CSI driver Helm values

How the driver reaches the storage cluster, what the controller does in the background, and how the node plugin behaves on the NVMe-oF connection.

charts/mgx-csi-driver/values.yaml in the driver repo is the full, commented list; this page is the part you are likely to set.

#Cluster connection

Rendered into a Secret named mgxcsi-secret.

Value	Default	Description
`csiSecret.clusterConfig.protocol`	`http`	Management API protocol.
`csiSecret.clusterConfig.nodes`	`[localhost:8082]`	Management API endpoints, `host:port`.
`csiSecret.clusterConfig.cluster`	`main`	Target cluster name.
`csiSecret.clusterConfig.ns`	`main`	Target namespace within the cluster.
`csiSecret.clusterConfig.username`	`""`	API user.
`csiSecret.clusterConfig.password`	`""`	API password.

#Controller

Value	Default	Description
`controller.replicas`	`1`	Controller StatefulSet replicas.
`controller.timeoutVolumeCheck`	`2`	Minutes between reconciler scans of every PV. `0` disables the reconciler.
`controller.idleVolumeMin`	`10`	Minutes a volume may stay unattached before the reconciler idles or stops it on the backend.

#What the chart creates

Value	Default	Description
`storageclass.create`	`true`	Create the `mgxcsi-sc` StorageClass.
`volumeSnapshotClass.create`	`true`	Create the `mgxcsi-snapshotclass` VolumeSnapshotClass.
`externalSnapshotter.enabled`	`false`	Deploy a snapshot controller. Only if the cluster has none — one serves every CSI driver.
`externalSnapshotter.customResourceDefinitions.enabled`	`false`	Let the chart create the snapshot CRDs.
`serviceAccount.create`	`true`	Create the ServiceAccount.
`rbac.create`	`true`	Create the RBAC objects.

#Node NVMe-oF tuning

Defaults are sensible. Reach for these when diagnosing connection or teardown behaviour, not before.

Value	Default	Description
`node.nrIoQueues`	`2`	NVMe-oF I/O queues.
`node.queueSize`	`16`	NVMe-oF queue depth.
`node.fastIoFailTmo`	`0`	Seconds to queue I/O on a lost controller before failing fast. `0` fails immediately.
`node.ctrlLossTmo`	`10`	Seconds of reconnect attempts before the controller is removed. Must be at least `fastIoFailTmo`.
`node.reconnectDelay`	`2`	Seconds between reconnect attempts.
`node.keepAliveTmo`	`5`	Keep-alive timeout, seconds.
`node.nvmeTimeout`	`30`	Deadline for `nvme connect`/`disconnect` and the cleanup that follows, seconds.
`node.mkfsFsckTimeout`	`120`	Per-command deadline for `fsck`, `mkfs` and `mount`, seconds.

#Volume clean on unpublish

Off by default. When enabled, unpublishing a volume asks the backend to clean it and waits for READY again, which trims freed space rather than leaving it allocated.

Value	Default	Description
`node.volumeCleanEnabled`	`false`	Call `storage.volume_clean` on unpublish and wait for `READY`.
`node.volumeCleanPollInterval`	`2`	Seconds between `volume_get` probes while waiting.
`node.volumeCleanReadyTimeout`	`330`	Total budget waiting for `READY`, seconds.
`node.volumeCleanFstrimTimeout`	`300`	`fstrim` timeout forwarded to `volume_clean`, seconds.

#Driver flags

The chart starts the mgxcsi binary with --controller or --node. The flags mirror the values above; these are the ones without a value of their own:

Flag	Default	Description
`--drivername`	`csi.migrx.io`	CSI driver name, and the `provisioner` a StorageClass must name.
`--endpoint`	`unix://tmp/mgxcsi.sock`	CSI gRPC endpoint.
`--nodeid`	required for node	Node identifier.