I've seen these kinds of messages on VMs when I/O overwhelms disk caching. The default settings are not necessarily the best for a VM environment that is backed by fast disk arrays.
There are 2 kernel parameters that can be adjusted to control disk caching. They are:
- vm.dirty_background_ratio - this controls when the OS background processes will start flushing the cache to disk
- vm.dirty_ratio - this is the max amount of memory used for cache, once you hit this limit, all I/O blocks until dirty pages are written to disk
When I ran into this problem a month or two ago, I found this blog post
https://lonesysadmin.net/2013/12/22/...m-dirty_ratio/ that has a good explanation of different options for tuning the cache depending on application behaviour. It's not the newest post I've found on this problem, but they all pointed to the same 2 kernel parameters, usually with less information about what they do.