When a GPFS client machine rebooted, the GPFS File System was unmounted with no error sign. When NSD issues the command “mmstartup -N client_node“, similarly, there is no error sign.
But if you do a
# mmdsh -v -N all "/usr/lpp/mmfs/bin/mmfsadm dump waiters" > all.waiters
You may see something like
.......Sync handler: on ThCond 0x1015236D30 (0xFFFFC20015236D30) (wait for inodeFlushFlag), reason 'waiting for the flush flag'......
This occurs when a revoke comes in and an mmapped file needs to be flushed to disk. GPFS tells Linux to flush all dirty mapped pages, and the thread then waits for Linux to report that this has been completed. So something in the kernel is preventing all the dirty pages from being flushed. I guess the best way is to use the NSD nodes to issue a command to do a
# mmshutdown -N client_node # mmstartup -N client_node