Documentation
¶
Overview ¶
Package config contains functions and types used for managing dragonboat's configurations.
Index ¶
Examples ¶
Constants ¶
View Source
const ( // NoCompression is the CompressionType value used to indicate not to use // any compression. NoCompression = pb.NoCompression // Snappy is the CompressionType value used to indicate that google snappy // is used for data compression. Snappy = pb.Snappy )
Variables ¶
This section is empty.
Functions ¶
func IsValidAddress ¶
IsValidAddress returns a boolean value indicating whether the input address is valid.
Types ¶
type CompressionType ¶
type CompressionType = pb.CompressionType
CompressionType is the type of the compression.
type Config ¶
type Config struct {
// NodeID is a non-zero value used to identify a node within a Raft cluster.
NodeID uint64
// ClusterID is the unique value used to identify a Raft cluster.
ClusterID uint64
// CheckQuorum specifies whether the leader node should periodically check
// non-leader node status and step down to become a follower node when it no
// longer has the quorum.
CheckQuorum bool
// Whether to use PreVote for this node. PreVote is described in the section
// 9.7 of the raft thesis.
PreVote bool
// ElectionRTT is the minimum number of message RTT between elections. Message
// RTT is defined by NodeHostConfig.RTTMillisecond. The Raft paper suggests it
// to be a magnitude greater than HeartbeatRTT, which is the interval between
// two heartbeats. In Raft, the actual interval between elections is
// randomized to be between ElectionRTT and 2 * ElectionRTT.
//
// As an example, assuming NodeHostConfig.RTTMillisecond is 100 millisecond,
// to set the election interval to be 1 second, then ElectionRTT should be set
// to 10.
//
// When CheckQuorum is enabled, ElectionRTT also defines the interval for
// checking leader quorum.
ElectionRTT uint64
// HeartbeatRTT is the number of message RTT between heartbeats. Message
// RTT is defined by NodeHostConfig.RTTMillisecond. The Raft paper suggest the
// heartbeat interval to be close to the average RTT between nodes.
//
// As an example, assuming NodeHostConfig.RTTMillisecond is 100 millisecond,
// to set the heartbeat interval to be every 200 milliseconds, then
// HeartbeatRTT should be set to 2.
HeartbeatRTT uint64
// SnapshotEntries defines how often the state machine should be snapshotted
// automcatically. It is defined in terms of the number of applied Raft log
// entries. SnapshotEntries can be set to 0 to disable such automatic
// snapshotting.
//
// When SnapshotEntries is set to N, it means a snapshot is created for
// roughly every N applied Raft log entries (proposals). This also implies
// that sending N log entries to a follower is more expensive than sending a
// snapshot.
//
// Once a snapshot is generated, Raft log entries covered by the new snapshot
// can be compacted. This involves two steps, redundant log entries are first
// marked as deleted, then they are physically removed from the underlying
// storage when a LogDB compaction is issued at a later stage. See the godoc
// on CompactionOverhead for details on what log entries are actually removed
// and compacted after generating a snapshot.
//
// Once automatic snapshotting is disabled by setting the SnapshotEntries
// field to 0, users can still use NodeHost's RequestSnapshot or
// SyncRequestSnapshot methods to manually request snapshots.
SnapshotEntries uint64
// CompactionOverhead defines the number of most recent entries to keep after
// each Raft log compaction. Raft log compaction is performance automatically
// every time when a snapshot is created.
//
// For example, when a snapshot is created at let's say index 10,000, then all
// Raft log entries with index <= 10,000 can be removed from that node as they
// have already been covered by the created snapshot image. This frees up the
// maximum storage space but comes at the cost that the full snapshot will
// have to be sent to the follower if the follower requires any Raft log entry
// at index <= 10,000. When CompactionOverhead is set to say 500, Dragonboat
// then compacts the Raft log up to index 9,500 and keeps Raft log entries
// between index (9,500, 1,0000]. As a result, the node can still replicate
// Raft log entries between index (9,500, 1,0000] to other peers and only fall
// back to stream the full snapshot if any Raft log entry with index <= 9,500
// is required to be replicated.
CompactionOverhead uint64
// OrderedConfigChange determines whether Raft membership change is enforced
// with ordered config change ID.
//
// When set to true, ConfigChangeIndex is required for membership change
// requests. This behaves like an optimistic write lock forcing clients to
// linearize membership change requests explicitly. (recommended)
//
// When set to false (default), ConfigChangeIndex is ignored for membership
// change requests. This may cause a client to request a membership change
// based on stale membership data.
OrderedConfigChange bool
// MaxInMemLogSize is the target size in bytes allowed for storing in memory
// Raft logs on each Raft node. In memory Raft logs are the ones that have
// not been applied yet.
// MaxInMemLogSize is a target value implemented to prevent unbounded memory
// growth, it is not for precisely limiting the exact memory usage.
// When MaxInMemLogSize is 0, the target is set to math.MaxUint64. When
// MaxInMemLogSize is set and the target is reached, error will be returned
// when clients try to make new proposals.
