see: https://discord.com/channels/817392104711651328/1044802643253547089 It seems like garden will just timeout waiting for the non-existant pods to be ready

Recently came across Garden, exactly what we're looking for. Thanks! I'm wondering if it is possible for a developer to run one service locally and other services in cluster. Essentially, each dev in our team has their own namespace separated cluster. My idea is if the microservice they are developing they can deploy (locally or remote) but all other deps and services are already in cluster. This will help running e2e's but with latest code changes this dev is working on. Is this is a possibility with Garden.

Is there similar functionality to the postSyncCommand available for devMode? I'm pondering the case for compiled languages where one syncs over a binary and want to restart the service and similar.

Hey guys! Just creating this one for visibility, hopefully somebody can take a look at this one, also if there’s a possibility of getting some context on the Code that handles this part I might be able to try to implement a solution for it! https://github.com/garden-io/garden/issues/3367

Hi all, trying to use garden in our drone ci. we have a simple deploy step that runs garden deploy for our stack. I think this issue is intermittent becuase it does not always happen, however ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🌍 Running in namespace staging in environment staging ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Failed getting status for service 'secrets' (from module 'secrets'). Here is the output: ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Error: read ECONNRESET ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ It seems like garden util is failing to sync as well from the logs: │ 2022/12/06 20:27:07 [1543] rsync: [Receiver] safe_read failed to read 1 bytes: Connection reset by peer (104) │ │ 2022/12/06 20:27:07 [1543] rsync error: error in rsync protocol data stream (code 12) at io.c(279) [Receiver=3.2.5] │ │ 2022/12/06 20:27:08 [1544] connect from UNDETERMINED (10.240.0.203) │ │ 2022/12/06 20:27:09 [1545] connect from UNDETERMINED (10.240.0.203) this causes some of our deploy steps to never resolve, has anyone else run into this issue?

I can't seem to figure out how garden determines when it needs to deploy an update to a helm chart, as far as I can tell there should be no change but every deployment causes a new revision to the helm chart release in k8s

I have a module of type Kubernetes that is deploying various resources, including a primary StatefulSet. It appears that the postStart hooks defined in the statefulset are not executing. Would there be any Garden-related reason for this?

In a recent release of Garden, support for syncing of files outside of the garden module's root using absolute paths instead of relative ones was supposed to have been implemented. However, in my testing locally it appears like files aren't being synced. My test scenario was like this. First I tested the module using a relative reference to a file I wanted to sync: /src/my-test-service/garden/garden.yaml

type: container
name: my-test-container
services:
  - name: "my-test-http"
    devMode:
      sync:
        - source: Dockerfile
          target: /Dockerfile

Then I exec into the container and verify that the Dockerfile has been mounted to /Dockerfile Next, I modified /src/my-test-service/garden/garden.yaml like so:

type: container
name: my-test-container
services:
  - name: "my-test-http"
    devMode:
      sync:
        - source: /src/my-test-service/garden/Dockerfile
          target: /Dockerfile

And after deploying again in dev mode, the /Dockerfile mount does not show up. Is this a possible regression, or is there any chance that absolute paths weren't field tested before being marked as supported?

Is there a way to change a particular module's type and all the corresponding config based on the env it's being deployed to? In the specific example I'm trying to figure out it's a module named db where we would want to use a helm chart in one env and just a simple container in a different one.

Not 100% Garden but Garden-adjacent - one of the steps we have in creating our Garden cluster is to create the secret object that contains the config for the docker-based credential helper. It seems like the EKS 1.24 upgrade instructions involves moving over to the kubelet-based credential helper configuration (which has the appearance of a K8s API object, although it's provided as a configuration file to the kubelet). I'm not sure if this is actually needed or if the existing secret object for ECR auth is enough - has anybody upgraded their Garden clusters to EKS 1.24 and has already dealt with this?

Hi! I have a helm chart that requires a version of helm 3.9+, but it looks like garden internally uses helm 3.7.2 currently (garden version 0.12.44). Is there a way to use the system helm installation, or to upgrade the internal helm version used?

Hi All, Looking for some advice on the approach that people have used for local Kubernetes KinD clusters. If I understand correctly, docker images that are built locally need to be uploaded into KinD using

kind load docker-image ...

