Containers are the new IaaS

containers-are-the-new-iaas

At work, as with many other companies, we're actively investing in new platforms, including container platforms and public cloud. We use Kubernetes based container platforms both on-premise and in the cloud, but are also very adamant that the container platforms should only be used for application workload that is correctly designed for cloud-native deployments: we do not want to see vendors packaging full operating systems in a container and then shouting they are now container-ready.

Sadly, we notice more and more vendors abusing containerization to wrap their products in and selling it as 'cloud-ready' or 'container-ready'. For many vendors, containers allow them to bundle everything as if it were an appliance, but without calling it an appliance - in our organization, we have specific requirements on appliances to make sure they aren't just pre-build systems that lack the integration, security, maintainability and supportability capabilities that we would expect from an appliance.

Even developers are occasionally tempted to enlarge container images with a whole slew of middleware and other services, making it more monolithic solutions than micro-services, just running inside a container because they can. I don't feel that this evolution is beneficial (or at least not yet), because the maintainability and supportability of these images can be very troublesome.

This evolution is similar to the initial infrastructure-as-a-service offerings, where the focus was on virtual machines: you get a platform on top of which your virtual machines run, but you remain responsible for the virtual machine and its content. But unlike virtual machines, where many organizations have standardized management and support services deployed for, containers are often shielded away or ignored. But the same requirements should be applied to containers just as to virtual machines.

Let me highlight a few of these, based on my Process view of infrastructure.

Cost and licensing

Be it on a virtual machine or in a container, the costs and licensing of the products involved must be accounted for. For virtual machines, this is often done through license management tooling that facilitates tracking of software deployments and consumption. These tools often use agents running on the virtual machines (and a few run at the hypervisor level so no in-VM agents are needed).

Most software products also use licensing metrics that are tailored to (virtual) hardware (like processors) or deployments (aka nodes, i.e. a per-operating-system count). Software vendors often have the right to audit software usage, to make sure companies do not abuse their terms and conditions.

Now let's tailor that to a container environment, where platforms like Kubernetes can dynamically scale up the number of deployments based on the needs. Unlike more static virtual machine-based deployments, we now have a more dynamic environment. How do you measure software usage here? Running software license agents inside containers isn't a good practice. Instead, we should do license scanning in the images up-front, and tag resources accordingly. But not many license management tooling is already container-aware, let alone aligned with a different way of working.

But "our software license management tooling is not container-ready yet" is not an adequate answer to software license audits, nor will the people in the organization that are responsible for license management be happy with such situations.

Product lifecycle

Next to the licensing part, companies also want to track which software versions are being used: not just for vulnerability management purposes, but also to make sure the software remains supported and fit for purpose.

On virtual machines, regular software scanning and inventory setup can be done to report on the software usage. And while on container environments this can be easily done at the image level (which software and versions are available in which containers) this often provides a pre-deployment view, and doesn't tell us if a certain container is being used or not, nor if additional deployments have been triggered since the container is launched.

Again, deploying in-container scanning capabilities seems to be contra-productive here. Having an end-to-end solution that detects and registers software titles and products based on the container images, and then provides insights into runtime deployments (and history) seems to be a better match.

Authorization management (and access control)

When support teams need to gain access to the runtime environment (be it for incident handling, problem management, or other operational tasks) most companies will already have a somewhat finer-grained authorization system in place: you don't want to grant full system administrator rights if they aren't needed.

For containers, this is often not that easy to accomplish: the design of container platforms is tailored to situations where you don't want to standardize on in-container access: runtimes are ephemeral, and support is handled through logging and metric, with adaptation to the container images and rolling out new versions. If containers are starting to get used for more classical workloads, authorization management will become a more active field to work out.

Consider a database management system within the container alongside the vendor software. Managing this database might become a nightmare, especially if it is only locally accessible (within the container or pod). And before you yell how horrible such a setup would be for a container platform... yes, but it is still a reality for some.

Auditing

Auditing is a core part of any security strategy, logging who did what, when, from where, on what, etc. For classical environments, audit logging, reporting and analysis are based upon static environment details: IP addresses, usernames, process names, etc.

In a container environment, especially when using container orchestration, these classical details are not as useful. Sure, they will point to the container platform, but IP addresses are often shared or dynamically assigned. Usernames are dynamically generated or are pooled resources. Process identifiers are not unique either.

Auditing for container platforms needs to consider the container-specific details, like namespaces. But that means that all the components involved in the auditing processes (including the analysis frameworks, AI models, etc.) need to be aware of these new information types.

In the case of monolithic container usage, this can become troublesome as the in-container logging often has no knowledge of the container-specific nature, which can cause problems when trying to correlate information.

Conclusion

I only touched upon a few processes here. Areas such as quality assurance and vulnerability management are also challenges for instance, as is data governance. None of the mentioned processes are impossible to solve, but require new approaches and supporting services, which make the total cost of ownership of these environments higher than your business or management might expect.

The rise of monolithic container usage is something to carefully consider. In the company I work for, we are strongly against this evolution as the enablers we would need to put in place are not there yet, and would require significant investments. It is much more beneficial to stick to container platforms for the more cloud-native setups, and even in those situations dealing with ISV products can be more challenging than when it is only for internally developed products.

Feedback? Comments? Don't hesitate to drop me an email, or join the discussion on Twitter.