selee
/
GPU-Live
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
GPU-Live/template/stg-server/ddn/exa-csi-driver/README.md

550 lines
22 KiB
Markdown
Raw Blame History

# Exascaler-csi-file-driver
Releases can be found here - https://github.com/DDNStorage/exa-csi-driver/releases
## Feature List
|Feature|Feature Status|CSI Driver Version|CSI Spec Version|Kubernetes Version|
|--- |--- |--- |--- |--- |
|Static Provisioning|GA|>= 1.0.0|>= 1.0.0|>=1.18|
|Dynamic Provisioning|GA|>= 1.0.0|>= 1.0.0|>=1.18|
|RW mode|GA|>= 1.0.0|>= 1.0.0|>=1.18|
|RO mode|GA|>= 1.0.0|>= 1.0.0|>=1.18|
|Expand volume|GA|>= 1.0.0|>= 1.1.0|>=1.18|
|StorageClass Secrets|GA|>= 1.0.0|>=1.0.0|>=1.18|
|Mount options|GA|>= 1.0.0|>= 1.0.0|>=1.18|
|Topology|GA|>= 2.0.0|>= 1.0.0|>=1.17|
## Access Modes support
|Access mode| Supported in version|
|--- |--- |
|ReadWriteOnce| >=1.0.0 |
|ReadOnlyMany| >=2.2.3 |
|ReadWriteMany| >=1.0.0 |
|ReadWriteOncePod| >=2.2.3 |
## Openshift Certification
|Openshift Version| CSI driver Version| EXA Version|
|---|---|---|
|v4.13| >=v2.2.3|v6.3.0|
|v4.14| >=v2.2.4|v6.3.0|
|v4.15| >=v2.2.4|v6.3.0|
## Requirements
- Kubernetes cluster must allow privileged pods, this flag must be set for the API server and the kubelet
([instructions](https://github.com/kubernetes-csi/docs/blob/735f1ef4adfcb157afce47c64d750b71012c8151/book/src/Setup.md#enable-privileged-pods)):
```
--allow-privileged=true
```
- Required API server and kubelet feature gates for k8s version < 1.16 (skip this step for k8s >= 1.16)
([instructions](https://github.com/kubernetes-csi/docs/blob/735f1ef4adfcb157afce47c64d750b71012c8151/book/src/Setup.md#enabling-features)):
```
--feature-gates=ExpandInUsePersistentVolumes=true,ExpandCSIVolumes=true,ExpandPersistentVolumes=true
```
- Mount propagation must be enabled, the Docker daemon for the cluster must allow shared mounts
([instructions](https://github.com/kubernetes-csi/docs/blob/735f1ef4adfcb157afce47c64d750b71012c8151/book/src/Setup.md#enabling-mount-propagation))
## Installation
Clone or untar driver (depending on where you get the driver)
```bash
git clone -b <driver version> https://github.com/DDNStorage/exa-csi-driver.git /opt/exascaler-csi-file-driver
```
e.g:-
```bash
git clone -b 2.2.4 https://github.com/DDNStorage/exa-csi-driver.git /opt/exascaler-csi-file-driver
```
or
```bash
rpm -Uvh exa-csi-driver-1.0-1.el7.x86_64.rpm
```
## Openshift
### Prerequisites
Internal OpenShift image registry needs to be patched to allow building lustre modules with KMM.
