I’m running into some odd behavior from clustered LVM commands on a 3-node pacemaker-based cluster. When other processes on the nodes are doing somewhat heavy sequential I/O traffic, LVM commands such as lvs, pvs, vgs, lvdisplay, pvdisplay, and vgdisplay have widely variable, often very long execution times. The same LVM command done repetitively will have execution times varying from less than a second to 10-15 seconds, more rarely up to around 60 seconds. Other read-heavy commands such as find, wc, and sum have much less variability and much shorter tails in the distribution of their execution times.
Is it known what causes the cLVM commands to be so slow at times?
Perchance, are there known ways to get cLVM commands to be less slow?
Or, is this just par for the course for clustered LVM?