The sds-replicated-volume module: FAQ

What is difference between LVM and LVMThin?

• LVM is simpler and has high performance that is similar to that of native disk drives, but it does not support snapshots;
• LVMThin allows for snapshots and overprovisioning; however, it is slower than LVM.

How do I get info about the space used?

There are two options:

1. Using the Grafana dashboard:

• Navigate to Dashboards –> Storage –> LINSTOR/DRBD
  In the upper right corner, you’ll see the amount of space used in the cluster.

  Caution! Raw space usage in the cluster is displayed. Suppose you create a volume with two replicas. In this case, these values must be divided by two to see how many such volumes can be in your cluster.

2. Using the LINSTOR command line:

kubectl exec -n d8-sds-replicated-volume deploy/linstor-controller -- linstor storage-pool list

Caution! Raw space usage for each node in the cluster is displayed. Suppose you create a volume with two replicas. In this case, those two replicas must fully fit on two nodes in your cluster.

How do I set the default StorageClass?

Set the spec.IsDefault field to true in the corresponding ReplicatedStorageClass custom resource.

How do I add the existing LVM Volume Group or LVMThin pool?

1. Manually add the storage.deckhouse.io/enabled=true LVM tag to the Volume Group:

   vgchange myvg-0 --add-tag storage.deckhouse.io/enabled=true
   

   This VG will be automatically discovered and a corresponding LVMVolumeGroup resource will be created in the cluster for it.

2. Specify this resource in the ReplicatedStoragePool parameters in the spec.lvmVolumeGroups[].name field (note that for the LVMThin pool, you must additionally specify its name in spec.lvmVolumeGroups[].thinPoolName).

How do I evict DRBD resources from a node?

1. Upload the evict.sh script to the host that has administrative access to the Kubernetes API server (for the script to work, you need to have kubectl and jq installed):

kubectl -n d8-sds-replicated-volume cp -c sds-replicated-volume-controller $(kubectl -n d8-sds-replicated-volume get po -l app=sds-replicated-volume-controller -o jsonpath='{.items[0].metadata.name}'):/tools/evict.sh ./evict.sh
chmod 700 evict.sh

2. Fix all faulty LINSTOR resources in the cluster. Run the following command to filter them:

kubectl -n d8-sds-replicated-volume exec -ti deploy/linstor-controller -- linstor resource list --faulty

3. Check that all the pods in the d8-sds-replicated-volume namespace are in the Running state:

kubectl -n d8-sds-replicated-volume get pods | grep -v Running

How do I evict DRBD resources from a node without deleting it from LINSTOR and Kubernetes

1. Run the evict.sh script in interactive mode (--delete-resources-only):

./evict.sh --delete-resources-only

To run the evict.sh script in non-interactive mode, add the --non-interactive flag followed by the name of the node to evict the resources from. In this mode, the script will perform all the necessary actions automatically (no user confirmation is required). Example:

./evict.sh --non-interactive --delete-resources-only --node-name "worker-1"

Caution! After the script finishes its job, the node will still be in the Kubernetes cluster albeit in SchedulingDisabled status. In LINSTOR, the AutoplaceTarget=false property will be set for this node, preventing the LINSTOR scheduler from creating resources on this node.

2. Run the following command to allow DRBD resources and pods to be scheduled on the node again:

alias linstor='kubectl -n d8-sds-replicated-volume exec -ti deploy/linstor-controller -- linstor'
linstor node set-property "worker-1" AutoplaceTarget
kubectl uncordon "worker-1"

3. Run the following command to check the AutoplaceTarget property for all nodes (the AutoplaceTarget field will be empty for nodes that are allowed to host LINSTOR resources):

alias linstor='kubectl -n d8-sds-replicated-volume exec -ti deploy/linstor-controller -- linstor'
linstor node list -s AutoplaceTarget

Troubleshooting

Problems can occur at different levels of component operation. This cheat sheet will help you quickly navigate through the diagnosis of various problems with the LINSTOR-created volumes:

Some common problems are described below.

linstor-node fail to start because the drbd module cannot be loaded

1. Check the status of the linstor-node pods:

kubectl get pod -n d8-sds-replicated-volume -l app=linstor-node

2. If some of those pods got stuck in Init state, check the DRBD version as well as the bashible logs on the node:

cat /proc/drbd
journalctl -fu bashible

The most likely reasons why bashible is unable to load the kernel module:

• You have the in-tree version of the DRBDv8 module preloaded, whereas LINSTOR requires DRBDv9. Verify the preloaded module version using the following command: cat /proc/drbd. If the file is missing, then the module is not preloaded and this is not your case.

