Cluster Mode
In cluster mode, DolphinDB configuration files are saved under the config directory.
-
controller.cfg: the configuration file for controller nodes
-
agent.cfg: the configuration file for agent nodes
-
cluster.cfg: the configuration file for data nodes
-
cluster.nodes: network information and node mode of each node in the cluster.
For a high-availability cluster, all the configuration files cluster.cfg and cluster.node are synchronously managed by the raft group. After the configuration parameters are modified through the web interface, the interface will automatically synchronize the parameters to the configuration files in the cluster.
Controller
Configuration parameters for controllers can be specified in controller.cfg, or in command line when starting a controller node. localSite and mode must be specified.
Parameter | Description |
---|---|
localSite | IP address, port number and alias of the controller node. |
mode=controller | Node mode. Must specify "controller" here for a controller. |
clusterConfig=cluster.cfg | Configuration file for the cluster. The default value is cluster.cfg. It can only be specified in command line. |
nodesFile=cluster.nodes | Specify IP address, port number, alias and node mode of all nodes in a cluster. It is loaded when the controller is started. The default value is cluster.nodes. It can only be specified in command line. |
dataSync=0 | Whether database logs are forced to persist to disk before the transaction is committed. If dataSync=1, the database log (including redo log, undo log, edit log of data nodes,and edit log of controller nodes) must be written to disk before each transaction is committed. This guarantees that data written to the database is not lost in case of systemcrash or power outage. If dataSync=0 (default), the log files are written to cache before a transaction is committed. The operating system will write the log files to disk at a later time. We may experience loss of data or corruption of the database in case of system crash or power outage. |
dfsMetaDir =/home/DolphinDB/server | The directory to save the metadata of the distributed file system on the controller node. The default value is the DolphinDB home directory specified by parameter home. |
dfsMetaLogFilename =DFSMetaLog | The edit log file for metadata of the distributed file system on the controller node. The default value is DFSMetaLog. |
PublicName | Internet IP or domain name of the controller node. It must be a domain name if enableHttps=true. |
datanodeRestartInterval=0 | It is a non-negative integer of int type. Its unit is second. Defaults to 0, indicating that the system will not automatically start data/compute nodes. Configure this parameter to implement the following features: (1) Automatically start the data/compute node after the controller is started (the agent needs to be started); (2) The data/compute node will be automatically restarted after its offline time is larger than the set value (for a non-graceful shutdown and the agent needs to be started). |
datanodeRestartLevel=CRASH | Specifies the triggering condition for the
controller to automatically restart the data/compute node. It can
take the following values (case sensitive):
|
dfsRecoveryConcurrency | The number of concurrent recovery tasks in a node recovery. The default value is twice the number of all nodes except the agent. |
dfsRebalanceConcurrency | The number of concurrent rebalance tasks in a node rebalance. The default value is twice the number of all nodes except the agent. |
dfsChunkNodeHeartBeatTimeout |
An integer (of INT type) indicating the time interval (in seconds) for marking a data node/compute node/agent as down, i.e., if the controller has not received heartbeat message from a node for more than this time interval, the node will be marked and treated as down by default. The default value is 8. |
The following configuration parameters are for high availability and are specified in controller.cfg.
Parameter | Description |
---|---|
dfsHAMode=Raft | Whether the controller nodes form a raft group. It can take the value of either Raft or None. The default value is None. |
dfsReplicationFactor=2 | The number of all replicas for each data chunk. The default value is 2. |
dfsReplicaReliabilityLevel=0 | Whether multiple replicas can reside on the same
node.
|
dfsRecoveryWaitTime=0 | Length of time (in milliseconds) the controller waits after a data node becomes unavailable before restoring the data to other data nodes. The default value is 0indicating no automatic recovery. |
raftElectionTick=800 | Specify a time interval (in 10ms): [raftElectionTick, 2*raftElectionTick]. After receiving the last heartbeat from the leader, if a follower does not receive the next heartbeat after a random waiting time within the specified interval, it will send a request for leader election. The default value is 800 (8 seconds) and the interval is [8s, 16s]. Note: All controllers of the raft group must have the same configurations. |
The number of online data nodes must be no smaller than the value of dfsReplicationFactor, otherwise an exception will be thrown.
Agent
Configuration parameters for agents can be specified in agent.cfg, or in command line when starting an agent node. All of the following 3 parameters must be specified.