// MaxInMemLogSize is recommended to be significantly larger than the biggest
// proposal you are going to use.
MaxInMemLogSize uint64
// SnapshotCompressionType is the compression type to use for compressing
// generated snapshot data. No compression is used by default.
SnapshotCompressionType CompressionType
// EntryCompressionType is the compression type to use for compressing the
// payload of user proposals. When Snappy is used, the maximum proposal
// payload allowed is roughly limited to 3.42GBytes. No compression is used
// by default.
EntryCompressionType CompressionType
// DisableAutoCompactions disables auto compaction used for reclaiming Raft
// log entry storage spaces. By default, compaction request is issued every
// time when a snapshot is created, this helps to reclaim disk spaces as
// soon as possible at the cost of immediate higher IO overhead. Users can
// disable such auto compactions and use NodeHost.RequestCompaction to
// manually request such compactions when necessary.
DisableAutoCompactions bool
// IsNonVoting indicates whether this is a non-voting Raft node. Described as
// non-voting members in the section 4.2.1 of Diego Ongaro's thesis, they are
// used to allow a new node to join the cluster and catch up with other
// existing ndoes without impacting the availability. Extra non-voting nodes
// can also be introduced to serve read-only requests.
IsNonVoting bool
// IsObserver indicates whether this is a non-voting Raft node without voting
// power.
//
// Deprecated: use IsNonVoting instead.
IsObserver bool
// IsWitness indicates whether this is a witness Raft node without actual log
// replication and do not have state machine. It is mentioned in the section
// 11.7.2 of Diego Ongaro's thesis.
//
// Witness support is currently experimental.
IsWitness bool
}
Config is used to configure Raft nodes.
type EngineConfig ¶
type EngineConfig struct {
// ExecShards is the number of execution shards in the first stage of the
// execution engine. Default value is 16. Once deployed, this value can not
// be changed later.
ExecShards uint64
// CommitShards is the number of commit shards in the second stage of the
// execution engine. Default value is 16.
CommitShards uint64
// ApplyShards is the number of apply shards in the third stage of the
// execution engine. Default value is 16.
ApplyShards uint64
// SnapshotShards is the number of snapshot shards in the forth stage of the
// execution engine. Default value is 48.
SnapshotShards uint64
// CloseShards is the number of close shards used for closing stopped
// state machines. Default value is 32.
CloseShards uint64
}
EngineConfig is the configuration for the execution engine.
func GetDefaultEngineConfig ¶
func GetDefaultEngineConfig() EngineConfig
GetDefaultEngineConfig returns the default EngineConfig instance.
func (EngineConfig) IsEmpty ¶
func (ec EngineConfig) IsEmpty() bool
IsEmpty returns a boolean value indicating whether EngineConfig is an empty one.
func (EngineConfig) Validate ¶
func (ec EngineConfig) Validate() error
Validate return an error value when the EngineConfig is invalid.
type ExpertConfig ¶
type ExpertConfig struct {
// Engine is the cponfiguration for the execution engine.
Engine EngineConfig
// FS is the filesystem instance used in tests.
FS IFS
// TestNodeHostID is the NodeHostID value to be used by the NodeHost instance.
// This field is expected to be used in tests only.
TestNodeHostID uint64
}
ExpertConfig contains options for expert users who are familiar with the internals of Dragonboat. Users are recommended not to set ExpertConfig unless it is absoloutely necessary.
func GetDefaultExpertConfig ¶
func GetDefaultExpertConfig() ExpertConfig
GetDefaultExpertConfig returns the default ExpertConfig.
type NodeHostConfig ¶
type NodeHostConfig struct {
// DeploymentID is used to determine whether two NodeHost instances belong to
// the same deployment and thus allowed to communicate with each other. This
// helps to prvent accidentially misconfigured NodeHost instances to cause
// data corruption errors by sending out of context messages to unrelated
// Raft nodes.
// For a particular dragonboat based application, you can set DeploymentID
// to the same uint64 value on all production NodeHost instances, then use
// different DeploymentID values on your staging and dev environment. It is
// also recommended to use different DeploymentID values for different
// dragonboat based applications.
// When not set, the default value 0 will be used as the deployment ID and
// thus allowing all NodeHost instances with deployment ID 0 to communicate
// with each other.
DeploymentID uint64
// WALDir is the directory used for storing the WAL of Raft entries. It is
// recommended to use low latency storage such as NVME SSD with power loss
// protection to store such WAL data. Leave WALDir to have zero value will
// have everything stored in NodeHostDir.
WALDir string
// NodeHostDir is where everything else is stored.
NodeHostDir string
// RTTMillisecond defines the average Rround Trip Time (RTT) in milliseconds
// between two NodeHost instances. Such a RTT interval is internally used as
// a logical clock tick, Raft heartbeat and election intervals are both
// defined in term of how many such logical clock ticks (RTT intervals).
// Note that RTTMillisecond is the combined delays between two NodeHost
// instances including all delays caused by network transmission, delays
// caused by NodeHost queuing and processing. As an example, when fully
// loaded, the average Rround Trip Time between two of our NodeHost instances
// used for benchmarking purposes is up to 500 microseconds when the ping time
// between them is 100 microseconds. Set RTTMillisecond to 1 when it is less
// than 1 million in your environment.
