ipfs-cluster/consensus/crdt/config.go

package crdt

import (
	"encoding/json"
	"errors"
	"fmt"
	"time"

	"github.com/ipfs-cluster/ipfs-cluster/api"
	"github.com/ipfs-cluster/ipfs-cluster/config"

	peer "github.com/libp2p/go-libp2p/core/peer"

	"github.com/kelseyhightower/envconfig"
)

var configKey = "crdt"
var envConfigKey = "cluster_crdt"

// Default configuration values
var (
	DefaultClusterName          = "ipfs-cluster"
	DefaultPeersetMetric        = "ping"
	DefaultDatastoreNamespace   = "/c" // from "/crdt"
	DefaultRebroadcastInterval  = time.Minute
	DefaultTrustedPeers         = []peer.ID{}
	DefaultTrustAll             = true
	DefaultBatchingMaxQueueSize = 50000
	DefaultRepairInterval       = time.Hour
)

// BatchingConfig configures parameters for batching multiple pins in a single
// CRDT-put operation.
//
// MaxBatchSize will trigger a commit whenever the number of pins in the batch
// reaches the limit.
//
// MaxBatchAge will trigger a commit when the oldest update in the batch
// reaches it. Setting both values to 0 means batching is disabled.
//
// MaxQueueSize specifies how many items can be waiting to be batched before
// the LogPin/Unpin operations block.
type BatchingConfig struct {
	MaxBatchSize int
	MaxBatchAge  time.Duration
	MaxQueueSize int
}

// Config is the configuration object for Consensus.
type Config struct {
	config.Saver

	hostShutdown bool

	// The topic we wish to subscribe to
	ClusterName string

	// TrustAll specifies whether we should trust all peers regardless of
	// the TrustedPeers contents.
	TrustAll bool

	// Any update received from a peer outside this set is ignored and not
	// forwarded. Trusted peers can also access additional RPC endpoints
	// for this peer that are forbidden for other peers.
	TrustedPeers []peer.ID

	// Specifies whether to batch CRDT updates for increased
	// performance.
	Batching BatchingConfig

	// The interval before re-announcing the current state
	// to the network when no activity is observed.
	RebroadcastInterval time.Duration

	// The name of the metric we use to obtain the peerset (every peer
	// with valid metric of this type is part of it).
	PeersetMetric string

	// All keys written to the datastore will be namespaced with this prefix
	DatastoreNamespace string

	// How often the underlying crdt store triggers a repair when the
	// datastore is marked dirty.
	RepairInterval time.Duration

	// Tracing enables propagation of contexts across binary boundaries.
	Tracing bool
}

type batchingConfigJSON struct {
	MaxBatchSize int    `json:"max_batch_size"`
	MaxBatchAge  string `json:"max_batch_age"`
	MaxQueueSize int    `json:"max_queue_size,omitempty"`
}

type jsonConfig struct {
	ClusterName         string             `json:"cluster_name"`
	TrustedPeers        []string           `json:"trusted_peers"`
	Batching            batchingConfigJSON `json:"batching"`
	RepairInterval      string             `json:"repair_interval"`
	RebroadcastInterval string             `json:"rebroadcast_interval,omitempty"`

	PeersetMetric      string `json:"peerset_metric,omitempty"`
	DatastoreNamespace string `json:"datastore_namespace,omitempty"`
}

// ConfigKey returns the section name for this type of configuration.
func (cfg *Config) ConfigKey() string {
	return configKey
}

// Validate returns an error if the configuration has invalid values.
func (cfg *Config) Validate() error {
	if cfg.ClusterName == "" {
		return errors.New("crdt.cluster_name cannot be empty")
	}

	if cfg.PeersetMetric == "" {
		return errors.New("crdt.peerset_metric needs a name")
	}

	if cfg.RebroadcastInterval <= 0 {
		return errors.New("crdt.rebroadcast_interval is invalid")
	}

	if cfg.Batching.MaxQueueSize <= 0 {
		return errors.New("crdt.batching.max_queue_size is invalid")
	}

	if cfg.RepairInterval < 0 {
		return errors.New("crdt.repair_interval is invalid")
	}
	return nil
}

// LoadJSON takes a raw JSON slice and sets all the configuration fields.
func (cfg *Config) LoadJSON(raw []byte) error {
	jcfg := &jsonConfig{}
	err := json.Unmarshal(raw, jcfg)
	if err != nil {
		return fmt.Errorf("error unmarshaling %s config", configKey)
	}

	cfg.Default()

	return cfg.applyJSONConfig(jcfg)
}

func (cfg *Config) applyJSONConfig(jcfg *jsonConfig) error {
	config.SetIfNotDefault(jcfg.ClusterName, &cfg.ClusterName)

