scaledecider

type ScaleDecider ¶

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

ScaleDecider makes decisions based on the following assumptions:

1. All pending tasks cannot fit into all agents when receiving the snapshots from the scheduler, i.e. we need to launch new agents to fit the pending tasks.
2. All tasks, agents, and instances don't have empty identifiers.
3. All tasks, agents, and instances are not duplicated.

ScaleDecider ignores the agents that cannot be associated with any instances. ScaleDecider considers the following two cases:

1. Instances that can be associated with agents.
2. Instances that cannot be associated with agents. There are several possible causes: a. The provider is starting up the instances. b. The instances are already running but agents on them are starting up. c. The agents are disconnected to the master due to misconfiguration or some unknown reason.

func New(
	resourcePool string,
	maxIdlePeriod, maxStartingPeriod,
	maxDisconnectPeriod time.Duration,
	minInstanceNum int,
	maxInstanceNum int,
	db db.DB,
) *ScaleDecider

func (s *ScaleDecider) CalculateInstanceStates()

func (s *ScaleDecider) CalculateNumInstancesToLaunch() int

func (s *ScaleDecider) FindInstancesToTerminate() sproto.TerminateDecision

func (s *ScaleDecider) RecordInstanceStats(slots int) error

func (s *ScaleDecider) UpdateInstanceSnapshot(instances []*model.Instance) bool

func (s *ScaleDecider) UpdateInstancesEndStats(instIDs []string) error

func (s *ScaleDecider) UpdateScalingInfo(info *sproto.ScalingInfo)