Skip to content

Latest commit

 

History

History
135 lines (93 loc) · 4.12 KB

README.md

File metadata and controls

135 lines (93 loc) · 4.12 KB

Description

vsc-zk contains scripts to build zookeeper trees and tools using zookeeper.

vsc-zk

zookeeper tools

vsc-zk contains scripts to build zookeeper trees and tools using zookeeper.

  • zkinitree builds a tree in zookeeper from a config file.

  • vsc-zkrsync uses zookeeper to distribute rsync across multiple nodes when rsyncing (large) (shared) filesystems.

    A large fraction of rsync calls is spent in gathering the required metadata, before performing the actual data synchronisation. This issue becomes worse in the case of an incremental rsync. Distributing the load across multiple processes may lead to a significant performance gain even though other bottlenecks, such as access to the metadata, can become apparent.

    The implementation uses zookeeper to coordinate the distribution of collecting and sync the data across many processes and/or nodes.

Installation of Zookeeper

A zookeeper server with proper ACLs to a base znode (zookeeper node) is required.

If no such server is available, installation can be performed as follows:

zkversion=3.4.6
basepath=`mktemp -d` # not permanent, choose other directory as desired

mkdir -p $basepath/zk/server/data
cd $basepath/zk/server
# use another mirror if needed
wget http://apache.belnet.be/zookeeper/zookeeper-$zkversion/zookeeper-$zkversion.tar.gz
tar xzf zookeeper-$zkversion.tar.gz

rm -f zkServer.sh
cat > zkServer.sh <<EOF
#!/bin/bash
./zookeeper-$zkversion/bin/zkServer.sh \$@
EOF
chmod +x zkServer.sh

cat > zk.conf <<EOF
tickTime=2000
dataDir=$PWD/data
clientPort=2181
EOF

Start zookeeper server

`./zkServer.sh start $PWD/zk.conf `

Note: You should have Java (JDK 6 or greater) installed to run Zookeeper. More information can be found in the Zookeeper documentation.

Dependencies

Vsc-zk requires:

  • ZooKeeper 3.4 or above, since it is using transactions.
  • Kazoo 1.3.1 or above (pip install kazoo)
  • recent version of vsc-base (>= 1.6.7)

Build the initial zookeeper tree

Use zkinitree to build an initial tree in zookeeper from a config file. It will create paths with specified ACLs.

An example config file can be found in the examples folder (zkinitree.cfg)

Usage of zkrsync

Example usage for N-parallelised rsync :

Put reusable values in a config file, e.g. zkrs.conf:

[MAIN]
servers=<servers> 
session=<session>
domain=<domain>
daemon=true
user=<user>
passwd=<password>
depth=<depth>
#debug=true
#dryrun=false
#delete=false

Start N+1 sources (first source client will be the Master)

zkrsync -S -r <sourcepath> --configfiles=zkrs.conf

Start N destinations:

zkrsync -D -r <destpath> --configfiles=zkrs.conf

Testing pathbuilding: add the pathsonly option

run zkrsync -H to see all options

Default logging goes to /tmp/zkrsync/<session>-<source|dest>-<pid>.log. Use the logfile option to change the template. Use the debug option (or short -d) for more verbose logging.

When running in daemon mode, a pidfile will be generated. Default location is /tmp/zkrsync/<session>-<source|dest>-<pid>.pid (pid is pid of process that starts the daemon). This can also be templated with the pidfile option.

If anything (zookeeper-related) goes wrong (no cleanup has been done)

  • kill all running source and destination clients (of that session)
  • wait about 20 seconds (ymmv), making sure all zookeeper connections have timed out
  • run exactly one source client, if the previous shutdown was not finished cleanly, zookeeper will recognise this and attempt a new cleanup before exiting.
  • Should the previous step fail to clean up, there might still be a running client. Make sure they are all killed
  • Start a new zkrsync process

Global remarks:

  • Always define a session to make sure you do not mix up different sessions
  • Make sure parameters path, dryrun and delete are all the same: These parameters are not checked (at this moment) and each client will use its own parameters. This can lead to inconsistencies (e.g., some paths runs with the dry-run option while others are effectively synced.)
  • Parameter depth is only used on pathbuilding, so the Source Master will always provide this.