	// Whenever we parse JSON, TrustAll is false unless an '*' peer exists
	cfg.TrustAll = false
	cfg.TrustedPeers = []peer.ID{}

	for _, p := range jcfg.TrustedPeers {
		if p == "*" {
			cfg.TrustAll = true
			cfg.TrustedPeers = []peer.ID{}
			break
		}
		pid, err := peer.Decode(p)
		if err != nil {
			return fmt.Errorf("error parsing trusted peers: %s", err)
		}
		cfg.TrustedPeers = append(cfg.TrustedPeers, pid)
	}

	cfg.Batching.MaxBatchSize = jcfg.Batching.MaxBatchSize

	config.SetIfNotDefault(jcfg.Batching.MaxQueueSize, &cfg.Batching.MaxQueueSize)
	config.SetIfNotDefault(jcfg.PeersetMetric, &cfg.PeersetMetric)
	config.SetIfNotDefault(jcfg.DatastoreNamespace, &cfg.DatastoreNamespace)
	config.ParseDurations(
		"crdt",
		&config.DurationOpt{Duration: jcfg.RebroadcastInterval, Dst: &cfg.RebroadcastInterval, Name: "rebroadcast_interval"},
		&config.DurationOpt{Duration: jcfg.Batching.MaxBatchAge, Dst: &cfg.Batching.MaxBatchAge, Name: "max_batch_age"},
		&config.DurationOpt{Duration: jcfg.RepairInterval, Dst: &cfg.RepairInterval, Name: "repair_interval"},
	)
	return cfg.Validate()
}

// ToJSON returns the JSON representation of this configuration.
func (cfg *Config) ToJSON() ([]byte, error) {
	jcfg := cfg.toJSONConfig()

	return config.DefaultJSONMarshal(jcfg)
}

func (cfg *Config) toJSONConfig() *jsonConfig {
	jcfg := &jsonConfig{
		ClusterName:         cfg.ClusterName,
		PeersetMetric:       "",
		RebroadcastInterval: "",
	}

	if cfg.TrustAll {
		jcfg.TrustedPeers = []string{"*"}
	} else {
		jcfg.TrustedPeers = api.PeersToStrings(cfg.TrustedPeers)
	}

	jcfg.Batching.MaxBatchSize = cfg.Batching.MaxBatchSize
	jcfg.Batching.MaxBatchAge = cfg.Batching.MaxBatchAge.String()
	if cfg.Batching.MaxQueueSize != DefaultBatchingMaxQueueSize {
		jcfg.Batching.MaxQueueSize = cfg.Batching.MaxQueueSize
		// otherwise leave as 0/hidden
	}

	if cfg.PeersetMetric != DefaultPeersetMetric {
		jcfg.PeersetMetric = cfg.PeersetMetric
		// otherwise leave empty/hidden
	}

	if cfg.DatastoreNamespace != DefaultDatastoreNamespace {
		jcfg.DatastoreNamespace = cfg.DatastoreNamespace
		// otherwise leave empty/hidden
	}

	if cfg.RebroadcastInterval != DefaultRebroadcastInterval {
		jcfg.RebroadcastInterval = cfg.RebroadcastInterval.String()
	}

	jcfg.RepairInterval = cfg.RepairInterval.String()

	return jcfg
}

// Default sets the configuration fields to their default values.
func (cfg *Config) Default() error {
	cfg.ClusterName = DefaultClusterName
	cfg.RebroadcastInterval = DefaultRebroadcastInterval
	cfg.PeersetMetric = DefaultPeersetMetric
	cfg.DatastoreNamespace = DefaultDatastoreNamespace
	cfg.TrustedPeers = DefaultTrustedPeers
	cfg.TrustAll = DefaultTrustAll
	cfg.Batching = BatchingConfig{
		MaxBatchSize: 0,
		MaxBatchAge:  0,
		MaxQueueSize: DefaultBatchingMaxQueueSize,
	}
	cfg.RepairInterval = DefaultRepairInterval
	return nil
}

// ApplyEnvVars fills in any Config fields found
// as environment variables.
func (cfg *Config) ApplyEnvVars() error {
	jcfg := cfg.toJSONConfig()

	err := envconfig.Process(envConfigKey, jcfg)
	if err != nil {
		return err
	}

	return cfg.applyJSONConfig(jcfg)
}

// ToDisplayJSON returns JSON config as a string.
func (cfg *Config) ToDisplayJSON() ([]byte, error) {
	return config.DisplayJSON(cfg.toJSONConfig())
}

func (cfg *Config) batchingEnabled() bool {
	return cfg.Batching.MaxBatchSize > 0 &&
		cfg.Batching.MaxBatchAge > 0
}