Boot and Shutdown Times

In a distrubuted, auto-scaling cluster, it is critical that applications can boot promptly, so that services can automatically scale up and start serving requests quickly. Ideally they should boot within ten seconds. This is one of the 12-factor app principles.

Likewise, if the application is sent the kill signal, it should be able to shut down promptly. Ten seconds is a good rule of thumb here too. If they do not, they could be consuming cluster resources, and blocking other applications from being scheduled on the node. This could delay cluster maintenance or scaling of other services.

Slow Frameworks

When dealing with languages like Java or Python, and frameworks like Spring, Hibernate, and Anaconda, it is not unusual to have hundreds of dependencies, and boot times measured in minutes. These often create problems in clusters. That is one reason lower level system languages are sometimes preferred, they are more light-weight and even complicated apps tend to boot in under ten seconds.