[Problem Description]
[Active-active +Dorado18000 V6 6.0.1 +Solaris is preferred at the local end of ALUA] In the solaris10u11 sparc+QLE2562+Veritas DMP5.1SP1+ZFS scenario, after a zfs file system is created on the dmp aggregated disk, the drive letter of the active-active LUN is abnormal, causing path loss.
[Symptom Description]
1. On Solaris OS, create a zpool of the Zfs file system on the drive letter after DMP aggregation. Run the DMP command to check whether the drive letter name and disk array information are changed.
Before creation:
After the drive is created, the drive letter changes to 0 _ 0.
2. About 5 minutes after the zpool is created, the drive letter name and array information are restored, but the path is lost and becomes disabled. (The number of lost paths is not fixed. Sometimes, only some paths are lost. Sometimes, all paths at one end of the HyperMetro pair are lost.)
This problem occurs when:
1. Huawei storage uses HyperMetro LUNs to map them to Solaris hosts.
2. The Veritas DMP multipathing software is used on the host. Then, the ZFS file system is created on the drive letter after the DMP aggregation.
[Cause]
According to the comparison analysis, the possible causes of the current problem are as follows:
1. When DMP takes over HyperMetro LUNs of Huawei storage, some private information is stored in management data structures such as Dmpnodes of hosts. (This may be caused by the information about the arrays at both ends of the HyperMetro LUN. Huawei storage devices have a private authentication interface with DMP to query the information about the arrays masqueraded by the HyperMetro LUN. DMP needs to aggregate drive letters based on the information.) . These fields may conflict with zpool. When zpool is created, the OS overwrites these fields. As a result, the DMP displays abnormal information.
2. Five minutes after the zpool is created (The default routine test and recovery period of the DMP is 5 minutes.) The DMP requeries and restores some fields through routine tests, but not all fields are restored. As a result, the displayed information is restored, but the status of some trails is incorrect.
[Location Method]
1. Check whether the operation meets the conditions for reoccurring.
2. Run the DMP command on the host to check whether the preceding problem occurs.
3. Check the DMP logs on the host. After the zpool is created, the logs contain "Enable Disk Array 0" and then the original path is removed.
4. About 5 minutes after the Zpoo is created, the DMP logs contain Enabled Disk array huawei-xsg1xxxxx logs. Disk array 0 is also disabled.
[Solution]
Due to event A, there is no way to obtain confirmation information about this issue from Veritas. Therefore, the current problem can only be avoided. The workarounds are as follows:
1. In active-active LUN scenarios, if the zfs file system is required, the multipathing stms of the Solaris operating system must be used.
2. In active-active LUN scenarios, if DMP multipathing is required, only the Vxfs file system of Veritas Storage Foundation can be used.
3. Non-active LUNs do not have the preceding restrictions. No workaround is required.
[Appendix]
NA
[Applicability]
All OceanStor V3/V5 series versions
All OceanStor Dorado V3 series versions
All OceanStor Dorado V6 series versions
