This week, Mirantis and Pivotal are announcing a partnership to make Cloud Foundry Platform-as-a-Service (PaaS) fast and easy to consume, and put into robust production on Mirantis OpenStack. As a result of the collaboration, Pivotal customers will soon gain the option of using Mirantis OpenStack’s easy-to-deploy OpenStack distribution to create cost-effective, robust, production IaaS under Pivotal Cloud Foundry and Pivotal’s related products. Mirantis OpenStack users will gain the ability to deploy — rapidly and simply — these leading container-based PaaS solutions, enabling sophisticated application lifecycle management.
The combo of Mirantis OpenStack and Pivotal Cloud Foundry will support continuous integration and delivery and power other critical PaaS use-cases: the products’ respective toolkits can work together to empower full-stack DevOps. So great are the potential synergies that Mirantis has recently joined the Cloud Foundry Foundation, with the goal of supporting the community and facilitating broader collaboration.
As our Runbook (see below) demonstrates, it’s readily possible to install Cloud Foundry on Mirantis OpenStack today — though what we’re presenting here is really only suitable for PoC ‘tire-kicking’ deployments (and is scaled for very small target environments). However, Mirantis and Pivotal can help today with production-scale deployments (and have mutual customers in production).
Pivotal is soon expected to release their version 1.5, with full support for OpenStack (and certified on Mirantis OpenStack), including a high-level installer permitting easy, automated installation. When PCF 1.5 is generally available, Mirantis will produce a similar (but shorter) runbook document, providing advice for installing Pivotal CF 1.5 on Mirantis OpenStack.
Cloud Foundry on MOS
Our step-by-step runbook can help adventurous users install community Cloud Foundry (build 205) in a Proof of Concept (PoC) configuration on a Mirantis OpenStack cluster (version 5.0.1+ or 6.0+, i.e., Icehouse or Juno), or on a hosted Mirantis OpenStack cluster (Mirantis OpenStack On Demand with Fuel version 5.0.1+ or 6.0+, i.e., Icehouse or Juno generation) with a minimum of one free compute node supporting a minimum of about 24 vCPUs and about 1TB block storage.
The installation process involves manually building and installing a recent version of community Cloud Foundry using a Pivotal-developed tool called BOSH, that unifies release engineering, deployment and lifecycle management for small- and large-scale cloud software. The version of Cloud Foundry installed (cf 205) aligns with Pivotal Cloud Foundry version 1.4, which provides formal support for VMware vSphere, vCloud Air and Amazon AWS (but not OpenStack).
Full support for OpenStack from Pivotal — plus easier installation at production scales — awaits release of Pivotal Cloud Foundry 1.5. Meanwhile, those with access to a Mirantis OpenStack datacenter (bare metal or virtualized) of at least minimal capacity will be able to experience Cloud Foundry via this PoC, learn its architecture and play with its tools.
CloudFoundry on OpenStack, basic logical diagram. Deployment proceeds from the Inception VM, on which the BOSH CLI, its OpenStack provider plugin, and other components are installed for remote access via SSH. Creating the inception VM on the target cluster increases security and speeds up deployment, by reducing time required to upload stemcells and release components.
The BOSH Director VM works as the ‘inception’ VM’s agent: this VM drives OpenStack via API commands under Admin authentication (cloud-wide/toplevel or within a project/tenant). CloudFoundry cluster deployment begins with the creation of several compiler VMs, which are transitory — these compile CloudFoundry’s packages, which are then deployed to non-transitory target VMs.
Gross capacity requirements for PoC Cloud Foundry deployment detailed in Mirantis Runbook. To limit resource requirements, a custom OpenStack flavor (cf.dea) is defined for Cloud Foundry ‘Runner’ nodes (Droplet Execution Agents – analogous to nova compute nodes in an OpenStack cluster). In a production deployment, Pivotal reports that 8 vCPUs and 32GB RAM is optimal for each DEA.