```bash
oc patch configs.imageregistry.operator.openshift.io cluster --type merge --patch '{"spec":{"managementState":"Managed"}}'
oc patch configs.imageregistry.operator.openshift.io cluster --type merge --patch '{"spec":{"storage":{"emptyDir":{}}}}'
oc patch configs.imageregistry.operator.openshift.io/cluster --type merge -p '{"spec":{"defaultRoute":true}}'
```
### Building lustre rpms
You will need a vm with the kernel version matching that of the Openshift nodes. To check on nodes:
```bash
oc get nodes
oc debug node/c1-pk6k4-worker-0-2j6w8
uname -r
```
On the builder vm, install the matching kernel and log in to it, example for Rhel 9.2:
```bash
# login to subscription-manager
subscription-manager register --username <username> --password <password> --auto-attach
# list available kernels
yum --showduplicates list available kernel
yum install kernel-<version>-<revision>.<arch> # e.g.: yum install kernel-5.14.0-284.25.1.el9_2.x86_64
grubby --info=ALL | grep title
title="Red Hat Enterprise Linux (5.14.0-284.11.1.el9_2.x86_64) 9.2 (Plow)" <---- 0
title="Red Hat Enterprise Linux (5.14.0-284.25.1.el9_2.x86_64) 9.2 (Plow)" <---- 1
grub2-set-default 1
reboot
```
Copy EXAScaler client tar from the EXAScaler server:
```bash
scp root@<exa-server-ip>:/scratch/EXAScaler-<version>/exa-client-<version>.tar.gz .
tar -xvf exa-client-<version>.tar.gz
cd exa-client
./exa_client_deploy.py -i
```
This will build the rpms and install the client.
Upload the rpms to any repository available from the cluster and change deploy/openshift/lustre-module/lustre-dockerfile-configmap.yaml lines 12-13 accordingly.
```
RUN git clone https://github.com/Qeas/rpms.git # change this to your repo with matching rpms
RUN yum -y install rpms/*.rpm
```
### Loading lustre modules in Openshift
Before loading the lustre modules, make sure to install Openshift Kernel Module Management (KMM) via Openshift console.
```bash
oc create -n openshift-kmm -f deploy/openshift/lustre-module/lustre-dockerfile-configmap.yaml
oc apply -n openshift-kmm -f deploy/openshift/lustre-module/lnet-mod.yaml
```
Wait for the builder pod (e.g. `lnet-build-5f265-build`) to finish. After builder finishes,
you should have `lnet-8b72w-6fjwh` running on each worker node.
```bash
# run ko2iblnd-mod if you are using Infiniband network
oc apply -n openshift-kmm -f deploy/openshift/lustre-module/ko2iblnd-mod.yaml
```
Make changes to `deploy/openshift/lustre-module/lnet-configuration-ds.yaml` line 38 according to the cluster's network
```
lnetctl net add --net tcp --if br-ex # change interface according to your cluster
```
Configure lnet and install lustre
```
oc apply -n openshift-kmm -f deploy/openshift/lustre-module/lnet-configuration-ds.yaml
oc apply -n openshift-kmm -f deploy/openshift/lustre-module/lustre-mod.yaml
```
### Installing the driver
Make sure that `openshift: true` in `deploy/openshift/exascaler-csi-file-driver-config.yaml`.
Create a secret from the config file and apply the driver yaml.
```bash
oc create -n openshift-kmm secret generic exascaler-csi-file-driver-config --from-file=deploy/openshift/exascaler-csi-file-driver-config.yaml
oc apply -n openshift-kmm -f deploy/openshift/exascaler-csi-file-driver.yaml
```
### Uninstall
```bash
oc delete -n openshift-kmm secret exascaler-csi-file-driver-config
oc delete -n openshift-kmm -f deploy/openshift/exascaler-csi-file-driver.yaml
oc delete -n openshift-kmm -f deploy/openshift/lustre-module/lustre-mod.yaml
oc delete -n openshift-kmm -f deploy/openshift/lustre-module/lnet-configuration-ds.yaml
oc delete -n openshift-kmm -f deploy/openshift/lustre-module/ko2iblnd-mod.yaml
oc delete -n openshift-kmm -f deploy/openshift/lustre-module/lnet-mod.yaml
oc delete -n openshift-kmm -f deploy/openshift/lustre-module/lustre-dockerfile-configmap.yaml
oc get images | grep lustre-client-moduleloader | awk '{print $1}' | xargs oc delete image
```
## Kubernetes
### Using helm chart
Make changes to `deploy/helm-chart/values.yaml` and `deploy/helm-chart/exascaler-csi-file-driver-config.yaml` according to your Kubernetes and EXAScaler clusters environment
`helm install -n ${namespace} exascaler-csi-file-driver deploy/helm-chart/`
#### Uninstall
`helm uninstall exascaler-csi-file-driver`
#### Upgrade
1. Make any necessary changes to the chart, for example new driver version: `tag: "v2.2.5"` in `deploy/helm-chart/values.yaml`.
