This guide details the prerequisites required to run Pravega in production (Distributed mode only).
Setup an HDFS storage cluster running HDFS version 2.7+. HDFS is used as Tier 2 storage and must have sufficient capacity to store contents of all streams. The storage cluster is recommended to be run alongside Pravega on separate nodes.
Install the latest Java 8 from java.oracle.com. Packages are available for all major operating systems.
Pravega requires Zookeeper 3.5.1-alpha+. At least 3 Zookeeper nodes are recommended for a quorum. No special configuration is required for Zookeeper but it is recommended to use a dedicated cluster for Pravega.
This specific version of Zookeeper can be downloaded from Apache at zookeeper-3.5.1-alpha.tar.gz.
For installing Zookeeper see the Getting Started Guide.
Pravega requires Bookkeeper 4.4.0+. At least 3 Bookkeeper servers are recommended for a quorum.
This specific version of Bookkeeper can be downloaded from Apache at bookkeeper-server-4.4.0-bin.tar.gz.
For installing Bookkeeper see the Getting Started Guide.
Some specific Pravega instructions are shown below. All sets assuming being run from the
The following configuration options should be changed in the
# Comma separated list of <zp-ip>:<port> for all ZK servers zkServers=localhost:2181 # Alternatively specify a different path to the storage for /bk journalDirectory=/bk/journal ledgerDirectories=/bk/ledgers indexDirectories=/bk/index zkLedgersRootPath=/pravega/bookkeeper/ledgers
Initializing Zookeeper paths¶
The following paths need to be created in Zookeeper. From the
zookeeper-3.5.1-alpha directory on the Zookeeper servers run:
bin/zkCli.sh -server $ZK_URL create /pravega bin/zkCli.sh -server $ZK_URL create /pravega/bookkeeper
Before starting the bookie, it needs to be formatted:
bin/bookkeeper shell metaformat -nonInteractive
Start the bookie: