PowerScale: nodes panic: Assertion Failure kernel:pdm_entries_get_cacheable_entry+0x48e

Multiple PowerScale nodes panic with stack:
 

ipfw2 (+ipv6) initialized, divert loadable, nat loadable, default to accept, logging disabled
mce2: Interface stopped DISTRIBUTING, possible flapping
panic @ time 1642530761.327, thread 0xfffffea7a8a76580: Assertion Failure
time = 1642530761
cpuid = 24, TSC = 0xe457b789d574ac
Panic occurred in module kernel loaded at 0xffffffff80200000:

Stack: --------------------------------------------------
kernel:isi_assert_halt+0x2e
kernel:pdm_entries_get_cacheable_entry+0x48e
kernel:pdm_entries_get+0x95
kernel:pdm_entries_get_global+0x33
kernel:pdm_get_painted_domids+0x2e2
kernel:pdm_member_generate_entry+0xda
kernel:pdm_member_get_membership_entry+0x1b7
kernel:pdm_member_init_operation+0x16d
kernel:ifm_init_operation+0xb7
kernel:txn_i_include_vnode_to_list+0xc0
kernel:pdm_vget_adsio_txn_include+0x12c
kernel:pdm_domain_paint_adsdir+0x41
kernel:pdm_unlink+0x1f0
kernel:bam_rename+0x2bcb
kernel:ifs_vnop_wraprename+0x96
kernel:VOP_RENAME_APV+0x9b
isi_lwext.ko:lwextsvc_rename+0xe35
kernel:amd64_syscall+0x380
--------------------------------------------------
*** FAILED ASSERTION !pdm_lk_is_held(domid, PDM_EXCLUSIVE) @ /b/mnt/src/sys/ifs/pdm/pdm_ops.c:443:
Disabling swatchdog
Dumping stacks (40960 bytes)

Caused by the following defect: PSCALE-63084: FAILED ASSERTION !pdm_lk_is_held(domid, PDM_EXCLUSIVE) @ /b/mnt/src/sys/ifs/pdm/pdm_ops.c:443 During ADS rename

This issue is related to ADS (alternate data streams) workflow. Alternate Data Streams is a data structure used within Windows that stores metadata-types of information about a file, such as comments about a file. It does not exist in FreeBSD (which is the underlying file system of OneFS on Isilon), but OneFS supports it and it is treated as a file within OneFS.

The race condition occurs when OneFS renames an ADS file across a domain boundary (in or out of snapshot) in order to overwrite a hard linked ADS file. This causes a node to become unresponsive when an operation was attempted. As a result, data unavailability (DU) may occur.

How to determine whether a cluster is at risk for this issue?

The cluster is running OneFS code older than 9.3.0.0.

If workflow does not use ADS and 'rename' operations, then OneFS will not experience this issue.