2. Run `helm upgrade -n ${namespace} exascaler-csi-file-driver deploy/helm-chart/`
### Using docker load and kubectl commands
```bash
docker load -i /opt/exascaler-csi-file-driver/bin/exascaler-csi-file-driver.tar
```
#### Verify version
kubectl get deploy/exascaler-csi-controller -o jsonpath="{..image}"
2. Copy /opt/exascaler-csi-file-driver/deploy/kubernetes/exascaler-csi-file-driver-config.yaml to /etc/exascaler-csi-file-driver-v1.0/exascaler-csi-file-driver-config.yaml
```
cp /opt/exascaler-csi-file-driver/deploy/kubernetes/exascaler-csi-file-driver-config.yaml /etc/exascaler-csi-file-driver-v1.0/exascaler-csi-file-driver-config.yaml
```
Edit `/etc/exascaler-csi-file-driver-v1.0/exascaler-csi-file-driver-config.yaml` file. Driver configuration example:
```yaml
exascaler_map:
exa1:
mountPoint: /exaFS # mountpoint on the host where the exaFS will be mounted
exaFS: 192.168.88.114@tcp2:192.168.98.114@tcp2:/testfs # default path to exa filesystem where the PVCs will be stored
managementIp: 10.204.86.114@tcp # network for management operations, such as create/delete volume
zone: zone-1
exa2:
mountPoint: /exaFS-zone-2 # mountpoint on the host where the exaFS will be mounted
exaFS: 192.168.78.112@tcp2:/testfs/zone-2 # default path to exa filesystem where the PVCs will be stored
managementIp: 10.204.86.114@tcp # network for management operations, such as create/delete volume
zone: zone-2
exa3:
mountPoint: /exaFS-zone-3 # mountpoint on the host where the exaFS will be mounted
exaFS: 192.168.98.113@tcp2:192.168.88.113@tcp2:/testfs/zone-3 # default path to exa filesystem where the PVCs will be stored
managementIp: 10.204.86.114@tcp # network for management operations, such as create/delete volume
zone: zone-3
debug: true
```
3. Create Kubernetes secret from the file:
```bash
kubectl create secret generic exascaler-csi-file-driver-config --from-file=/etc/exascaler-csi-file-driver-v1.0/exascaler-csi-file-driver-config.yaml
```
4. Register driver to Kubernetes:
```bash
kubectl apply -f /opt/exascaler-csi-file-driver/deploy/kubernetes/exascaler-csi-file-driver.yaml
```
## Usage
### Dynamically provisioned volumes
For dynamic volume provisioning, the administrator needs to set up a _StorageClass_ in the PV yaml (/opt/exascaler-csi-file-driver/examples/exa-dynamic-nginx.yaml for this example) pointing to the driver.
For dynamically provisioned volumes, Kubernetes generates volume name automatically (for example `pvc-ns-cfc67950-fe3c-11e8-a3ca-005056b857f8-projectId-1001`).
```yaml
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: exascaler-csi-file-driver-sc-nginx-dynamic
provisioner: exa.csi.ddn.com
allowedTopologies:
- matchLabelExpressions:
- key: topology.exa.csi.ddn.com/zone
values:
- zone-1
mountOptions: # list of options for `mount -o ...` command
# - noatime #
parameters:
projectId: "100001" # Points to EXA project id to be used to set volume quota. Automatically generated by the driver if not provided.
exaMountUid: "1001" # Uid which will be used to access the volume in pod. Should be synced between EXA server and clients.
exaMountGid: "1002" # Gid which will be used to access the volume in pod. Should be synced between EXA server and clients.
bindMount: "false" # Determines, whether volume will bind mounted or as a separate lustre mount.
exaFS: "10.204.86.114@tcp:/testfs" # Overrides exaFS value from config. Use this to support multiple EXA filesystems.
