Containers initially came into use as a method to more efficiently deploy lightweight stateless applications. But a new containers report from container monitoring service provider Sysdig, based on the more than 90,000 containers it actively monitors, shows a marked increase in usage of Java Virtual Machines as well as databases such PostgreSQL and MongoDB being deployed using containers.
Apurva Davé, vice president of marketing at Sysdig, says the increase is indicative of containers being relied on to run more stateful enterprise-class applications.
Given the ephemeral nature of containers, the assumption often made is that they are not widely used to deploy these classes of applications. The Sysdig containers report finds 95 percent of containers live less than a week, while 69 percent of images are updated in the span of a week. But 67 percent of the services based on containers live beyond a week. That suggests that while containers and pods may come and go as developers update those services, the service itself remains in place, says Davé.
Nevertheless, a lot of containers get spun to perform a specific task, such as jobs within a Jenkins continuous integration/continuous development (CI/CD) platform. Overall, the Sysdig containers report finds 11 percent of containers stay alive for less than 10 seconds. Another 27 percent of containers churn between five to 10 minutes.
The primary container orchestration engine being used to manage all that churn is Kubernetes. More than half the containers (51 percent) being managed were running on a Kubernetes clusters, followed by 11 percent for Docker Swarm and 4 percent for Mesos. While usage of Mesos dropped in the last year, the report notes that clusters based on Mesos are on average 50 percent larger than Kubernetes. Docker Swarm clusters, in contrast, are on average 30 percent smaller than Kubernetes.
In terms of tools used to monitor, StatsD and Prometheus ranked highest. Half the customers employ a container registry, with the top instance being Google Container Registry (GCR), followed by Quay, Docker and Amazon Elastic Container Registry (ECR).
The Sysdig report also notes there is more use of containers other than Docker. Rkt now represents 12 percent of the containers running, while Mesos containers grew to 4 percent. Usage of LXC also grew slightly.
Finally, the report also finds that the median number of containers per host per customer climbed 50 percent, to 15 from 10. At the top end of the spectrum, Sysdig found one customer running 154 containers on a single host, up from a maximum of 95 the previous year. Davé says that while most containers are running on a virtual machine, more customers are starting to push the envelope in terms of the number of containers that can be hosted on a single host. That shift suggests over time there will be more reliance on bare-metal servers as organizations looks to maximize both server density and application performance.
Clearly, as IT organizations become more comfortable with containers, a major shift is underway that affects all classes of applications. Many IT operations teams were taken by surprise as developers began to embrace containers in large numbers to meet demands for faster application releases and updates, notes Davé. The question now is not whether enterprise IT organizations will be embracing containers, but rather to what degree.