github.com/Ilhicas/nomad@v1.0.4-0.20210304152020-e86851182bc3/website/content/docs/internals/scheduling/scheduling.mdx (about) 1 --- 2 layout: docs 3 page_title: Scheduling 4 sidebar_title: Internals 5 description: Learn about how scheduling works in Nomad. 6 --- 7 8 # Scheduling in Nomad 9 10 [![Nomad Data Model][img-data-model]][img-data-model] 11 12 There are four primary "nouns" in Nomad; jobs, nodes, allocations, and 13 evaluations. Jobs are submitted by users and represent a _desired state_. A job 14 is a declarative description of tasks to run which are bounded by constraints 15 and require resources. Tasks can be scheduled on nodes in the cluster running 16 the Nomad client. The mapping of tasks in a job to clients is done using 17 allocations. An allocation is used to declare that a set of tasks in a job 18 should be run on a particular node. Scheduling is the process of determining 19 the appropriate allocations and is done as part of an evaluation. 20 21 An evaluation is created any time the external state, either desired or 22 emergent, changes. The desired state is based on jobs, meaning the desired 23 state changes if a new job is submitted, an existing job is updated, or a job 24 is deregistered. The emergent state is based on the client nodes, and so we 25 must handle the failure of any clients in the system. These events trigger the 26 creation of a new evaluation, as Nomad must _evaluate_ the state of the world 27 and reconcile it with the desired state. 28 29 This diagram shows the flow of an evaluation through Nomad: 30 31 [![Nomad Evaluation Flow][img-eval-flow]][img-eval-flow] 32 33 The lifecycle of an evaluation begins with an event causing the evaluation to 34 be created. Evaluations are created in the `pending` state and are enqueued 35 into the evaluation broker. There is a single evaluation broker which runs on 36 the leader server. The evaluation broker is used to manage the queue of pending 37 evaluations, provide priority ordering, and ensure at least once delivery. 38 39 Nomad servers run scheduling workers, defaulting to one per CPU core, which are 40 used to process evaluations. The workers dequeue evaluations from the broker, 41 and then invoke the appropriate scheduler as specified by the job. Nomad ships 42 with a `service` scheduler that optimizes for long-lived services, a `batch` 43 scheduler that is used for fast placement of batch jobs, a `system` scheduler 44 that is used to run jobs on every node, and a `core` scheduler which is used 45 for internal maintenance. 46 47 Schedulers are responsible for processing an evaluation and generating an 48 allocation _plan_. The plan is the set of allocations to evict, update, or 49 create. The specific logic used to generate a plan may vary by scheduler, but 50 generally the scheduler needs to first reconcile the desired state with the 51 real state to determine what must be done. New allocations need to be placed 52 and existing allocations may need to be updated, migrated, or stopped. 53 54 Placing allocations is split into two distinct phases, feasibility checking and 55 ranking. In the first phase the scheduler finds nodes that are feasible by 56 filtering unhealthy nodes, those missing necessary drivers, and those failing 57 the specified constraints. 58 59 The second phase is ranking, where the scheduler scores feasible nodes to find 60 the best fit. Scoring is primarily based on bin packing, which is used to 61 optimize the resource utilization and density of applications, but is also 62 augmented by affinity and anti-affinity rules. Nomad automatically applies a job 63 anti-affinity rule which discourages colocating multiple instances of a task 64 group. The combination of this anti-affinity and bin packing optimizes for 65 density while reducing the probability of correlated failures. 66 67 Once the scheduler has ranked enough nodes, the highest ranking node is 68 selected and added to the allocation plan. 69 70 When planning is complete, the scheduler submits the plan to the leader which 71 adds the plan to the plan queue. The plan queue manages pending plans, provides 72 priority ordering, and allows Nomad to handle concurrency races. Multiple 73 schedulers are running in parallel without locking or reservations, making 74 Nomad optimistically concurrent. As a result, schedulers might overlap work on 75 the same node and cause resource over-subscription. The plan queue allows the 76 leader node to protect against this and do partial or complete rejections of a 77 plan. 78 79 As the leader processes plans, it creates allocations when there is no conflict 80 and otherwise informs the scheduler of a failure in the plan result. The plan 81 result provides feedback to the scheduler, allowing it to terminate or explore 82 alternate plans if the previous plan was partially or completely rejected. 83 84 Once the scheduler has finished processing an evaluation, it updates the status 85 of the evaluation and acknowledges delivery with the evaluation broker. This 86 completes the lifecycle of an evaluation. Allocations that were created, 87 modified or deleted as a result will be picked up by client nodes and will 88 begin execution. 89 90 [omega]: https://research.google.com/pubs/pub41684.html 91 [borg]: https://research.google.com/pubs/pub43438.html 92 [img-data-model]: /img/nomad-data-model.png 93 [img-eval-flow]: /img/nomad-evaluation-flow.png