MesosCon North America 2017

This talk will provide an overview of multi-tenancy in Apache Mesos, including recent work that was done to enable multi-tenant frameworks (with multi-role framework support and support for hierarchical roles). Topics include multi-tenant resource management, isolation, security, and other multi-tenancy concerns. This talk will also cover some upcoming work.

Mesos currently supports both local and external persistent volumes. However, both of them have some limitations. For instance, local persistent volume support requires tedious setup by operators for additional disks. External persistent volume support does not support dynamic provisioning, and Mesos doesn’t track the resources associated with the external volumes. This makes quota control, reservation, fair sharing almost impossible to implement. Our goal is to improve the experience for both of them. The solution is based on a general resource provider model in Mesos, and a standard pluggable interface when talking to storage platforms. This interface is called Container Storage Interface (CSI). This talk will give you an overview about CSI and the resource provider based solution for storage, and dive deeper into how storage platforms can be easily integrated with Mesos using CSI.

Microsoft has worked closely with Mesos and DC/OS communities to support customer workloads on Windows by expanding Mesos Ecosystem. In this presentation, we will review some of our work and learnings, then focus on the enhancement on CMake build system providing a single cross-platform developer experience, also discuss the roadmap for what’s coming next and show some cool demos!

Twitter use Apache Aurora on Apache Mesos to provide a scalable cluster that is used by the engineers to run their microservices. The platform has seen dramatic adoption due to the deploy features provided by Aurora, which is appealing to the engineers and the resource isolation features provided by Mesos, which improves the cluster utilization. At Twitter we use isolation of CPU, Memory, Disk and Network bandwidth to provide guaranteed access to resources for application running inside a container. One of the glaring exceptions of the container isolation is the lack of file system isolation. To this end we experimented with enabling Docker containers inside Twitter to provide an isolation layer between the host and the container. In this talk we will share our experiences at an attempt to using Docker at scale and the lessons we have learned along the journey.

In this talk we will focus on effective Mesos cluster utilization for non-production jobs, which are commonly used for development and testing. While many of such jobs are idle, they consume cluster resources, leaving a huge gap between allocated and actually utilized resources.
We will present our approach for detecting idle resources, offering them to other non-production jobs using Mesos oversubscription model, and isolating non-production jobs from production jobs. We will also highlight issues in Mesos’ approach to oversubscription and discuss possible solutions and workarounds.

Mesos, different from some other container orchestrators, has its own native container runtime. Based on the pure Linux Kernel namespaces and cgroups, Mesos supports different container image formats with advantages of extensible container storage, networking, and security.

As of today it support all major container image formats such as OCI or Docker Image format, as well as default storage and networking plugins formats with CSI and CNI.

Historically, Mesos has provided a command executor for running one off tasks by schedulers e.g., Marathon. For launching a group of co-located tasks, the recommended way was to implement the logic using a custom executor. With the support for task groups aka Pods in Mesos, the default executor is the new recommended way for running pods in production. Moreover, based on nested containers, debug containers become realistic to enter any containers namespaces.

In this talk you will learn the following:

- Overview of the Mesos Agent API’s for nested containers used by the default executor

- Best practices for running sidecar/adapter containers and transient tasks

- Health Checks and Probes (Non-interpreted health checks)

- Default Termination Policy for the default executor

- New planned upcoming features on the roadmap