aggregate

Documentation

Index

• func AddHookAfterTableDropped(hk HookAfterTableDropped)
• func AddHookAfterTableStatEmitted(hk HookAfterTableStatEmitted)
• func AllMetrics() (res []string)
• func Start(tom *tomb.Tomb)
• type ClusterStats
  • func SnapshotClusterStats() []ClusterStats
• type HookAfterTableDropped
• type HookAfterTableStatEmitted
• type NodeStat
• type PartitionStats
• type PerfClient
  • func NewPerfClient(metaAddrs []string) *PerfClient
  • func (m *PerfClient) Close()
  • func (m *PerfClient) GetNodeStats(filter string) ([]*NodeStat, error)
  • func (m *PerfClient) GetPartitionStats() ([]*PartitionStats, error)
• type PerfCounter
  • func (p *PerfCounter) String() string
• type PerfSession
  • func NewPerfSession(addr string) *PerfSession
  • func WrapPerf(addr string, session session.NodeSession) *PerfSession
  • func (c *PerfSession) Close()
  • func (c *PerfSession) GetPerfCounters(filter string) ([]*PerfCounter, error)
• type TableStats
• type TableStatsAggregator
  • func NewTableStatsAggregator(metaAddrs []string) TableStatsAggregator

Functions

func AddHookAfterTableDropped

func AddHookAfterTableDropped(hk HookAfterTableDropped)

AddHookAfterTableDropped adds a hook of event that a table is dropped.

func AddHookAfterTableStatEmitted

func AddHookAfterTableStatEmitted(hk HookAfterTableStatEmitted)

AddHookAfterTableStatEmitted adds a hook of event that a new TableStats is generated.

func AllMetrics

func AllMetrics() (res []string)

AllMetrics returns metrics tracked within this collector. The sets of metrics from cluster level and table level are completely equal.

func Start

func Start(tom *tomb.Tomb)

Start looping for metrics aggregation

Types

type ClusterStats

type ClusterStats struct {
	Timestamp time.Time

	Stats map[string]float64
}

ClusterStats is the aggregated metrics for all the TableStats in this cluster. For example, 3 tables with "write_qps" [25, 70, 100] are summed up to `Stats: {"write_qps" : 195}`.

func SnapshotClusterStats

func SnapshotClusterStats() []ClusterStats

SnapshotClusterStats takes a snapshot from the history. The returned array is ordered by time.

type HookAfterTableDropped

type HookAfterTableDropped func(appID int)

HookAfterTableDropped is a hook of event that a table is dropped.

type HookAfterTableStatEmitted

type HookAfterTableStatEmitted func(stats []TableStats, allStats ClusterStats)

HookAfterTableStatEmitted is a hook of event that new TableStats are generated. Each call of the hook handles a batch of tables.

type NodeStat

type NodeStat struct {
	// Address of the replica node.
	Addr string

	// perfCounter's name -> the value.
	Stats map[string]float64
}

NodeStat contains the stats of a replica node.

type PartitionStats

type PartitionStats struct {
	Gpid base.Gpid

	// Address of the primary where this partition locates.
	Addr string

	// perfCounter's name -> the value.
	Stats map[string]float64
}

PartitionStats is a set of metrics retrieved from this partition.

type PerfClient

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

PerfClient manages sessions to all replica nodes.

func NewPerfClient

func NewPerfClient(metaAddrs []string) *PerfClient

NewPerfClient returns an instance of PerfClient.

func (*PerfClient) Close

func (m *PerfClient) Close()

Close release the resources.

func (*PerfClient) GetNodeStats

func (m *PerfClient) GetNodeStats(filter string) ([]*NodeStat, error)

GetNodeStats retrieves all the stats matched with `filter` from replica nodes.

func (*PerfClient) GetPartitionStats

func (m *PerfClient) GetPartitionStats() ([]*PartitionStats, error)

GetPartitionStats retrieves all the partition stats from replica nodes. NOTE: Only the primaries are counted.

type PerfCounter

type PerfCounter struct {
	Name  string
	Value float64
}

PerfCounter is a Pegasus perf-counter.

func (*PerfCounter) String

func (p *PerfCounter) String() string

type PerfSession

type PerfSession struct {
	session.NodeSession

	Address string
}

PerfSession is a client to get perf-counters from a Pegasus ReplicaServer.

func NewPerfSession

func NewPerfSession(addr string) *PerfSession

NewPerfSession returns an instance of PerfSession.

func WrapPerf

func WrapPerf(addr string, session session.NodeSession) *PerfSession

WrapPerf returns an instance of PerfSession using an existed session.

func (*PerfSession) Close

func (c *PerfSession) Close()

Close terminates the session to replica.

func (*PerfSession) GetPerfCounters

func (c *PerfSession) GetPerfCounters(filter string) ([]*PerfCounter, error)

GetPerfCounters retrieves all perf-counters matched with `filter` from the remote node.

type TableStats

type TableStats struct {
	TableName string
	AppID     int

	Partitions map[int]*PartitionStats

	// the time when the stats was generated
	Timestamp time.Time

	// The aggregated value of table metrics.
	// perfCounter's name -> the value.
	Stats map[string]float64
}

TableStats has the aggregated metrics for this table.

type TableStatsAggregator

type TableStatsAggregator interface {
	Aggregate() (map[int32]*TableStats, *ClusterStats, error)

	Close()
}

TableStatsAggregator aggregates the metric on each partition into table-level metrics. It's reponsible for all tables in the pegasus cluster. After all TableStats have been collected, TableStatsAggregator sums them up into a ClusterStats. Users of this pacakage can use the hooks to watch every changes of the stats.

func NewTableStatsAggregator

func NewTableStatsAggregator(metaAddrs []string) TableStatsAggregator

NewTableStatsAggregator returns a TableStatsAggregator instance.