github.com/anth0d/nomad@v0.0.0-20221214183521-ae3a0a2cad06/website/content/docs/concepts/scheduling/scheduling.mdx (about)

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