deploymentwatcher

Documentation

Overview

deploymentwatcher creates and tracks Deployments, which hold meta data describing the process of upgrading a running job to a new set of Allocations. This encompasses settings for canary deployments and blue/green rollouts.

- The watcher is only enabled on the active raft leader. - func (w *deploymentWatcher) watch() is the main deploymentWatcher process

Index

• Constants
• type AllocUpdateBatcher
  • func NewAllocUpdateBatcher(ctx context.Context, batchDuration time.Duration, raft DeploymentRaftEndpoints) *AllocUpdateBatcher
  • func (b *AllocUpdateBatcher) CreateUpdate(allocs map[string]*structs.DesiredTransition, eval *structs.Evaluation) *BatchFuture
• type BatchFuture
  • func NewBatchFuture() *BatchFuture
  • func (f *BatchFuture) Results() (uint64, error)
  • func (f *BatchFuture) Set(index uint64, err error)
• type DeploymentRPC
• type DeploymentRaftEndpoints
• type JobRPC
• type Watcher
  • func NewDeploymentsWatcher(logger log.Logger, raft DeploymentRaftEndpoints, deploymentRPC DeploymentRPC, ...) *Watcher
  • func (w *Watcher) CancelDeployment(req *structs.DeploymentCancelRequest, resp *structs.DeploymentUpdateResponse) error
  • func (w *Watcher) FailDeployment(req *structs.DeploymentFailRequest, resp *structs.DeploymentUpdateResponse) error
  • func (w *Watcher) PauseDeployment(req *structs.DeploymentPauseRequest, resp *structs.DeploymentUpdateResponse) error
  • func (w *Watcher) PromoteDeployment(req *structs.DeploymentPromoteRequest, resp *structs.DeploymentUpdateResponse) error
  • func (w *Watcher) RunDeployment(req *structs.DeploymentRunRequest, resp *structs.DeploymentUpdateResponse) error
  • func (w *Watcher) SetAllocHealth(req *structs.DeploymentAllocHealthRequest, ...) error
  • func (w *Watcher) SetEnabled(enabled bool, state *state.StateStore)
  • func (w *Watcher) UnblockDeployment(req *structs.DeploymentUnblockRequest, resp *structs.DeploymentUpdateResponse) error

Constants

const (
	// LimitStateQueriesPerSecond is the number of state queries allowed per
	// second
	LimitStateQueriesPerSecond = 100.0

	// CrossDeploymentUpdateBatchDuration is the duration in which allocation
	// desired transition and evaluation creation updates are batched across
	// all deployment watchers before committing to Raft.
	CrossDeploymentUpdateBatchDuration = 250 * time.Millisecond
)

Types

type AllocUpdateBatcher

type AllocUpdateBatcher struct {
	// contains filtered or unexported fields
}

AllocUpdateBatcher is used to batch the updates to the desired transitions of allocations and the creation of evals.

func NewAllocUpdateBatcher

func NewAllocUpdateBatcher(ctx context.Context, batchDuration time.Duration, raft DeploymentRaftEndpoints) *AllocUpdateBatcher

NewAllocUpdateBatcher returns an AllocUpdateBatcher that uses the passed raft endpoints to create the allocation desired transition updates and new evaluations and exits the batcher when the passed exit channel is closed.

func (*AllocUpdateBatcher) CreateUpdate

func (b *AllocUpdateBatcher) CreateUpdate(allocs map[string]*structs.DesiredTransition, eval *structs.Evaluation) *BatchFuture

CreateUpdate batches the allocation desired transition update and returns a future that tracks the completion of the request.

type BatchFuture

type BatchFuture struct {
	// contains filtered or unexported fields
}

BatchFuture is a future that can be used to retrieve the index the eval was created at or any error in the creation process

func NewBatchFuture

func NewBatchFuture() *BatchFuture

NewBatchFuture returns a new BatchFuture

func (*BatchFuture) Results

func (f *BatchFuture) Results() (uint64, error)

Results returns the creation index and any error.

func (*BatchFuture) Set

func (f *BatchFuture) Set(index uint64, err error)

Set sets the results of the future, unblocking any client.

type DeploymentRPC

type DeploymentRPC interface{}

DeploymentRPC and JobRPC hold methods for interacting with peer regions in enterprise edition.

