Triton DataCenter (just "Triton" for short, formerly "SmartDataCenter" and "SDC") is an open-source cloud management platform that delivers next generation, container-based, service-oriented infrastructure across one or more data centers. With an emphasis on ease of installation and operation, Triton is proven at scale: Triton powers the Triton Cloud and private data centers (Triton Enterprise) worldwide.
This repository provides documentation for the overall Triton project and pointers to the other repositories that make up a complete Triton deployment. See the repository list.
A Triton DataCenter installation consists of two or more servers. All servers run SmartOS. One server acts as the management server, the head node, which houses the initial set of core services that drive Triton. The remainder are compute nodes (CNs) which run instances (containers and virtual machines).
- SmartOS zones provides high performance container virtualization. KVM support on top of zones means secure full Linux and Windows guest OS support.
- RESTful API and CLI tooling for customer self-service
- Complete operator portal (web app)
- Robust and observable service oriented architecture (implemented primarily in Node.js)
- Automated USB key installation
Triton consists of the following components:
- A public API for provisioning and managing instances (virtual machines), networks, users, images, etc.
- An operator portal
- A set of private APIs
- Agents running in the global zone of CNs for management and monitoring
For more details, see:
- The Triton Enterprise documentation.
- Triton DataCenter Architecture for overall architecture.
- Triton DataCenter Reference for an overview of each component.
Community discussion about Triton DataCenter happens in two main places:
The sdc-discuss mailing list. If you wish to send mail to the list you'll need to join, but you can view and search the archives online without being a member.
In the #smartos IRC channel on the Freenode IRC network.
Cloud on a Laptop (CoaL)
An easy way to try Triton DataCenter is by downloading and installing a Cloud on a Laptop (CoaL) build. CoaL is a VMware virtual appliance that provides a full Triton head node for development and testing.
The CoaL Setup document contains detailed instructions for downloading and installing the virtual appliance.
If you already have a CoaL and would like to update the installation, follow the instructions
for updating a Triton standup using
Installing Triton on a Physical Server
A Triton DataCenter server runs SmartOS which is a live image. This means that it boots from a USB flash drive (key). a physical USB key, inserting the key and booting the server from that key. To install Triton, first obtain the latest release USB build.
For Triton development only, the minimum server hardware is:
- 8 GB USB flash drive
- Intel Processors with VT-x and EPT support (all Xeon since Nehalem)
- 16 GB RAM
- 6 GB available storage
Hardware RAID is not recommended. Triton will lay down a ZFS ZPOOL across all available disks on install. You'll want much more storage if you're working with images and instances.
If setting up a Triton DataCenter pilot then you'll want to review the Hardware Selection Requirements, Networking Layout Requirements, and Installation Prerequisites which include IPMI and at least 10 gigabit Ethernet. The supported hardware components for SmartOS are described in the SmartOS Hardware Requirements. Joyent certified hardware for Triton DataCenter are all in the Joyent Manufacturing Database.
To install Triton, first download the latest release image:
curl -C - -O https://us-east.manta.joyent.com/Joyent_Dev/public/SmartDataCenter/usb-latest.tgz
Once you have downloaded the latest release image, you will need to write it to a USB key boot the head node server using the USB key, and follow the install prompts. All steps necessary to plan, install, and configure Triton DataCenter (Triton) are available in the Joyent customer documentation Installing Triton Elastic Container Infrastructure.
Triton is composed of several pre-built components:
- A SmartOS platform image. This is a slightly customized build of vanilla SmartOS for Triton.
- Virtual machine images for Triton services (e.g. imgapi, vmapi, adminui), which are provisioned as VMs at install time.
- Agents bundled into a single package installed into the global zone of each compute node.
Each component is built separately and then all are combined into CoaL and USB
builds (see the preceding sections) via the sdc-headnode
repository. The built components are
typically stored in a Manta object store,
e.g. Joyent's public Manta,
and pulled from there. For example, Joyent's builds push to
/Joyent_Dev/public/builds in Joyent's public Manta in us-east-1
You can build your own CoaL and USB images on Mac or SmartOS (see the sdc-headnode README). However, all other Triton components must be built using a running Triton (e.g. on the Joyent Cloud or in a local CoaL). See the building document for details on building each of the Triton components.
To report bugs or request features, submit issues here on GitHub, joyent/triton/issues (or on the GitHub issue tracker for the relevant project). For support of Joyent products and services, please contact Joyent customer support instead.
See the Contribution Guidelines for information about contributing changes to the project.
Triton DataCenter is very opinionated about how to architect a cloud. These opinions are the result of many years of deploying and debugging the Joyent public cloud. Design principles include the following:
- A VM's primary storage should be local disk, not over the network -- this avoids difficult to debug performance pathologies.
- Communication between internal APIs should occur in its own control plane (network) that is separate from the customer networks. Avoid communicating over the open Internet if possible.
- A provisioned VM should rely as little as possible on Triton services outside of the operating system for its normal operation.
- Installation and operation should require as little human intervention as possible.
The goals behind the design of Triton services include:
- All parts of the stack should be observable.
- The state of the running service should be simple to obtain.
- The internals of the system should make it straightforward to debug from a core file (from a crash or taken from a running process using gcore(1)).
- Services should be RESTful and accept JSON unless there is a compelling reason otherwise.
- Services should avoid keeping state and should not assume that there is only one instance of that service running. This allows multiple instances of a service to be provisioned for high availability.
- Node.js and C should be used for new services.
Dependencies and Related Projects
Triton DataCenter uses SmartOS as the host operating system. The SmartOS hypervisor provides both SmartOS zone (container) and KVM virtualization.
Joyent's open-source Manta project is an HTTP-based object store with built-in support to run arbitrary programs on data at rest (i.e., without copying data out of the object store). Manta runs on and integrates with Triton DataCenter.