is the only dependency. For example
| clace app create --dev --approve --spec python-fasthtml --param
| APP_MODULE=basic_ws:app ~/mycode/fasthtml/examples
| fasthtmlapp.localhost:/
|
| The goal with Clace is to build an application server for easily
| and securely managing internal tools.
| pkphilip wrote:
| Pretty cool!
| ajayvk wrote:
| Clace is built in go as a reverse proxy. Clace uses
| https://github.com/google/starlark-go as its configuration
| language. This allows full Hypermedia driven apps to be built
| in Clace, running within the main Clace process. For example:
| clace app create --approve
| github.com/claceio/apps/system/disk_usage /disk_usage
| clace app create --approve
| github.com/claceio/apps/utils/bookmarks /book
|
| installs a du like tool and a bookmark manager. These are
| Hypermedia based apps using HTMX which run in Starlark. Works
| across Linux/OSX/Windows, with no dependencies to install.
| Not even containers are required for these apps.
| mdaniel wrote:
| I hadn't heard about Containerfile but it seems to be the
| Podman ecosystem's version of Dockerfile (e.g.
| https://github.com/containers/common/blob/v0.60.2/docs/Conta...
| )
| ajayvk wrote:
| Yes, I used the Containerfile name since is more neutral. The
| file format is same across Containerfile and Dockerfile.
| Clace works with either file name.
| seabrookmx wrote:
| Neat! I currently use jwilder/nginx-proxy and docker compose to
| run a small fleet of apps on my home server. This all-in-one
| solution sounds like it would be more streamlined for single-
| machine deployments like this.
|
| I use Kubernetes in <day-job> and while I'm a big fan, it's
| incredible overkill for running a few services ala. Syncthing and
| Vaultwarden.
| ajayvk wrote:
| Clace is more of an AppServer than a PaaS solution. Docker
| Compose is not supported currently. If you have a postgres
| database and want to deploy multiple apps which access the same
| database, then Clace provides blue-green staged deployment,
| GitOps, OAuth support, auto-pause etc for those apps. The app
| updates are atomic (all-or-nothing). The postgres database
| itself will have to be managed outside of Clace. If you want
| each app to have its own database, then a PaaS solution which
| supports deployment of pre-packaged apps, including Docker
| Compose support is what you want.
|
| Clace is targeting use cases where you have external
| databases/REST API/CLI tools etc already and want to build and
| deploy multiple apps pointing to them. AppServer for deploying
| internal tools for use across a team is a target use case. For
| local dev, one use case is that Clace helps you set up a dev
| environment for webapps, with auto-reload, without having to
| setup any dependencies.
| byrnedo wrote:
| Check out https://github.com/skateco/skate
|
| I'm building it for exactly that reason. Multihost and supports
| k8s manifests.
| mdaniel wrote:
| I'm not trying to yuck your yum, but you'll want to be _very
| careful_ about the uncanny valley of squatting on k8s
| manifests since there is a _ton_ of functionality in those
| files and (as best I can tell) only by reading your readme
| can one tell which features actually work versus are just
| silently(?) swallowed
| byrnedo wrote:
| Yeah, you're right, I'd need to at least have some kind of
| table of supported attributes, or even mint my own schema
| with the subset that's valid for skate.
| ttymck wrote:
| Perhaps a tool that processes a k8s manifest and produces
| a modified manifest containing only the attributes that
| are supported?
| stackskipton wrote:
| I think not bundling this into Kubernetes is mistake if you ever
| want to escape homelab.
|
| Cold starts for JIT languages are problematic for anything
| customer facing so one must be kept running. So now I have two
| platforms and that always sucks as the SRE.
| ajayvk wrote:
| Clace currently is single-node. It uses a SQLite database for
| app metadata (which is what allows it to support atomic updates
| across multiple apps).
|
| I will be adding multi-node support soon. The user will have to
| bring their own Postgres database and load balancer. Multiple
| Clace instances will run in parallel, each spinning up
| containers locally using Docker/Podman. For the internal tools
| use case, I think that should scale to support most workloads.
|
| There will be cases where delegating container management to
| Kubernetes might make sense. Container management in Clace is a
| thin layer, it should be possible to use a K8S
| service/deployment wrapper instead of using local Docker or
| Podman.
|
| The auto-idle feature might cause latency for services with a
| high startup cost. For most apps, it should be fine. It can be
| disabled individually for apps.
| ajayvk wrote:
| Starting on K8s will make scaling down difficult. Starting
| without K8s for Clace makes it possible to control the
| developer experience better. Adding K8s and scaling up can be
| done later.
|
| At an abstract level, Kubernetes is used for managing compute
| (applications) and storage (databases/queues/file
| stores/volumes), stateless services and stateful data stores.
| Compute by itself is easy to scale. Managing storage and
| stateful data stores are where some of the complexity of K8s
| comes from.
|
| For many workloads, it makes sense to use a managed database
| (RDS/managed Redis/S3 etc). Data backups, performance etc are
| easier to handle with a managed service, at the cost of how
| much you pay. If the stateful data stores are externally
| managed, using K8s for compute might be overkill, especially
| when you consider the extras (like ArgoCD, Service Mesh, IDP
| etc) which need to added to make it a useful developer
| experience.
|
| Clace aims to provide a scalable solution for compute only
| scenarios, providing a great developer experience while
| targeting internal tools as the primary initial use case.
| ttymck wrote:
| Scale to zero makes me think it might be useful to specify a time
| window (cron?) where scaling to zero is allowed, i.e. 6pm-8am it
| can scale to zero but during the work day never fully scale to
| zero. Makes me wonder if this is a common pattern.
| ajayvk wrote:
| For webapps which are accessed as web pages or through API,
| looking at the API activity will indicate whether the app is
| being actively used. Clace currently uses no new API calls with
| last 3 minutes as indicating idleness. This is aggressive, it
| can be tuned as required
| https://clace.io/docs/container/config/. For apps using
| websockets or server-sent events, the API activity check will
| not be accurate.
|
| For background jobs and cron jobs, some kind of cron job can be
| defined in the app, to wake up the container. This is not
| supported currently. If an app has background jobs or idle
| check is not accurate, the auto-pause can be disabled.
___________________________________________________________________
(page generated 2024-09-11 23:01 UTC)