Hello, we are currently running VSphere and the thin provisioning feature is quite useful as most servers have a small drive footprint so we can fit in much more servers on the available storage.
However, since the space is not statically allocated and some servers are expanding over time it might lead to that they eventually consume the space on the whole datastore which essentially stops all machines on the datastore as well.
This is certainly to be avoided and since we want to utilize the extra space from thin provisioning the datastore is of course over allocated but is it possible to somehow enforce a growth limit when the datastore is almost out of storage so the system does not go down due to any unexpected expansion?
It is indeed better that one expanding server fails due to too little space instead of also bringing down all the other servers on the datastore so how can this be avoided iin a smoothful manner? Thanks!
As far as I understand Thin Provisioning on the vSphere side, you still provision a VM with a fixed amount of space. And the VM will grow inside that provisioned space. Not beyond it.
Check this screenshot: http://i.techrepublic.com.com/blogs/c-thin-space-used.jpg?tag=leftCol;post-2655
That I understand, however since most virtual machines do not need all allocated space we currently overallocate on the basis that most machines will not reach their maximum but in the rare cases it happens the datastore breaks down so it would be quite good if it was possible to enforce some limit/alert when the datastore space is getting dangerously low.
Datastore space usage alarms are key for any VMware installation but especially for thin-provisioned volumes. From your description, you say your servers have "a small drive footprint" -- does this mean you're not using a SAN? If so, that strikes me as a bigger issue than thin-provisioning.
If you are truly concerned about a sudden spurt that leaves you no time to respond (which shouldn't be the case in a properly-monitored environment) one option is to keep an unused spare drive online. Then set up an alarm on the thinly-provisioned datastore, one of whose actions (besides emailing you) is to run a vmkfstools command to add the spare drive as an emergency extent to the datastore. This is a bit of a hack but in principle it should work.