Did a quick test on codfw, for different scenarios:

• When pausing the cluster (no reads, no writes) -> VMs hang on syncing disk (cached writes work), but no major issues happen at libvirt level, at least for 30s
  • When unpausing everything goes back to normal (syncs go through)
• When pausing the cluster, and taking down the monitors -> same behavior as before
  • When bringing up the monitors, waiting for quorum (<1s), and unpausing, everything goes back to normal (same as before)
• When pausing the VMs, then the cluster, and taking down the monitors -> No issues on libvirt, at least for 1min
  • When bringing up the monitors, waiting for quorum (<1s), and unpausing the cluster and VMs, everything goes back to normal (though external connections to the VM time out, ex. ssh)

I have not tried to leave it not running for longer, but will try and report back (enough for timeouts to kick in)

I'll prepare a script to pause/unpause all VMs