github.com/iqoqo/nomad@v0.11.3-0.20200911112621-d7021c74d101/website/pages/docs/internals/architecture.mdx (about) 1 --- 2 layout: docs 3 page_title: Architecture 4 sidebar_title: Architecture 5 description: Learn about the internal architecture of Nomad. 6 --- 7 8 # Architecture 9 10 Nomad is a complex system that has many different pieces. To help both users and developers of Nomad 11 build a mental model of how it works, this page documents the system architecture. 12 13 ~> **Advanced Topic!** This page covers technical details 14 of Nomad. You do not need to understand these details to 15 effectively use Nomad. The details are documented here for 16 those who wish to learn about them without having to go 17 spelunking through the source code. 18 19 # Glossary 20 21 Before describing the architecture, we provide a glossary of terms to help 22 clarify what is being discussed: 23 24 - **Job** - A Job is a specification provided by users that declares a workload for 25 Nomad. A Job is a form of _desired state_; the user is expressing that the job should 26 be running, but not where it should be run. The responsibility of Nomad is to make sure 27 the _actual state_ matches the user desired state. A Job is composed of one or more 28 task groups. 29 30 - **Task Group** - A Task Group is a set of tasks that must be run together. For example, a 31 web server may require that a log shipping co-process is always running as well. A task 32 group is the unit of scheduling, meaning the entire group must run on the same client node and 33 cannot be split. 34 35 - **Driver** – A Driver represents the basic means of executing your **Tasks**. 36 Example Drivers include Docker, Qemu, Java, and static binaries. 37 38 - **Task** - A Task is the smallest unit of work in Nomad. Tasks are executed by drivers, 39 which allow Nomad to be flexible in the types of tasks it supports. Tasks 40 specify their driver, configuration for the driver, constraints, and resources required. 41 42 - **Client** - A Client of Nomad is a machine that tasks can be run on. All clients run the 43 Nomad agent. The agent is responsible for registering with the servers, watching for any 44 work to be assigned and executing tasks. The Nomad agent is a long lived process which 45 interfaces with the servers. 46 47 - **Allocation** - An Allocation is a mapping between a task group in a job and a client 48 node. A single job may have hundreds or thousands of task groups, meaning an equivalent 49 number of allocations must exist to map the work to client machines. Allocations are created 50 by the Nomad servers as part of scheduling decisions made during an evaluation. 51 52 - **Evaluation** - Evaluations are the mechanism by which Nomad makes scheduling decisions. 53 When either the _desired state_ (jobs) or _actual state_ (clients) changes, Nomad creates 54 a new evaluation to determine if any actions must be taken. An evaluation may result 55 in changes to allocations if necessary. 56 57 - **Server** - Nomad servers are the brains of the cluster. There is a cluster of servers 58 per region and they manage all jobs and clients, run evaluations, and create task allocations. 59 The servers replicate data between each other and perform leader election to ensure high 60 availability. Servers federate across regions to make Nomad globally aware. 61 62 - **Regions and Datacenters** - Nomad models infrastructure as regions and 63 datacenters. Regions may contain multiple datacenters. Servers are assigned to 64 a specific region, managing state and making scheduling decisions within that 65 region. Multiple regions can be federated together. For example, you may 66 have a `US` region with the `us-east-1` and `us-west-1` datacenters, 67 connected to the `EU` region with the `eu-fr-1` and `eu-uk-1` datacenters. 68 Requests that are made between regions are forwarded to the appropriate servers. 69 Data is _not_ replicated between regions. 70 71 - **Bin Packing** - Bin Packing is the process of filling bins with items in a way that 72 maximizes the utilization of bins. This extends to Nomad, where the clients are "bins" 73 and the items are task groups. Nomad optimizes resources by efficiently bin packing 74 tasks onto client machines. 75 76 # High-Level Overview 77 78 Looking at only a single region, at a high level Nomad looks like this: 79 80 [](/img/nomad-architecture-region.png) 81 82 Within each region, we have both clients and servers. Servers are responsible for 83 accepting jobs from users, managing clients, and [computing task placements](/docs/internals/scheduling/scheduling). 84 Each region may have clients from multiple datacenters, allowing a small number of servers 85 to handle very large clusters. 86 87 In some cases, for either availability or scalability, you may need to run multiple 88 regions. Nomad supports federating multiple regions together into a single cluster. 89 At a high level, this setup looks like this: 90 91 [](/img/nomad-architecture-global.png) 92 93 Regions are fully independent from each other, and do not share jobs, clients, or 94 state. They are loosely-coupled using a gossip protocol, which allows users to 95 submit jobs to any region or query the state of any region transparently. Requests 96 are forwarded to the appropriate server to be processed and the results returned. 97 Data is _not_ replicated between regions. 98 99 The servers in each region are all part of a single consensus group. This means 100 that they work together to elect a single leader which has extra duties. The leader 101 is responsible for processing all queries and transactions. Nomad is optimistically 102 concurrent, meaning all servers participate in making scheduling decisions in parallel. 103 The leader provides the additional coordination necessary to do this safely and 104 to ensure clients are not oversubscribed. 105 106 Each region is expected to have either three or five servers. This strikes a balance 107 between availability in the case of failure and performance, as consensus gets 108 progressively slower as more servers are added. However, there is no limit to the number 109 of clients per region. 110 111 Clients are configured to communicate with their regional servers and communicate 112 using remote procedure calls (RPC) to register themselves, send heartbeats for liveness, 113 wait for new allocations, and update the status of allocations. A client registers 114 with the servers to provide the resources available, attributes, and installed drivers. 115 Servers use this information for scheduling decisions and create allocations to assign 116 work to clients. 117 118 Users make use of the Nomad CLI or API to submit jobs to the servers. A job represents 119 a desired state and provides the set of tasks that should be run. The servers are 120 responsible for scheduling the tasks, which is done by finding an optimal placement for 121 each task such that resource utilization is maximized while satisfying all constraints 122 specified by the job. Resource utilization is maximized by bin packing, in which 123 the scheduling tries to make use of all the resources of a machine without 124 exhausting any dimension. Job constraints can be used to ensure an application is 125 running in an appropriate environment. Constraints can be technical requirements based 126 on hardware features such as architecture and availability of GPUs, or software features 127 like operating system and kernel version, or they can be business constraints like 128 ensuring PCI compliant workloads run on appropriate servers. 129 130 # Getting in Depth 131 132 This has been a brief high-level overview of the architecture of Nomad. There 133 are more details available for each of the sub-systems. The [consensus protocol](/docs/internals/consensus), 134 [gossip protocol](/docs/internals/gossip), and [scheduler design](/docs/internals/scheduling/scheduling) 135 are all documented in more detail. 136 137 For other details, either consult the code, ask in IRC or reach out to the mailing list.