Parameter | Description |
---|---|
localSite | Host address, port number and alias of the local node. |
mode=agent | Node mode. Must specify "agent" here for an agent. |
controllerSite | Host address, port number and alias of the controller site of the agent node. When an agent starts up, it uses this information to contact the controller. It must be identical as localSite of one of the controllers in controller.cfg. |
lanCluster=1 | Whether the cluster is within a LAN (local area network). lanCluster = 1: use UDP for heartbeats; lanCluster = 0: use TCP for heartbeats. Set lanCluster = 0 if the system is deployed in the cloud. |
Data Node
All configuration parameters in the standalone mode can be used for data nodes in the cluster mode. They are specified in cluster.cfg. Please refer to Standalone Mode for more details.
The following table lists the configuration parameters that apply to cluster mode only.
Parameter | Description |
---|---|
lanCluster=true | Whether the cluster is within a LAN (local area network). lanCluster = true (default): use UDP for heartbeats; lanCluster = false: use TCP for heartbeats. Set lanCluster = false if the system is deployed in the cloud. |
allowVolumeCreation=1 | Set the parameter to determine whether a volume can be created automatically if the volume does not exist. The value can be 0 or 1.The default value 1 indicates to create volumes automatically. If the parameter is set to 0, the system will report an error if the volume does not exist. |
volumeUsageThreshold=0.97 | Specifies the upper limit of volume usage on a data node. It is a floating-point value in (0, 1]. The default value is 0.97. When the volume usage on a data node reaches the threshold, new chunks are not allowed to be created on the node, but data can be written to existing chunks. The parameter can only be configured on the controller. |
volumes=/hdd/hdd1/volumes, /hdd/hdd2/volumes, /hdd/hdd3/volumes, /hdd/hdd4/volumes | The directory of data files. If it is not specified, the default directory is <HomeDir>/<nodeAlias>/storage. |
recoveryWorkers = 1 | The number of workers that can be used to recover chunks synchronously in node recovery. The default value is 1. |
memLimitOfQueryResult | The memory limit for a query result. The default value is min(50% * maxMemSize, 8G), and it must be smaller than 80% * maxMemSize. |
memLimitOfTaskGroupResult | In the Map phase of MapReduce, a single query is divided into several tasks, among which the remote tasks are sent to remote nodes in batches (task groups). The configuration parameter is used to set the memory limit of a task group sent from the current node. The default value is min(20% * maxMemSize, 2G),and it must be smaller than 50% * maxMemSize. |
memLimitOfAllTempResults |
Specifies the total memory limit (in GB) for all temporary results generated during SQL distributed queries. The default value is 20% * maxMemSize. If the memory usage exceeds the limit, temporary results are spilled to disk. |
Streaming-related configuration parameters are specified in cluster.cfg.
Parameter | Description |
---|---|
streamingHADir=<HomeDir>/log/streamLog | The directory to keep streaming raft log files. The default value is <HomeDir>/log/streamLog. Each data node should be configured with different streamingHADir. |
streamingHAMode=raft | Enable high-availability for streaming. |
streamingRaftGroups = 2:NODE1:NODE2:NODE3,3:NODE3:NODE4:NODE5 | Information about raft groups. Each raft group is represented by group ID and aliases of data nodes in the group, separated with colon (:). Raft group ID must be an integer greater than 1. Each raft group has at least 3 data nodes. Use comma (,) to seperate multiple raft groups. |
Asynchronous Replication
-
Master Cluster
The parameters clusterReplicationSlaveNum and clusterReplicationMode must be configured.
Parameter Description Node clusterReplicationSlaveNum Specifies the upper limit of slave clusters. The default value is 2. controller clusterReplicationMode Specifies whether the cluster is master or slave cluster. data node clusterReplicationWorkDir Specifies the working directory where data of write tasks on the master cluster is stored. The default value is <HomeDir>/clusterReplication. A larger-sized SSD is recommended. data node clusterReplicationSyncPersistence Specifies a Boolean value indicating whether to persist data of write tasks synchronously or asynchronously. Note: Synchronous persistence ensures data consistency but may affect transaction efficiency, while asynchronous persistence improves performance but may cause data loss in node crashes. data node -
Slave cluster
The parameters clusterReplicationMasterCtl and clusterReplicationMode must be configured.