mountPoint: /exaFS # Overrides mountPoint value from config. Use this to support multiple EXA filesystems.
mountOptions: ro,noflock
```
#### Example
Run Nginx pod with dynamically provisioned volume:
```bash
kubectl apply -f /opt/exascaler-csi-file-driver/examples/exa-dynamic-nginx.yaml
# to delete this pod:
kubectl delete -f /opt/exascaler-csi-file-driver/examples/exa-dynamic-nginx.yaml
```
### Static (pre-provisioned) volumes
The driver can use already existing Exasaler filesystem,
in this case, _StorageClass_, _PersistentVolume_ and _PersistentVolumeClaim_ should be configured.
#### _StorageClass_ configuration
```yaml
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: exascaler-csi-driver-sc-nginx-persistent
provisioner: exa.csi.ddn.com
allowedTopologies:
- matchLabelExpressions:
- key: topology.exa.csi.ddn.com/zone
values:
- zone-1
mountOptions: # list of options for `mount -o ...` command
# - noatime #
```
### Topology configuration
In order to configure CSI driver with kubernetes topology, use the `zone` parameter in driver config or storageClass parameters. Example config file with zones:
```bash
exascaler_map:
exa1:
exaFS: 10.3.196.24@tcp:/csi
mountPoint: /exaFS
zone: us-west
exa2:
exaFS: 10.3.196.24@tcp:/csi-2
mountPoint: /exaFS-zone2
zone: us-east
```
This will assign volumes to be created on Exascaler cluster that correspond with the zones requested by allowedTopologies values.
#### _PersistentVolume_ configuration
```yaml
apiVersion: v1
kind: PersistentVolume
metadata:
name: exascaler-csi-driver-pv-nginx-persistent
labels:
name: exascaler-csi-driver-pv-nginx-persistent
spec:
storageClassName: exascaler-csi-driver-sc-nginx-persistent
accessModes:
- ReadWriteMany
capacity:
storage: 1Gi
csi:
driver: exa.csi.ddn.com
volumeHandle: exa1:/exaFS/nginx-persistent
volumeAttributes: # volumeAttributes are the alternative of storageClass params for static (precreated) volumes.
exaMountUid: "1001" # Uid which will be used to access the volume in pod.
#mountOptions: ro, flock # list of options for `mount` command
```
CSI Parameters:
| Name | Description | Example |
|----------------|-------------------------------------------------------------------|--------------------------------------|
| `driver` | [required] installed driver name " exa.csi.ddn.com" | `exa.csi.ddn.com` |
| `volumeHandle` | [required] EXAScaler name and path to volume on EXAScaler filesystem | `exa1:/nginx-persistent` |
| `exaMountUid` | Uid which will be used to access the volume from the pod. | `1015` |
| `exaMountGid` | Gid which will be used to access the volume from the pod. | `1015` |
| `projectId` | Points to EXA project id to be used to set volume quota. Automatically generated by the driver if not provided. | `100001` |
| `managementIp` | Should be used if there is a separate network configured for management operations, such as create/delete volumes. This network should have access to all Exa filesystems in a isolated zones environment | `192.168.10.20@tcp2` |
| `bindMount` | Determines, whether volume will bind mounted or as a separate lustre mount. | `true` |
| `mountOptions` | Options that will be passed to mount command (-o <opt1,opt2,opt3>) | `ro,flock` |
| `minProjectId` | Minimum project ID number for automatic generation. Only used when projectId is not provided. | 10000 |
| `maxProjectId` | Maximum project ID number for automatic generation. Only used when projectId is not provided. | 4294967295 |
| `generateProjectIdRetries` | Maximum retry count for generating random project ID. Only used when projectId is not provided. | `5` |
| `zone` | Topology zone to control where the volume should be created. Should match topology.exa.csi.ddn.com/zone label on node(s). | `us-west` |
| `v1xCompatible` | [Optional] Only used when upgrading the driver from v1.x.x to v2.x.x. Provides compatibility for volumes that were created beore the upgrade. Set it to `true` to point to the Exa cluster that was configured before the upgrade | `false` |
| `tempMountPoint` | [Optional] Used when `exaFS` points to a subdirectory that does not exist on Exascaler and will be automatically created by the driver. This parameter sets the directory where Exascaler filesystem will be temporarily mounted to create the subdirectory. | `/tmp/exafs-mnt` |
| `volumeDirPermissions` | [Optional] Defines file permissions for mounted volumes. | `0777` |
#### _PersistentVolumeClaim_ (pointing to created _PersistentVolume_)
```yaml
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: exascaler-csi-driver-pvc-nginx-persistent
spec:
storageClassName: exascaler-csi-driver-cs-nginx-persistent
accessModes:
- ReadWriteMany
resources:
requests:
storage: 1Gi
selector:
matchLabels:
# to create 1-1 relationship for pod - persistent volume use unique labels
name: exascaler-csi-file-driver-pv-nginx-persistent
```
#### Example
Run nginx server using PersistentVolume.