• You have Secure Boot enabled. Since the DRBD module is compiled dynamically for your kernel (similar to dkms), it is not digitally signed. Currently, running the DRBD module with a Secure Boot enabled is not supported.

The Pod cannot start due to the FailedMount error

The Pod is stuck at the ContainerCreating phase

If the Pod is stuck at the ContainerCreating phase, and if the errors like those shown below are displayed when the kubectl describe pod command is invoked, then it means that the device is mounted on one of the nodes.

rpc error: code = Internal desc = NodePublishVolume failed for pvc-b3e51b8a-9733-4d9a-bf34-84e0fee3168d: checking
for exclusive open failed: wrong medium type, check device health

Use the command below to see if this is the case:

alias linstor='kubectl -n d8-sds-replicated-volume exec -ti deploy/linstor-controller -- linstor'
linstor resource list -r pvc-b3e51b8a-9733-4d9a-bf34-84e0fee3168d

The InUse flag will point to the node on which the device is being used. You will need to manually unmount the disk on that node.

Errors of the Input/output error type

These errors usually occur during the file system (mkfs) creation.

Check dmesg on the node where the pod is being run:

dmesg | grep 'Remote failed to finish a request within'

If the command output is not empty (the dmesg output contains lines like “Remote failed to finish a request within … “), most likely your disk subsystem is too slow for DRBD to run properly.

I have deleted the ReplicatedStoragePool resource, yet its associated Storage Pool in the LINSTOR backend is still there. Is it supposed to be like this?

Yes, this is the expected behavior. Currently, the sds-replicated-volume module does not process operations when deleting the ReplicatedStoragePool resource.

I am unable to update the fields in the ReplicatedStorageClass resource spec. Is this the expected behavior?

Yes, this is the expected behavior. Only the isDefault field is editable in the spec. All the other fields in the resource spec are made immutable.

When you delete a ReplicatedStorageClass resource, its child StorageClass in Kubernetes is not deleted. What can I do in this case?

The child StorageClass is only deleted if the status of the ReplicatedStorageClass resource is Created. Otherwise, you will need to either restore the ReplicatedStorageClass resource to a working state or delete the StorageClass yourself.

I noticed that an error occurred when trying to create a Storage Pool / Storage Class, but in the end the necessary entity was successfully created. Is this behavior acceptable?

This is the expected behavior. The module will automatically retry the unsuccessful operation if the error was caused by circumstances beyond the module’s control (for example, a momentary disruption in the Kubernetes API).

When running commands in the LINSTOR CLI, I get the “You’re not allowed to change state of linstor cluster manually. Please contact tech support” error. What to do?

In the sds-replicated-volume module, we have restricted the list of commands that are allowed to be run in LINSTOR, because we plan to automate all manual operations. Some of them are already automated, e.g., creating a Tie-Breaker in cases when LINSTOR doesn’t create them for resources with 2 replicas. Use the command below to see the list of allowed commands:

alias linstor='kubectl -n d8-sds-replicated-volume exec -ti deploy/linstor-controller -- linstor'
linstor --help

Service pods of sds-replicated-volume components fail to be created on the node I need

Most likely this is due to node labels.

• Check dataNodes.nodeSelector in the module settings:

kubectl get mc sds-replicated-volume -o=jsonpath={.spec.settings.dataNodes.nodeSelector}

• Check the selectors that sds-replicated-volume-controller uses:

kubectl -n d8-sds-replicated-volume get secret d8-sds-replicated-volume-controller-config  -o jsonpath='{.data.config}' | base64 --decode

• The d8-sds-replicated-volume-controller-config secret should contain the selectors that are specified in the module settings, as well as the kubernetes.io/os: linux selector.

• Make sure that the target node has all the labels specified in the d8-sds-replicated-volume-controller-config secret:

kubectl get node worker-0 --show-labels

• If there are no labels, add them to the NodeGroup or to the node via templates.

