Known issues

This page contains a list of known issues that may be observed when using Trident.

  • An upstream Kubernetes bug that could be encountered when rapidly attaching/detaching volumes has been fixed with Trident v20.07.1 and the v2.0.1 release of the CSI external-provisioner sidecar. Trident v20.07.1 will use the 2.0.1 release of the external-provisioner sidecar for all Kubernetes clusters running 1.17 and above. If using Kubernetes 1.16 or below, you must upgrade your Kubernetes cluster to 1.17 or above for Trident v20.07.1 to fix this issue.
  • A previously identified issue with updating the storage prefix for a Trident backend has been resolved with Trident v20.07.1. Users can work with backends that use an empty storage prefix ("") or one that includes "-".
  • With Trident v20.07.1 using the v2.0.1 release of the CSI external-provisioner sidecar, customers can set a blank fsType on the storageClass. For Kubernetes clusters running 1.17 or above, upgrading to Trident 20.07.1 would enable users to override the default ext4 fsType. Setting the fsType blank (“”) for NAS volumes will make Trident set fsType=nas on all created volumes.
  • Trident has continually improved the resiliency for iSCSI volumes. The v20.07.0 release implements multiple additional checks that eliminate accidental deletions from occuring.
  • Trident may create multiple unintended secrets when it attempts to patch Trident service accounts. This is a bug and is fixed with v20.07.1.
  • When installing Trident (using tridentctl or the Trident Operator) and using tridentctl to manage Trident, you must ensure the KUBECONFIG environment variable is set. This is necessary to indicate the Kubernetes cluster that tridentctl must work against. When working with multiple Kubernetes environments, care must be taken to ensure the KUBECONFIG file is sourced accurately.
  • To perform online space reclamation for iSCSI PVs, the underlying OS on the worker node may require mount options to be passed to the volume. This is true for RHEL/RedHat CoreOS instances, which require the discard mount option; ensure the discard mountOption is included in your StorageClass to support online block discard.
  • Although we provide a deployment for Trident, it should never be scaled beyond a single replica. Similarly, only one instance of Trident should be run per Kubernetes cluster. Trident cannot communicate with other instances and cannot discover other volumes that they have created, which will lead to unexpected and incorrect behavior if more than one instance runs within a cluster.
  • If Trident-based StorageClass objects are deleted from Kubernetes while Trident is offline, Trident will not remove the corresponding storage classes from its database when it comes back online. Any such storage classes must be deleted manually using tridentctl or the REST API.
  • If a user deletes a PV provisioned by Trident before deleting the corresponding PVC, Trident will not automatically delete the backing volume. In this case, the user must remove the volume manually via tridentctl or the REST API.
  • When using a backend across multiple Trident instances, it is recommended that each backend configuration file specify a different storagePrefix value for ONTAP backends or use a different TenantName for Element backends. Trident cannot detect volumes that other instances of Trident have created, and attempting to create an existing volume on either ONTAP or Element backends succeeds as Trident treats volume creation as an idempotent operation. Thus, if the storagePrefix or TenantName does not differ, there is a very slim chance to have name collisions for volumes created on the same backend.
  • ONTAP cannot concurrently provision more than one FlexGroup at a time unless the set of aggregates are unique to each provisioning request.
  • When using Trident over IPv6, the managementLIF and dataLIF in the backend definition must be specified within square brackets, like [fd20:8b1e:b258:2000:f816:3eff:feec:0].
  • If using the solidfire-san driver with OpenShift 4.5, make sure the underlying worker nodes use MD5 as the CHAP authentication algorithm. Refer to Worker node Preparation for instructions.