Apache Mesos — Distributed Systems Kernel for Data Center Resources

Introduction

Apache Mesos is a cluster manager that provides resource isolation and sharing across distributed applications. Originally developed at UC Berkeley and later adopted by organizations like Twitter and Apple, it pioneered the concept of a data center operating system — a single abstraction layer over heterogeneous compute resources.

What Apache Mesos Does

• Aggregates cluster resources (CPU, RAM, disk, ports, GPUs) into a unified pool
• Offers resources to frameworks (schedulers) using a two-level scheduling model
• Supports containerized workloads via the Mesos containerizer and Docker integration
• Provides fault tolerance through leader election (ZooKeeper) and agent recovery
• Enables co-location of diverse workloads (batch, real-time, stateful) on shared infrastructure

Architecture Overview

Mesos uses a master-agent model. The Mesos Master coordinates resource offers to registered frameworks (such as Marathon, Chronos, or Spark). Each Mesos Agent reports available resources and launches tasks assigned by frameworks. ZooKeeper handles master leader election for high availability. The two-level scheduling design lets each framework implement its own placement logic while Mesos handles resource isolation via Linux cgroups and namespaces.

Self-Hosting & Configuration

• Deploy masters (3 or 5 for HA) behind ZooKeeper for leader election
• Configure agents with resource attributes and roles for fine-grained allocation
• Use Marathon or Aurora as the long-running service scheduler on top of Mesos
• Set resource reservations and quotas per role to guarantee capacity for critical workloads
• Enable the Mesos containerizer for lightweight isolation or Docker for OCI image support

Key Features

• Two-level scheduling decouples resource management from application-specific placement
• Native support for GPUs, persistent volumes, and network isolation
• HTTP API for programmatic cluster interaction and custom framework development
• Multi-tenancy via roles, reservations, and resource quotas
• IP-per-container networking through CNI plugin integration

Comparison with Similar Tools

• Kubernetes — became the dominant container orchestrator; Mesos offers broader workload diversity but a smaller ecosystem
• HashiCorp Nomad — simpler cluster scheduler supporting containers, VMs, and batch jobs
• Docker Swarm — Docker-native orchestration with less flexibility than Mesos
• YARN — Hadoop resource manager focused on data processing rather than general workloads
• Slurm — HPC job scheduler oriented toward batch scientific computing

FAQ

Q: Is Apache Mesos still actively maintained? A: Mesos moved to the Apache Attic in 2024. While the codebase remains available, active development has wound down. Existing users should evaluate migration to Kubernetes or Nomad.

Q: What is Marathon? A: Marathon was the primary long-running service scheduler for Mesos, functioning similarly to Kubernetes Deployments. It managed application lifecycle, scaling, and health checks.

Q: Can Mesos run Kubernetes? A: There was a project (Kubernetes-Mesos) to run Kubernetes as a Mesos framework, but it is no longer maintained.

Q: Why did organizations move away from Mesos? A: Kubernetes gained a larger ecosystem of tools, cloud provider integrations, and community support, making it the pragmatic choice for most container orchestration needs.

Sources

• https://github.com/apache/mesos
• https://mesos.apache.org/documentation/