**Note:** Pre-configured filesystem should exist on the EXAScaler:
`/exaFS/nginx-persistent`.
```bash
kubectl apply -f /opt/exascaler-csi-file-driver/examples/nginx-persistent-volume.yaml
# to delete this pod:
kubectl delete -f /opt/exascaler-csi-file-driver/examples/nginx-persistent-volume.yaml
```
## Snapshots
To use CSI snapshots, the snapshot CRDs along with the csi-snapshotter must be installed.
```bash
kubectl apply -f deploy/kubernetes/snapshots/
```
After that the snapshot class for EXA CSI must be created
Snapshot parameter can be passed through snapshot class parameters as can be seen is `examples/snapshot-class.yaml`
List of available snapshot parameters:
| Name | Description | Example |
|----------------|-------------------------------------------------------------------|--------------------------------------|
| `snapshotFolder` | [Optional] Folder on ExaScaler filesystem where the snapshots will be created. | `csi-snapshots` |
| `snapshotUtility` | [Optional] Either `tar` or `dtar`. `dtar` is faster but requires `dtar` to be installed on all k8s nodes. Default is `tar`| `dtar` |
| `dtarPath` | [Optional] If `snapshotUtility` is `dtar` points to where the `dtar` utility is installed | `/opt/ddn/mpifileutils/bin/dtar` |
| `snapshotMd5Verify` | [Optional] Defines whether the driver should do md5sum check on the snapshot. Ensures that the snapshot is not corrupt but reduces performance. Default is `false` | `true` |
```bash
kubectl apply -f examples/snapshot-class.yaml
```
Now a snapshot can be created, for example
```bash
kubectl apply -f examples/snapshot-from-dynamic.yaml
```
We can create a volume using the snapshot
```bash
kubectl apply -f examples/nginx-from-snapshot.yaml
```
Exascaler CSI driver supports 2 snapshot modes: `tar` or `dtar`.
Default mode is `tar`.
To enable `dtar` set `snapshotUtility: dtar` in config.
Dtar is much faster but requires mpifileutils `dtar` installed on all the nodes as a prerequisite.
Installing `dtar` might differ depending on your OS. Here is an example for Ubuntu 22
```bash
# Install dependencies mpifileutils
cd
mkdir -p /opt/ddn/mpifileutils
cd /opt/ddn/mpifileutils
sudo apt-get install -y cmake libarchive-dev libbz2-dev libcap-dev libssl-dev openmpi-bin libopenmpi-dev libattr1-dev
export INSTALL_DIR=/opt/ddn/mpifileutils
git clone https://github.com/LLNL/lwgrp.git
cd lwgrp
./autogen.sh
./configure --prefix=$INSTALL_DIR
make -j 16
make install
cd ..
git clone https://github.com/LLNL/dtcmp.git
cd dtcmp
./autogen.sh
./configure --prefix=$INSTALL_DIR --with-lwgrp=$INSTALL_DIR
make -j 16
make install
cd ..
git clone https://github.com/hpc/libcircle.git
cd libcircle
sh ./autogen.sh
./configure --prefix=$INSTALL_DIR
make -j 16
make install
cd ..
git clone https://github.com/hpc/mpifileutils.git mpifileutils-clone
cd mpifileutils-clone
cmake ./ -DWITH_DTCMP_PREFIX=$INSTALL_DIR -DWITH_LibCircle_PREFIX=$INSTALL_DIR -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR -DENABLE_LUSTRE=ON -DENABLE_XATTRS=ON
make -j 16
make install
cd ..