type DeploymentRaftEndpoints

type DeploymentRaftEndpoints interface {
	// UpsertJob is used to upsert a job
	UpsertJob(job *structs.Job) (uint64, error)

	// UpdateDeploymentStatus is used to make a deployment status update
	// and potentially create an evaluation.
	UpdateDeploymentStatus(u *structs.DeploymentStatusUpdateRequest) (uint64, error)

	// UpdateDeploymentPromotion is used to promote canaries in a deployment
	UpdateDeploymentPromotion(req *structs.ApplyDeploymentPromoteRequest) (uint64, error)

	// UpdateDeploymentAllocHealth is used to set the health of allocations in a
	// deployment
	UpdateDeploymentAllocHealth(req *structs.ApplyDeploymentAllocHealthRequest) (uint64, error)

	// UpdateAllocDesiredTransition is used to update the desired transition
	// for allocations.
	UpdateAllocDesiredTransition(req *structs.AllocUpdateDesiredTransitionRequest) (uint64, error)
}

DeploymentRaftEndpoints exposes the deployment watcher to a set of functions to apply data transforms via Raft.

type JobRPC

type JobRPC interface{}

type Watcher

type Watcher struct {
	// contains filtered or unexported fields
}

Watcher is used to watch deployments and their allocations created by the scheduler and trigger the scheduler when allocation health transitions.

func NewDeploymentsWatcher

func NewDeploymentsWatcher(logger log.Logger,
	raft DeploymentRaftEndpoints,
	deploymentRPC DeploymentRPC, jobRPC JobRPC,
	stateQueriesPerSecond float64,
	updateBatchDuration time.Duration,
) *Watcher

NewDeploymentsWatcher returns a deployments watcher that is used to watch deployments and trigger the scheduler as needed.

func (*Watcher) CancelDeployment

func (w *Watcher) CancelDeployment(req *structs.DeploymentCancelRequest, resp *structs.DeploymentUpdateResponse) error

CancelDeployment is used to cancel a multiregion deployment. In single-region deployments, the deploymentwatcher has sole responsibility to cancel deployments so this RPC is never used.

func (*Watcher) FailDeployment

func (w *Watcher) FailDeployment(req *structs.DeploymentFailRequest, resp *structs.DeploymentUpdateResponse) error

FailDeployment is used to fail the deployment.

func (*Watcher) PauseDeployment

func (w *Watcher) PauseDeployment(req *structs.DeploymentPauseRequest, resp *structs.DeploymentUpdateResponse) error

PauseDeployment is used to toggle the pause state on a deployment. If the deployment is being unpaused, an evaluation is created.

func (*Watcher) PromoteDeployment

func (w *Watcher) PromoteDeployment(req *structs.DeploymentPromoteRequest, resp *structs.DeploymentUpdateResponse) error

PromoteDeployment is used to promote a deployment. If promote is false, deployment is marked as failed. Otherwise the deployment is updated and an evaluation is created.

func (*Watcher) RunDeployment

func (w *Watcher) RunDeployment(req *structs.DeploymentRunRequest, resp *structs.DeploymentUpdateResponse) error

RunDeployment is used to run a pending multiregion deployment. In single-region deployments, the pending state is unused.

func (*Watcher) SetAllocHealth

func (w *Watcher) SetAllocHealth(req *structs.DeploymentAllocHealthRequest, resp *structs.DeploymentUpdateResponse) error

SetAllocHealth is used to set the health of allocations for a deployment. If there are any unhealthy allocations, the deployment is updated to be failed. Otherwise the allocations are updated and an evaluation is created.

func (*Watcher) SetEnabled

func (w *Watcher) SetEnabled(enabled bool, state *state.StateStore)

SetEnabled is used to control if the watcher is enabled. The watcher should only be enabled on the active leader. When being enabled the state is passed in as it is no longer valid once a leader election has taken place.

func (*Watcher) UnblockDeployment

func (w *Watcher) UnblockDeployment(req *structs.DeploymentUnblockRequest, resp *structs.DeploymentUpdateResponse) error

UnblockDeployment is used to unblock a multiregion deployment. In single-region deployments, the blocked state is unused.