Parameter Description Node clusterReplicationMasterCtl Specifies ip:port of the controller in the master cluster. For a high-availability (HA) cluster, you can specify any controller in a raft group. controller clusterReplicationMode Specifies whether the cluster is master or slave cluster. data node clusterReplicationExecutionUsername Specifies the username (default=admin) used to execute operations regarding cluster replication. The specified user must have relevant privileges on transactions, otherwise the asynchronous tasks will fail. data node clusterReplicationExecutionPassword Specifies the user password (default=123456) for clusterReplicationExecutionUsername. Note: since version 2.00.13/3.00.1, the user is not required to specify this parameter.data node clusterReplicationQueue A positive integer specifying the number of execution queues. The default value is 4 times the number of data nodes. controller clusterReplicationWorkerNum The number of workers (task execution threads) on each data node. The default value is max(workerNum/4, 4). data node
Tiered Storage
For more information, see Tiered Storage.
To enable tiered storage, configure the storage path for cold data first.
Parameter | Description |
---|---|
coldVolumes= [file://home/mypath/hdd](file://home/mypath/hdd), s3://bucket1/data | Configures the storage path of cold data. After the tiered storage is enabled with functions moveHotDataToColdVolume and setRetentionPolicy, cold data will be moved from volumes to coldVolumes. |
To store expired data in the cloud, DolphinDB provides the following AWS configurations.
Parameters | Description |
---|---|
s3AccessKeyId | ID of an Amazon S3 account |
s3SecretAccessKey | Secret access key of an Amazon S3 account |
s3Region | Region of an Amazon S3 bucket |
s3Endpoint | Endpoint to access Amazon S3 |
-
When configuring s3Endpoint, you can add a BOOLEAN value after the endpoint to indicate whether to access the endpoint via HTTP protocol (true) or HTTPS protocol (false, default). For example,
s3Endpoint=192.168.1.160:980,true
indicates accessing 192.168.1.160:980 via HTTP protocol. -
With s3Endpoint configured, it is not necessary to configure s3Region.
Parameter Specification
In cluster.cfg, we can specify configuration parameter values in the following 4 ways:
-
Node alias qualified configuration parameters. Node aliases are defined in cluster.nodes.
nodeA.volumes = /DFSRoot/nodeA nodeB.volumes = /DFSRoot/nodeB
-
(Node alias + wildcard character ("?" or "%")) qualified configuration parameters. "?" represents a single character; "%" represents 0, 1 or multiple characters.
%8821.volumes = /DFSRoot/data8821 %8822.volumes = /DFSRoot/data8822 DFSnode?.maxMemSize=16
-
Use macro variable
<ALIAS>
for assignments with node aliases. For a cluster with 2 data nodes nodeA and nodeB:volumes = /DFSRoot/<ALIAS>
which is equivalent to
nodeA.volumes = /DFSRoot/nodeA nodeB.volumes = /DFSRoot/nodeB
Note: The macro variable
<ALIAS>
cannot be used to specify the volume for a single node, otherwise the controller cannot be started. For example, do not specify configuration parameter asnodeA.volumes = /DFSRoot/<ALIAS>
. -
If the parameters are not qualified by node aliases, they indicate common configuration parameter values for all the data nodes in the cluster.
// for Windows maxConnections=64 // for Linux maxConnections=512 maxMemSize=12
Parameter Assignment Order
A configuration parameter may appear in the command line or multiple configuration files. It may be assigned different values in different locations. DolphinDB checks for parameter values in the order below. If a parameter is assigned value in a step, then assignments in all subsequent steps are ignored.
-
command line
-
cluster.nodes
-
qualified by specific node aliases in cluster.cfg
-
qualified by node aliases and wildcard characters in cluster.cfg
-
common configuration in cluster.cfg
Inter-node Communication
Parameter | Description |
---|---|
tcpNoDelay=1 | Whether to enable the TCP_NODELAY socket option. The default value is false. |
tcpUserTimeout=300000 | Set the socket option TCP_USER_TIMEOUT. The default value is 300000 (in ms). |
Parameter | Description |
---|---|
lanCluster=true | Whether the cluster is within a LAN (local area network). lanCluster = true (default): use UDP for heartbeats; lanCluster = false: use TCP for heartbeats. Set lanCluster = false if the system is deployed in the cloud. |
dfsChunkNodeHeartBeatTimeout |
An integer (of INT type) indicating the time interval (in seconds) for marking a data node/compute node/agent as down, i.e., if the controller has not received heartbeat message from a node for more than this time interval, the node will be marked and treated as down by default. The default value is 8. |
Parameter | Description |
---|---|
maxConnectionPerSite=2 | The maximum number of connections from the local node to remote nodes. |