Hi All, we are planning to install an self-managed OCP cluster on GCP cloud. Had couple of questions regarding the same

1) Does manage support GCP cloud storage for doclinks. GCP does offer a S3 compatiable API just wanted to confirm if someone has configured the same for doclinks using the system properties mxe.cosaccesskey, mxe.cossecretkey etc

2) Also while sizing the infra, the calculator recommended 3 master and 3 worker nodes wherein each worker is suggested to have 400GB of storage. We are currently planning to go with standard-pd (an ssd) which will be mounted to each worker node. Checking online we could see that a storage type of RWX (read write many) is required for mas installation  in a cluster and standard-pd does not support RWX, can anyone confirm on this whether worker node storage itself needs to be changed to a RWX compatible storage? Or if the nodes need additional RWX storage to be provisioned? Just to add the database will reside outside the cluster and doclinks will be handled in GCP object storage preferably so not sure which components in the cluster itself require RWX storage?

@Ivan Lagunov would appreciate your inputs here

Looking forward to your insights and recommendations, especially around the doclinks configuration and storage setup. Any guidance based on prior implementations or best practices would be highly appreciated. Thanks in advance

Hi All, we are planning to install an self-managed OCP cluster on GCP cloud. Had couple of questions regarding the same

1) Does manage support GCP cloud storage for doclinks. GCP does offer a S3 compatiable API just wanted to confirm if someone has configured the same for doclinks using the system properties mxe.cosaccesskey, mxe.cossecretkey etc

2) Also while sizing the infra, the calculator recommended 3 master and 3 worker nodes wherein each worker is suggested to have 400GB of storage. We are currently planning to go with standard-pd (an ssd) which will be mounted to each worker node. Checking online we could see that a storage type of RWX (read write many) is required for mas installation  in a cluster and standard-pd does not support RWX, can anyone confirm on this whether worker node storage itself needs to be changed to a RWX compatible storage? Or if the nodes need additional RWX storage to be provisioned? Just to add the database will reside outside the cluster and doclinks will be handled in GCP object storage preferably so not sure which components in the cluster itself require RWX storage?

@Ivan Lagunov would appreciate your inputs here

Looking forward to your insights and recommendations, especially around the doclinks configuration and storage setup. Any guidance based on prior implementations or best practices would be highly appreciated. Thanks in advance

Hi Abhinav,

seems like You are confusing primary disk (with ephemeral storage and CoreOS) with persistent storage (used by pods). Primary disks are fine using any kind of block storage, so standard-pd should be fine. Just keep in mind that master nodes require very fast disks to operate properly. See: https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/scalability_and_performance/recommended-performance-and-scalability-practices-2#recommended-etcd-practices

As for the persistent storage: GCP has native Filestore driver which You can use for RWX capable storage. If You use a CSI driver and a StorageClass then PVCs will create a share (using Filestore driver) in GCP and bind it to the PVC dynamically. No need to create those disks upfront. See the table here: https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/storage/understanding-persistent-storage#pv-access-modes_understanding-persistent-storage

In Your case with DB outside of cluster and attachments in Obejct Storage You might not need RWX capable storage, but that would depend on the actual configuration of MAS apps. For example any PVC used by Manage (maybe for migration manager or jms backing store) should be RWX (in case of non SNO setup) as it can be mounted by all ServerBundle pods and maxinst pod, which may run in different nodes.

Hi All, we are planning to install an self-managed OCP cluster on GCP cloud. Had couple of questions regarding the same

1) Does manage support GCP cloud storage for doclinks. GCP does offer a S3 compatiable API just wanted to confirm if someone has configured the same for doclinks using the system properties mxe.cosaccesskey, mxe.cossecretkey etc

2) Also while sizing the infra, the calculator recommended 3 master and 3 worker nodes wherein each worker is suggested to have 400GB of storage. We are currently planning to go with standard-pd (an ssd) which will be mounted to each worker node. Checking online we could see that a storage type of RWX (read write many) is required for mas installation  in a cluster and standard-pd does not support RWX, can anyone confirm on this whether worker node storage itself needs to be changed to a RWX compatible storage? Or if the nodes need additional RWX storage to be provisioned? Just to add the database will reside outside the cluster and doclinks will be handled in GCP object storage preferably so not sure which components in the cluster itself require RWX storage?

@Ivan Lagunov would appreciate your inputs here

Looking forward to your insights and recommendations, especially around the doclinks configuration and storage setup. Any guidance based on prior implementations or best practices would be highly appreciated. Thanks in advance

Ivan and Witold thank you so much for your inputs.

@Witold Wierzchowski You mentioned that since my DB will reside outside and attachments are being handles in GCS I may not require RWX capable storage except for the case of jms related filestore or anything which needs to be shared by the worker nodes pods. However I was reading below article

https://cloud.redhat.com/experts/gcp/mas/

which is MAS installation on managed Openshift service by GCP where they seem to have provisioned a filestore storage class and using it for UDS_STORAGE_CLASS, DRO_STORAGE_CLASS  and couple more which are internal components of MAS. This had me worried because I was not expecting to see RWX storage to be a pre-requisite for MAS installation. As you also mentioned that in my case I most probably don't need it but the article above says otherwise.

Maybe I have drawn wrong conclusions from the same, would really like to know your thoughts on this

Thanks Again for your time and inputs

Abhinav

Hi Abhinav,

seems like You are confusing primary disk (with ephemeral storage and CoreOS) with persistent storage (used by pods). Primary disks are fine using any kind of block storage, so standard-pd should be fine. Just keep in mind that master nodes require very fast disks to operate properly. See: https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/scalability_and_performance/recommended-performance-and-scalability-practices-2#recommended-etcd-practices

As for the persistent storage: GCP has native Filestore driver which You can use for RWX capable storage. If You use a CSI driver and a StorageClass then PVCs will create a share (using Filestore driver) in GCP and bind it to the PVC dynamically. No need to create those disks upfront. See the table here: https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/storage/understanding-persistent-storage#pv-access-modes_understanding-persistent-storage

In Your case with DB outside of cluster and attachments in Obejct Storage You might not need RWX capable storage, but that would depend on the actual configuration of MAS apps. For example any PVC used by Manage (maybe for migration manager or jms backing store) should be RWX (in case of non SNO setup) as it can be mounted by all ServerBundle pods and maxinst pod, which may run in different nodes.

Hi All, we are planning to install an self-managed OCP cluster on GCP cloud. Had couple of questions regarding the same

1) Does manage support GCP cloud storage for doclinks. GCP does offer a S3 compatiable API just wanted to confirm if someone has configured the same for doclinks using the system properties mxe.cosaccesskey, mxe.cossecretkey etc

2) Also while sizing the infra, the calculator recommended 3 master and 3 worker nodes wherein each worker is suggested to have 400GB of storage. We are currently planning to go with standard-pd (an ssd) which will be mounted to each worker node. Checking online we could see that a storage type of RWX (read write many) is required for mas installation  in a cluster and standard-pd does not support RWX, can anyone confirm on this whether worker node storage itself needs to be changed to a RWX compatible storage? Or if the nodes need additional RWX storage to be provisioned? Just to add the database will reside outside the cluster and doclinks will be handled in GCP object storage preferably so not sure which components in the cluster itself require RWX storage?

@Ivan Lagunov would appreciate your inputs here

Looking forward to your insights and recommendations, especially around the doclinks configuration and storage setup. Any guidance based on prior implementations or best practices would be highly appreciated. Thanks in advance