package allocator

import "github.com/docker/swarmkit/manager/allocator"

Package allocator aims to manage allocation of different cluster-wide resources on behalf of the manager. In particular, it manages a set of independent allocator processes which can mostly execute concurrently with only a minimal need for coordination.

One of the instances where it needs coordination is when to move a task to ALLOCATED state. Since a task can move to ALLOCATED state only when all task allocators have completed their service of allocation, they all have to agree on that. The way this achieved in `allocator` is by creating a `taskBallot` to which all task allocators register themselves as mandatory voters. For each task that needs allocation, each allocator indepdently votes to indicate the completion of their allocation. Once all registered voters have voted then the task is moved to ALLOCATED state.

Other than the coordination needed for task ALLOCATED state, all the allocators function fairly indepdently.

Index ¶

• type Allocator
• func New(store *store.MemoryStore) (*Allocator, error)
• func (a *Allocator) Run(ctx context.Context) error
• func (a *Allocator) Stop()

Types ¶

type Allocator ¶

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

Allocator controls how the allocation stage in the manager is handled.

func New ¶

func New(store *store.MemoryStore) (*Allocator, error)

New returns a new instance of Allocator for use during allocation stage of the manager.

func (*Allocator) Run ¶

func (a *Allocator) Run(ctx context.Context) error

Run starts all allocator go-routines and waits for Stop to be called.

func (*Allocator) Stop ¶

func (a *Allocator) Stop()

Stop stops the allocator

Source Files ¶

allocator.go doc.go network.go

Directories ¶