pimcore cluster setup with ci/cd

[talsan74]

We are using Pimcore 11.x, which has been deployed on Kubernetes. Our setup includes persistent volume paths mounted as an NFS file system, which are as follows:

1. var/classes  
2. var/config/<bundle>  
3. public/var  
4. var/tmp  
5. config/  

Additionally, we use S3 Flysystem paths for the following:

1. pimcore.asset.storage  
2. pimcore.asset_cache.storage  
3. pimcore.thumbnail.storage  
4. pimcore.version.storage  
5. pimcore.recycle_bin.storage  
6. pimcore.admin.storage  

We have a Docker CI/CD setup. During initial deployment, running /vendor/bin/pimcore-install with the skeleton doesn't cause issues while Dockerizing the application. However, upgrading Pimcore is challenging. For example:

Installing a new bundle and Dockerizing it for deployment to Kubernetes.
Mounting S3 Flysystem paths during the Docker build process is feasible, but mounting the NFS file system during the same process is not.

Question:
What steps should we follow to upgrade Pimcore while using mounted S3 Flysystem paths during the Docker build process, particularly for:

• Installing new bundles.

A detailed explanation of the steps would be highly helpful to upgrade without data loss.

[dpfaffenbauer]

why do you need to mount anything during docker build?

[talsan74]

Correct, which is why finding a solution for that. The approach would be to have the Docker build only copy the code of skeleton and bundles, and then, inside the Kubernetes pod, execute vendor/bin/pimcore-install and enable the installed bundle. Right?"

[dpfaffenbauer]

Yes

[talsan74]

I copied the skeleton and downloaded the required packages during the Docker build process. Inside Kubernetes, I mounted the var/classes directory as an EFS volume. Then, I executed the vendor/bin/pimcore-install command within the pod.

When installing a new bundle, the Docker build retrieves the required dependencies. The latest Docker image is then deployed, and Kubernetes executes the install:bundle command. This workflow seems to be working well for me.

Below is myconfiguration:

##config.yaml

#save datahub config to database
pimcore_data_hub:
    config_location:
        data_hub:
            write_target:
                type: 'settings-store'

##flysystem.yaml

services:
   assets_s3:
        class: 'Aws\S3\S3Client'
        arguments:
            -  endpoint: 'https://s3.ap-south-1.amazonaws.com'
               region: 'ap-south-1'
               version: 'latest'
               credentials:
                   key: '%env(AWS_ACCESS_KEY_ID)%'
                   secret: '%env(AWS_SECRET_ACCESS_KEY)%'
flysystem:
   storages:
       pimcore.document_static.storage:
           adapter: 'aws'
           visibility: public
           options:
#               directory: /var/www/html/public/var/tmp/pages
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: documents
       pimcore.asset.storage:
           # Storage for asset source files, directory structure is equal to the asset tree structure
           adapter: 'aws'
           visibility: public
           options:
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: assets
       pimcore.asset_cache.storage:
           # Storage for cached asset files, e.g. PDF and image files generated out of Office files or videos
           # which are then used by the thumbnail engine as source files
           adapter: 'aws'
           visibility: public
           options:
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: cache
       pimcore.thumbnail.storage:
           # Storage for image and video thumbnails, directory structure is equal to the source asset tree
           adapter: 'aws'
           visibility: public
           options:
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: thumbnail
       pimcore.version.storage:
           # Storage for serialized versioning data of documents/asset/data objects
           adapter: 'aws'
           visibility: public
           options:
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: version
       pimcore.recycle_bin.storage:
           # Storage for serialized recycle bin data of documents/asset/data objects
           adapter: 'aws'
           visibility: public
           options:
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: recycle_bin
       pimcore.admin.storage:
           # Storage for shared admin resources, such as the user avatar, custom logos, ...
           adapter: 'aws'
           visibility: public
           options:
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: admin
       pimcore.temp.storage:
           # Storage for temporary files that work under a batch process, such as export csv
           # must not be `local` for stateless environment
           adapter: 'aws'
           visibility: private
           options:
#               directory: '%kernel.project_dir%/var/tmp'
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: temp-dir
       pimcore.application_log.storage:
           # Storage for application logs
           adapter: 'aws'
           visibility: private
           options:
               client: 'assets_s3'
               bucket: '%env(AWS_BUCKET)%'
               prefix: application-log

I believe the var/classes directory requires NFS mount to update or add classes through the UI. Is there an alternative solution to avoid using NFS for this while still allowing classes to be edited via the UI?

[dpfaffenbauer]

Kubernetes is executing it how? docker entrypoints or a separate job? We use a separate job to do it.

Re classes: I don't think there is.... We usually don't allow users to update classes in prod/stage. We do it locally in the dev env and deploy these changes then. You can rebuild class definitions that come from dev env with `bin/console pimcore:deployment:classes-rebuild```

[tamiltalsan]

Having separate kubernetes job to execute vendor/bin/pimcore-install command for one time

Thank you so much for your help.