. In my current setup I have added an

exec

module that does this and depends on the

container

module to get the image name to load. It works, but it feels messy. So, what are others doing? Is it possible to use the in-cluster registry with KinD and get garden to publish the images? I want to support both Docker Desktop (default) and KinD (multi-node testing) so if using the in-cluster registry is a viable option (for KinD) I would only want to configure this for KinD ... not sure how to accomplish that when there is only a single

local-kubernetes

provider configuration. Any thoughts and advice would be much appreciated. Thanks, Si

Here's a snippet from my helm module configuration:

devMode:
  sync:
    - source: .
      target: /app
      exclude: [.vscode, .github, .husky, "node_modules", node_modules]
      mode: one-way-safe

the container module is in the root folder of a remote source repository. The node modules folder is in the root of that same repository. I get the following error in the container logs: System limit for number of file watchers reached The error then references many paths from the node_modules folder.

Hey, I'd like to have a single instance of a service (functioning as a reverse proxy) in a separated namespace. I know I can make that happen using k8s module type but then I'd have to give up on in-cluster building. Is there a way to enforce a namespace using container module type?

Hi! Just ran into the following issue: https://github.com/garden-io/garden/issues/2092 I have a postgers module and I want to create multiple tasks that each task volume a schema.sql file and run it, but sadlly it looks like it might be impossible using Garden (and I don't want to use other tools 😦 ). Any idea if this is possible or will be in the future? It doesn't seems possible to volume a local file/dir into Garden container at the moment. Thanks!

I am currently using garden to keep a staging environment up to date in our CI. (Github Actions) Consecutive runs are "unaware" of each other and helm deployments fail because the k8s resources already exist. As a dirty fix, I added a step that removes the entire staging namespace before each run. I am aware that pushing the .garden folder to the repository would solve this as the context from previous runs is there. But we use the same garden project for individual dev environments, and having a .garden folder pushed to the repo would therefore cause problems. Is there a clean way to solve this problem?

Hi! Garden producing such error when scanning

templated

module. Basically, I have setup as described further:

root-dir/
  project.garden.yml
  tpls/
    tpl.garden.yml
  modules/
    # these modules are templated modules that might use root-dir/tpls/tpl.garden.yml ModuleTemplate
    mdl-one/garden.yml
    mdl-two/garden.yml
    etc.

Modules in modules/ directory are of various types and can depend on each other. There is only one particular

templated

module that cannot be properly processed, while other

templated

modules work fine. From silly logs I found that it is loaded and added twice:

{"msg":"Load module and workflow configs from /root-dir/modules/mdl-one/garden.yml","section":"","timestamp":"2022-12-21T09:10:35.620Z","level":"silly","allSections":[]}
{"msg":"Load module and workflow configs from /root-dir/modules/mdl-one/garden.yml","section":"","timestamp":"2022-12-21T09:10:35.622Z","level":"silly","allSections":[]}
{"msg":"Adding module mdl-one","section":"","timestamp":"2022-12-21T09:10:36.581Z","level":"silly","allSections":[]}
{"msg":"Adding module mdl-one","section":"","timestamp":"2022-12-21T09:10:36.583Z","level":"silly","allSections":[]}

With holidays upon us, much of the Garden crew is out enjoying some rest and time with their families. Your posts may take some time to be seen and replied to. 💟 from Garden

I will likely need to deploy certain services to GPU enabled nodes on GKE. Is it possible to specify pass through values for the generated PodSpec such as described here? https://cloud.google.com/kubernetes-engine/docs/how-to/autopilot-gpus

I would like to be able to reference the programmatically generated image locations of a container module in another module. How are they generated? In my

project.garden.yml

I have

deploymentRegistry:
      hostname: registry-hostname
      namespace: my/namespace

The container module I want to know the image location/tag for has

kind: Module
type: container
name: module-name
repositoryUrl: repository-address

I'd like to be able to get the full image repository and tag. Should it be something like:

registry-hostname/my/namespace/module-name

with a tag of

latest

?

Hi folks. Just curious if there is a way to apply a nodeSelector to a module of type container. It's easy to do this with Kubernetes module type, of course, but specifically, I'm trying to do this for a task (e2e tests). They chew up a ton of resources and I would like to run them on specific nodes. Thanks!

I saw elsewhere on discord that there are plans to deprecate the built in

cert-manager

option here: https://docs.garden.io/advanced/cert-manager-integration#enabling-and-configuring-cert-manager Probably worth indicating in the docs that this won't be supported going forward. I'm looking for recommendations for the least headache-inducing approach to get TLS set up. I am fine with wildcard certificates for our environment subdomain setups (currently

app.domain.com

,

*.dev.domain.com

,

*.preview.domain.com

). I would use the garden implementation if I didn't see the message about deprecation.