• If there are labels, check if the target node has the storage.deckhouse.io/sds-replicated-volume-node= label attached. If there is no label, check if the sds-replicated-volume-controller is running and if it is running, examine its logs:

kubectl -n d8-sds-replicated-volume get po -l app=sds-replicated-volume-controller
kubectl -n d8-sds-replicated-volume logs -l app=sds-replicated-volume-controller

I have not found an answer to my question and am having trouble getting the module to work. What do I do?

Information about the reasons for the failure is saved to the Status.Reason field of the ReplicatedStoragePool and ReplicatedStorageClass resources. If the information provided is not enough to identify the problem, refer to the sds-replicated-volume-controller logs.

Migrating from the Deckhouse Kubernetes Platform linstor built-in module to sds-replicated-volume

Note that the LINSTOR control-plane and its CSI will be unavailable during the migration process. This will make it impossible to create/expand/delete PVs and create/delete pods using the LINSTOR PV during the migration.

Please note! User data will not be affected by the migration. Basically, the migration to a new namespace will take place. Also, new components will be added (in the future, they will take over all LINSTOR volume management functionality).

Migration steps

1. Make sure there are no faulty LINSTOR resources in the cluster. The command below should return an empty list:

alias linstor='kubectl -n d8-linstor exec -ti deploy/linstor-controller -- linstor'
linstor resource list --faulty

Caution! You should fix all LINSTOR resources before migrating.

2. Disable the linstor module:

kubectl patch moduleconfig linstor --type=merge -p '{"spec": {"enabled": false}}'

3. Wait for the d8-linstor namespace to be deleted.

kubectl get namespace d8-linstor

4. Create a ModuleConfig resource for sds-node-configurator.

kubectl apply -f -<<EOF
apiVersion: deckhouse.io/v1alpha1
kind: ModuleConfig
metadata:
  name: sds-node-configurator
spec:
  enabled: true
  version: 1
EOF

5. Wait for the sds-node-configurator module to become Ready.

kubectl get moduleconfig sds-node-configurator

6. Create a ModuleConfig resource for sds-replicated-volume.

Caution! Failing to specify the settings.dataNodes.nodeSelector parameter in the sds-replicated-volume module settings would result in the value for this parameter to be derived from the linstor module when installing the sds-replicated-volume module. If this parameter is not defined there as well, it will remain empty and all the nodes in the cluster will be treated as storage nodes.

kubectl apply -f - <<EOF
apiVersion: deckhouse.io/v1alpha1
kind: ModuleConfig
metadata:
  name: sds-replicated-volume
spec:
  enabled: true
  version: 1
EOF

7. Wait for the sds-replicated-volume module to become Ready.

kubectl get moduleconfig sds-replicated-volume

8. Check the sds-replicated-volume module settings.

kubectl get moduleconfig sds-replicated-volume -oyaml

9. Wait for all pods in the d8-sds-replicated-volume and d8-sds-node-configurator namespaces to become Ready or Completed.

kubectl get po -n d8-sds-node-configurator
kubectl get po -n d8-sds-replicated-volume

10. Override the linstor command alias and check the LINSTOR resources:

alias linstor='kubectl -n d8-sds-replicated-volume exec -ti deploy/linstor-controller -- linstor'
linstor resource list --faulty

If there are no faulty resources, then the migration was successful.

Migrating to ReplicatedStorageClass

Note that StorageClasses in this module are managed via the ReplicatedStorageClass resource. StorageClasses should not be created manually.

When migrating from the linstor module, delete old StorageClasses and create new ones via the ReplicatedStorageClass resource (refer to the table below).

Note that in the old StorageClasses, you should pick up the option from the parameter section of the StorageClass itself, while for the new StorageClass, you should specify the corresponding option in ReplicatedStorageClass.

On top of these, the following parameters are available:

• reclaimPolicy (Delete, Retain) - corresponds to the reclaimPolicy parameter of the old StorageClass
• zones - list of zones to be used for hosting resources ( the actual names of the zones in the cloud). Please note that remote access to a volume with data is only possible within a single zone!
• volumeAccess can be “Local” (access is strictly within the node), “EventuallyLocal” (the data replica will be synchronized on the node with the running pod a while after the start), “PreferablyLocal” (remote access to the volume with data is allowed, volumeBindingMode: WaitForFirstConsumer), “Any” (remote access to the volume with data is allowed, volumeBindingMode: Immediate).
• If you need to use volumeBindingMode: Immediate, set the volumeAccess parameter of the ReplicatedStorageClass to Any.

