Preemptibillity

A preemptible instance is a VM instance type that is much cheaper than the regular Compute Engine machine types that we discussed in the previous section. The reason for this is that a preemptible instance uses spare resources in Google's infrastructure, but they might be terminated (that is, preempted) at any time if Google requires the resources held by this VM.

Preemptible instances cost only a fraction of what other VM instances do and so if you have a fault-tolerant application, for instance, a processing-only node in a Hadoop cluster, a preemptible instance might make a lot of sense and particularly so, if you are budget conscious.

Preemptible instances will definitely be terminated after running for 24 hours. Hence, do not ever use preemptible instances for long running jobs. When using a preemptible instance for relatively short jobs, the probability of termination is typically quite low. This probability of termination will vary based on the day, the zone, the network conditions, and other factors such as migrations and maintenance. Preemptible instances, unlike other VM types, cannot migrate; that is, they cannot stay alive during software updates and they will be forcibly restarted during maintenance. The first step in the preemption process is that the Google Cloud Platform or Compute Engine will send your instance a soft off signal where your machine has 30 seconds in which to clean up through a shutdown script and give up control. If it does not do so, Compute Engine will forcibly take control by sending a mechanical off signal. Therefore, if you use preemptible instances, do ensure that you have a well written shutdown script associated with the preemptible instance.