Hi 👋, I'm trying to mount a volume locally into a container, but it seems to be failing with

Deployment/contact-api-pod: Error: failed to start container "contact-api-pod": Error response from daemon: error while creating mount source 
path '/host/.aws': mkdir /host: file exists

. I'm trying to mount it in a service

services:
  - name: contact-api-pod
    ...
    volumes:
      - name: aws
        hostPath: /host/.aws
        containerPath: /root/.aws

and I'm running

minikube mount $HOME:/host

locally to add the local file to minikube locally. Am I missing something? Thanks in advance!

Hey, I am having authentication issues deploying a garden project on AWS, I have followed the steps in the docs https://docs.garden.io/guides/in-cluster-building#using-in-cluster-building-with-ecr, but still having issues. See attached image. PS: using the demo start project. I am trying to write a guide on setting up garden in AWS as a detailed guide isn’t available yet as the doc says "Cluster creation & configuration guide coming soon!"

Is it possible to deploy a container without having local Docker? Is it possible to use Podman? If so, how?

I have a script running through a Garden Exec in a workflow. The step's

{steps.step.log}

just outputs "Step command includes global options that will be ignored...". Nothing else in the steps object has any of the logs I'd expect from my script. Is there any way to save or bubble stdout that happens within the Garden Exec, or do I need to just save to file from the script in question? Thanks!

It looks like garden's container module only includes reference to a static number of replicas for services. Are there plans to include some basic horizontal scaling in the template? It would be nice if I didn't have to BYO helm charts.

Hey guys! I been using Garden for a couple of months now and I have a couple of questions about repositories structures and what's the best practice to handle them with Garden. Context: Company-A has 100 repositories with different kind of services (apis, frontends, workers, AI,etc). As this would be too much of a load for Engineers local computers to handle bootstrapping all the repositories with a single

garden deploy

is not a feasible option. Another option that can be used is that for example; imagine we are developing

service-a

and this service need some other services to work: - service-b - service-c - frontend In this case I would run

garden deploy

in

service-a

folder and that will reuse the other services using repositoryUrl's as modules, this will allow engineers to only deploy in their local computers the related services they are working on. Is this approach recommended? Is there any examples of a case like this? Question 2: Is there a way to make that garden works using a folder within the code repository? For example; if we create a folder called

garden-config

and store

garden.yaml, project.garden.yml, modules.yml, garden.env

in there it would be possible to run garden-deploy from that folder but use the files one level above? Does garden has a hard requirement to be at the same level of the dockerfile? Thanks a lot ❤️

For context, I'm using the recommended approach in https://docs.garden.io/guides/namespaces and it's great. However, when a single cluster is used to host multiple namespaces (one per developer, or one per preview deployment), there are also resources that are still global to the cluster. In my case, this includes - ingress - cert-manager When I want to tear down a preview environment or a developer wants to clear our their setup, how do I prevent these cluster level resources from getting removed? Would it makes sense to create unique garden environments for these modules and then template these modules to be disabled for all other environments? That would require deploying them separately, such as

garden deploy --env=devGlobal

, but it would mean when a dev does

garden delete environment dev

it wouldn't try to remove them.

I've been looking for a way to create a persistent volume claim that has files pre-loaded into it from the host machine. I know this can be done using the hostPath parameter on a

volume

module, but since this not supported for real kubernetes clusters, I was trying to find a more portable approach. The use case is that we'd like to load some startup scripts for a containers from a

scripts

folder in our Git repo. Right now, we are using a

configmap

since mounting this as a volume will convert all of the key/values from the

data

section into mounted files - however, this approach forces us to in-line our bash script as yaml text, and this prevents our linters from being able to catch bash scripting issues. Additionally - while I know that this can also be done by specifying a custom Dockerfile which copies in files from the module build directory, we're trying to avoid the "pull image -> copy files/build new image -> republish" cycle that this causes since we have a very large number of containers that are built from a template and are working on reducing the deployment times. Here's a reference template for what I'm trying to do:

kind: ModuleTemplate
name: service-template
inputsSchemaPath: schema.json
modules:
  # We'd like to pre-load some scripts into the persistent volume created by this claim
  - name: "${inputs.serviceName}-scripts-pvc"
    type: persistentvolumeclaim
    spec:
      accessModes:
        - ReadWriteOnce
      resources:
        requests:
          storage: 1Gi

  - type: container
    name: ${inputs.serviceName}-container
    image: "${inputs.imageName}"
    services:
      - name: "${inputs.serviceName}-http"
        command:
            - bash
            - /scripts/startup.sh
        volumes:
          - name: scripts-pvc
            containerPath: "/scripts"
            module: "${inputs.serviceName}-scripts-pvc"