You can read more about working with ReplicatedStorageClass resources here.

Migrating to ReplicatedStoragePool

The ReplicatedStoragePool resource allows you to create a Storage Pool in LINSTOR. It is recommended to create this resource for the Storage Pools that already exist in LINSTOR and specify the existing LVMVolumeGroups in this resource. In this case, the controller will see that the corresponding Storage Pool has been created and leave it unchanged, while the status.phase field of the created resource will be set to Created. Refer to the sds-node-configurator documentation to learn more about LVMVolumeGroup resources. To learn more about working with ReplicatedStoragePool resources, click here.

Migrating from sds-drbd module to sds-replicated-volume

Note that the module control-plane and its CSI will be unavailable during the migration process. This will make it impossible to create/expand/delete PVs and create/delete pods using DRBD PV during the migration.

Please note! User data will not be affected by the migration. Basically, the migration to a new namespace will take place. Also, new components will be added (in the future, they will take over all module volume management functionality).

Migration steps

1. Make sure there are no faulty DRBD resources in the cluster. The command below should return an empty list:

alias linstor='kubectl -n d8-sds-drbd exec -ti deploy/linstor-controller -- linstor'
linstor resource list --faulty

Caution! You should fix all DRBD resources before migrating.

2. Disable the sds-drbd module:

kubectl patch moduleconfig sds-drbd --type=merge -p '{"spec": {"enabled": false}}'

3. Wait for the d8-sds-drbd namespace to be deleted.

kubectl get namespace d8-sds-drbd

4. Create a ModuleConfig resource for sds-replicated-volume.

Caution! Failing to specify the settings.dataNodes.nodeSelector parameter in the sds-replicated-volume module settings would result in the value for this parameter to be derived from the sds-drbd module when installing the sds-replicated-volume module. If this parameter is not defined there as well, it will remain empty and all the nodes in the cluster will be treated as storage nodes.

kubectl apply -f - <<EOF
apiVersion: deckhouse.io/v1alpha1
kind: ModuleConfig
metadata:
  name: sds-replicated-volume
spec:
  enabled: true
  version: 1
EOF

5. Wait for the sds-replicated-volume module to become Ready.

kubectl get moduleconfig sds-replicated-volume

6. Check the sds-replicated-volume module settings.

kubectl get moduleconfig sds-replicated-volume -oyaml

7. Wait for all pods in the d8-sds-replicated-volume namespaces to become Ready or Completed.

kubectl get po -n d8-sds-replicated-volume

8. Override the linstor command alias and check the DRBD resources:

alias linstor='kubectl -n d8-sds-replicated-volume exec -ti deploy/linstor-controller -- linstor'
linstor resource list --faulty

If there are no faulty resources, then the migration was successful.

Caution! The resources DRBDStoragePool and DRBDStorageClass will be automatically migrated to ReplicatedStoragePool and ReplicatedStorageClass during the process, no user intervention is required for this. The functionality of these resources will not change. However, it is worth checking if there are any DRBDStoragePool or DRBDStorageClass left in cluster. If they exist after the migration, please inform our support team.

Why is it not recommended to use RAID for disks that are used by the sds-replicated-volume module?

DRBD with a replica count greater than 1 provides de facto network RAID. Using RAID locally may be inefficient because:

• Redundant RAID dramatically increases the overhead in terms of space utilization. Here is an example: Suppose, a DBRDStorageClass is used with replication set to ConsistencyAndAvailability. With this setting, DRBD will store data in three replicas (one replica per three different hosts). If RAID1 is used on these hosts, a total of 6 GB of disk space will be required to store 1 GB of data. Redundant RAID is worth using for easier server maintenance when the storage costs are irrelevant. RAID1 in this case will allow you to change disks on servers without having to move data replicas from the “problem” disk.

• As for RAID0, the performance gain will be unnoticeable, since data replication will be performed over the network and the network is likely to be the bottleneck. On top of that, decreased storage reliability on the host will potentially lead to data unavailability given that in DRBD, switching from a faulty replica to a healthy one is not instantaneous.

Why do you recommend using local disks (and not NAS)?

Motivation: DRBD replicates data over the network. If NAS is used, the network load will increase as well as the read/write latency.