CSV Volumes Stuck in Redirected Access After MPIO Path Failure

Question

CSV Volumes Stuck in Redirected Access After MPIO Path Failure

Manish Rahul 60

Hi everyone,

We’re seeing several Cluster Shared Volumes on a hyperconverged failover cluster switch into Redirected Access mode. Since then, storage latency has increased by roughly 300%.

The issue appears to be related to MPIO. One of the secondary SAN paths is no longer being claimed correctly, and we found what looks like a stale DSM registry entry on the affected nodes.

What’s the recommended way to completely clear the existing MPIO path claims and force Windows to rediscover/rebuild them using mpclaim?

Also, are there any specific checks we should run before bringing the CSVs back to normal direct I/O mode?

0 comments

2 answers

Your answer

Answer 1

Dear Manish Rahul,

When Cluster Shared Volumes get stuck in Redirected Access mode due to MPIO issues, the correct approach is to completely reset and rebuild the multipath claims. On the affected nodes, you must first remove all existing DSM registrations using mpclaim -r -i -d <DeviceID> or, if you want to clear everything, mpclaim -r -i -a. This removes the stale entries from the registry under HKLM\SYSTEM\CurrentControlSet\Control\MPDEV and HKLM\SYSTEM\CurrentControlSet\Services\MSDSM\Parameters\DsmParameters. Afterward, restart the server and run mpclaim -n -d <DeviceID> or mpclaim -n -a to reclaim the devices with MSDSM. Before reverting the CSVs to direct I/O mode, verify that all paths are correctly visible using mpclaim -s -d and that the DSM policy (Round Robin or Failover Only) is consistent across all nodes. It is also advisable to run Get-ClusterSharedVolumeState and Get-ClusterResource to verify that the volumes are back in “Online” status and have no pending redirections. Only when all multipath claims have been rebuilt and the SAN zoning is correct can you safely revert the CSVs to normal direct access without the risk of latency spikes.

If my answer is useful for you, please hit Accept the answer to support me.

Thank you,

Domic.

Answer 2

Use a staged recovery approach rather than only clearing claims.

Check the current MPIO state on the affected nodes.
- In Event Viewer and Device Manager, look for MPIO or storage errors.
- Run mpclaim -s -d to confirm which paths are visible and whether only some paths are missing.
- Pay particular attention to Event IDs 153, 129, and 140, and also cluster/storage events such as 51, 55, 98, 1069, 1205, 5120, 5121, and 5124.
Verify the basic storage path conditions before rebuilding claims.
- Confirm recent cabling, zoning, HBA, firmware, or driver changes.
- If only some paths are visible, verify zoning, cabling, and HBA status on both the server and the SAN.
- Make sure MPIO, DSM, and Storport are up to date.
- If using a third-party DSM, confirm it is certified for the operating system version.
Force Windows to refresh storage path discovery.
- On the affected computer, run:
```
   Update-HostStorageCache
   mpclaim -n -d
```
- If these commands do not restore path detection, restart the computer.
Review and rebuild the MPIO configuration.
- Open MPIO by running mpiocpl and review Discover Multi-Paths for incorrect or stale hardware IDs.
- The documented command-line tools for reviewing or updating MPIO discovery are:
```
   mpclaim -e
   mpclaim -r -i -a
```
- Also review available hardware and settings with:
```
   Get-MPIOAvailableHW
   Get-MPIOSetting
```
Validate SAN policy and path layout.
- In Command Prompt, verify SAN policy is not preventing disks from coming online:
```
   diskpart san policy=OnlineAll
```
- For iSCSI environments, each connection should use a different network adapter. If the system detects only one path, rebuild the iSCSI connections.
Before returning CSVs to direct I/O, validate cluster and disk health.
- In Failover Cluster Manager, check disk resource status.
- If a disk resource is stuck in Online Pending or fails with 1069 or 1205, increase the pending timeout:
```
   Get-ClusterResource "<Resource_Name>" | Set-ClusterParameter PendingTimeout 300000
```
- Check for file system issues with:
```
   chkdsk <drive>: /f /r
```
- Warning: if a disk is RAW and cannot be fixed, the documented recovery path is to delete and re-create the partition, then restore from backup.
For CSVs in redirected mode with high latency, also check for non-MPIO causes.
- Redirected mode is slower than direct mode.
- Review for network bottlenecks or misconfiguration, network adapter resource exhaustion, physical disk or HBA failure, switch or cable faults, insufficient CSV capacity, and unsupported filter drivers.

The context does not provide a supported mpclaim sequence to completely wipe all existing path claims. It does support refreshing host storage cache, rediscovering paths with mpclaim -n -d, reviewing claims with mpclaim -s -d and mpclaim -e, and rebuilding discovery with mpclaim -r -i -a. Use those steps together with SAN, DSM, and cluster validation before expecting CSVs to return to direct mode.