Clustered deployment

Configuration for distributing work across a cluster

Clustering of Vyne server instances allows for distribution of CPU intensive tasks, especially projections.

This is especially well suited for queries that return large numbers of records, which need to be projected. Distributing the work over a cluster provides significant performance improvements, as the work can be parallelized.

Understanding how work is distributed

When running in a multi-node cluster, the node that receives the query becomes the work-coordinator for that query.

Distributed projections

The work-coordinator is responsible for issuing the initial queries that find the source data. As data arrives on the work-coordinator, it is then added onto a queue for other nodes to pick up and project.

As worker nodes pick up projection tasks, they perform projection work (transforming and enriching data, calling remote services as required), and return the results back to the work-coordinator, who then sends the results back to the query issuer.


HTTP Caches

Caches are not shared between worker nodes, so remote services may receive a higher number of calls during parallelized enrichment.

Preferring work on remote nodes

As the cluster size grows, the work can be parallelized across a greater number of nodes. While this provides improved throughput, the work coordinator incurs a heavier workload in serialization and deserialization of work tasks and responses.

To account for this the vyne.projection.distributionRemoteBias allows tuning at which point the work is preferentially distributed to remote nodes, versus the query coordinator. Once this value is exceeded, the query coordinator node will perform a lower proportion of projection work in a query.

Work Distribution Configuration

For projection work to be distributed across all vyne nodes in a cluster the following configuration options are provided

Config parameterDescriptionDefault setting
vyne.projection.distributionModeDefines where projection work will be performed. On the local node or the cluster LOCAL, DISTRIBUTEDLOCAL
vyne.projection.distributionPacketSizeDefines the number of records in each work packet distributed to other vyne servers. Applicable to DISTRIBUTED distributionMode only100
vyne.projection.distributionRemoteBiasDefines the number of cluster members before work is preferentially distributed to remote nodes10

Cluster Configuration

Config parameterDescriptionDefault setting
vyne.hazelcast.discoverySpecifies the discovery mechanism hazelcast will use to find other instances. multicast, aws or eurekamulticast
vyne.hazelcast.eurekaUriURI of the coordinating eureka instance when discovery is set to eurekahttp://
vyne.hazelcast.memberTagDenotes which member type this is within the hazelcast clustervyne-query-service
vyne.hazelcast.networkInterfaceShould only be specified if the deployment environment requires that hazelcast only bind to a single network interface. The network interface supplied should be specified in hazelcast networking naming e.g. 10.10.., 172.16..
vyne.hazelcast.useMetadataForHostAndPortSpecific to eureka hazelcast discovery. Specifies if metadata should be published to and used for hazelcast node and port discoverytrue
vyne.hazelcast.awsPortScanRangeSpecific to aws hazelcast discovery. Specifies the port ranges that will be scanned from running hazelcast on ec2 instancestrue
vyne.hazelcast.taskPoolSizeSpecifies the number of threads dedicated to processing distributed work loads2
vyne.hazelcast.taskQueueSizeSpecifies the size of the queue of tasks on each cluster node. A value of 0 sets an infinite queue size0

Cluster formation and discovery

Vyne services can form a cluster discovering other nodes via one of three discovery mechanisms:

multicastVyne services attempt to use multicast to discover other cluster nodes
awsVyne services attempt use AWS metadata to discover other cluster nodes
eurekaVyne services use Eureka metadata to discover other cluster nodes

Example configurations


Enable clustering using discovery via local network multicast with distributed projections

        distributionMode: DISTRIBUTED

        discovery: multicast


Enable clustering using AWS discovery with distributed projections

The vyne query server should be run on a EC2 instances where AWS_REGION is specified as an environment variable and the EC2 instance is authorised to query ec2 instances via IAM.

AWS_REGION may alternatively be specified as a JVM property -DAWS_REGION=eu-west_2

        distributionMode: DISTRIBUTED

        discovery: aws


Enable clustering using Eureka discovery with distributed projections

        distributionMode: DISTRIBUTED

        discovery: eureka
        eurekaUri: http://eureka-server:8761/eureka/
        useMetadataForHostAndPort: true