```
If you use a different `INSTALL_DIR` path, pass it in config using `DtarPath` parameter.
## Updating the driver version
To update to a new driver version, you need to follow the following steps:
1. Remove the old driver version
```bash
kubectl delete -f /opt/exascaler-csi-file-driver/deploy/kubernetes/exascaler-csi-file-driver.yaml
kubectl delete secrets exascaler-csi-file-driver-config
rpm -evh exa-csi-driver
```
2. Download the new driver version (git clone or new ISO)
3. Copy and edit config file
```bash
cp /opt/exascaler-csi-file-driver/deploy/kubernetes/exascaler-csi-file-driver-config.yaml /etc/exascaler-csi-file-driver-v1.1/exascaler-csi-file-driver-config.yaml
```
If you are upgrading from v1.x.x to v2.x.x, config file structure will change to a map of Exascaler clusters instead of a flat structure. To support old volume that were created using v1.x.x, old config should be put in the config map with `v1xCompatible: true`. For example:
v1.x.x config
```bash
exaFS: 10.3.196.24@tcp:/csi
mountPoint: /exaFS
debug: true
```
v2.x.x config with support of previously created volumes
```bash
exascaler_map:
exa1:
exaFS: 10.3.196.24@tcp:10.3.199.24@tcp:/csi
mountPoint: /exaFS
v1xCompatible: true
exa2:
exaFS: 10.3.1.200@tcp:10.3.2.200@tcp:10.3.3.200@tcp:/csi-fs
mountPoint: /mnt2
debug: true
```
Only one of the Exascaler clusters can have `v1xCompatible: true` since old config supported only 1 cluster.
4. Update version in /opt/exascaler-csi-file-driver/deploy/kubernetes/exascaler-csi-file-driver.yaml
```
image: exascaler-csi-file-driver:v2.2.5
```
5. Load new image
```bash
docker load -i /opt/exascaler-csi-file-driver/bin/exascaler-csi-file-driver.tar
```
6. Apply new driver
```bash
kubectl create secret generic exascaler-csi-file-driver-config --from-file=/etc/exascaler-csi-file-driver-v1.1/exascaler-csi-file-driver-config.yaml
kubectl apply -f /opt/exascaler-csi-file-driver/deploy/kubernetes/exascaler-csi-file-driver.yaml
```
7. Verify version
```bash
kubectl get deploy/exascaler-csi-controller -o jsonpath="{..image}"
```
## Troubleshooting
### Driver logs
To collect all driver related logs, you can use the `kubectl logs` command.
All in on command:
```bash
mkdir exa-csi-logs
for name in $(kubectl get pod -owide | grep exascaler | awk '{print $1}'); do kubectl logs $name --all-containers > exa-csi-logs/$name; done
```
To get logs from all containers of a single pod
```bash
kubectl logs <pod_name> --all-containers
```
Logs from a single container of a pod
```bash
kubectl logs <pod_name> -c driver
```
#### Driver secret
```bash
kubectl get secret exascaler-csi-file-driver-config -o json | jq '.data | map_values(@base64d)' > exa-csi-logs/exascaler-csi-file-driver-config
```
#### PV/PVC/Pod data
```bash
kubectl get pvc > exa-csi-logs/pvcs
kubectl get pv > exa-csi-logs/pvs
kubectl get pod > exa-csi-logs/pods
```
#### Extended info about a PV/PVC/Pod:
```bash
kubectl describe pvc <pvc_name> > exa-csi-logs/<pvc_name>
kubectl describe pv <pv_name> > exa-csi-logs/<pv_name>
kubectl describe pod <pod_name> > exa-csi-logs/<pod_name>
```
Reference in New Issue View Git Blame Copy Permalink