RTTMillisecond uint64
// RaftAddress is a DNS name:port or IP:port address used by the transport
// module for exchanging Raft messages, snapshots and metadata between
// NodeHost instances. It should be set to the public address that can be
// accessed from remote NodeHost instances.
//
// When the NodeHostConfig.ListenAddress field is empty, NodeHost listens on
// RaftAddress for incoming Raft messages. When hostname or domain name is
// used, it will be resolved to IPv4 addresses first and Dragonboat listens
// to all resolved IPv4 addresses.
//
// By default, the RaftAddress value is not allowed to change between NodeHost
// restarts. AddressByNodeHostID should be set to true when the RaftAddress
// value might change after restart.
RaftAddress string
// RaftEventListener is the listener for Raft events, such as Raft leadership
// change, exposed to user space. NodeHost uses a single dedicated goroutine
// to invoke all RaftEventListener methods one by one, CPU intensive or IO
// related procedures that can cause long delays should be offloaded to worker
// goroutines managed by users. See the raftio.IRaftEventListener definition
// for more details.
RaftEventListener raftio.IRaftEventListener
// SystemEventsListener allows users to be notified for system events such
// as snapshot creation, log compaction and snapshot streaming. It is usually
// used for testing purposes or for other advanced usages, Dragonboat
// applications are not required to explicitly set this field.
SystemEventListener raftio.ISystemEventListener
// MaxSendQueueSize is the maximum size in bytes of each send queue.
// Once the maximum size is reached, further replication messages will be
// dropped to restrict memory usage. When set to 0, it means the send queue
// size is unlimited.
MaxSendQueueSize uint64
// MaxReceiveQueueSize is the maximum size in bytes of each receive queue.
// Once the maximum size is reached, further replication messages will be
// dropped to restrict memory usage. When set to 0, it means the queue size
// is unlimited.
MaxReceiveQueueSize uint64
// NotifyCommit specifies whether clients should be notified when their
// regular proposals and config change requests are committed. By default,
// commits are not notified, clients are only notified when their proposals
// are both committed and applied.
NotifyCommit bool
// Expert contains options for expert users who are familiar with the internals
// of Dragonboat. Users are recommended not to use this field unless
// absoloutely necessary. It is important to note that any change to this field
// may cause an existing instance unable to restart, it may also cause negative
// performance impacts.
Expert ExpertConfig
}
NodeHostConfig is the configuration used to configure NodeHost instances.
Example ¶
nhc := NodeHostConfig{
WALDir: "/data/wal",
NodeHostDir: "/data/dragonboat-data",
RTTMillisecond: 200,
// RaftAddress is the public address that will be used by others to contact
// this NodeHost instance.
RaftAddress: "node01.raft.company.com:5012",
}
_ = nhc
Output:
func (*NodeHostConfig) GetDeploymentID ¶
func (c *NodeHostConfig) GetDeploymentID() uint64
GetDeploymentID returns the deployment ID to be used.
func (*NodeHostConfig) GetRaftAddressValidator ¶
func (c *NodeHostConfig) GetRaftAddressValidator() RaftAddressValidator
GetRaftAddressValidator creates a RaftAddressValidator based on the specified NodeHostConfig instance.
func (*NodeHostConfig) GetTargetValidator ¶
func (c *NodeHostConfig) GetTargetValidator() TargetValidator
GetTargetValidator returns a TargetValidator based on the specified NodeHostConfig instance.
func (*NodeHostConfig) Prepare ¶
func (c *NodeHostConfig) Prepare() error
Prepare sets the default value for NodeHostConfig.
func (*NodeHostConfig) Validate ¶
func (c *NodeHostConfig) Validate() error
Validate validates the NodeHostConfig instance and return an error when the configuration is considered as invalid.
type RaftAddressValidator ¶
RaftAddressValidator is the validator used to validate user specified RaftAddress values.
type RaftRPCFactoryFunc
deprecated
type RaftRPCFactoryFunc func(NodeHostConfig, raftio.MessageHandler, raftio.ChunkHandler) raftio.ITransport
RaftRPCFactoryFunc is the factory function that creates the transport module instance for exchanging Raft messages between NodeHosts.
Deprecated: Use TransportFactory instead.
type TargetValidator ¶
TargetValidator is the validtor used to validate user specified target values.
type TransportFactory ¶
type TransportFactory interface {
// Create creates a transport module.
Create(NodeHostConfig,
raftio.MessageHandler, raftio.ChunkHandler) raftio.ITransport
// Validate validates the RaftAddress of the NodeHost. When using a custom
// transport module, users are granted full control on what address type to
// use for the NodeHostConfig.RaftAddress field, it can be of the traditional
// IP:Port format or any other form. The Validate method is used to validate
// that a received address is of the valid form.
Validate(string) bool
}
TransportFactory is the interface used for creating custom transport modules.
Source Files
Click to show internal directories.
Click to hide internal directories.