I’m working on adding support for High Availability Namenode support within hadoop-conf.

hadoop-conf

First, some background on hadoop-conf. It’s my Hadoop-on-a-laptop deployment setup tool, simply a Makefile with some shell scripts and XSL templates. It just developed slowly over the past year to ease development with Hadoop.

Its advantages are:

  • Generates Hadoop configurations in a developer-friendly (i.e. hackable) way using a Makefile
  • Configures difficult-to-configure system services (DNS, Kerberos, ntpd, ntpclient)
  • Modifies core system properties on a Linux VM, avoiding any need for modifications to your main development OS (* but see Caveat below)
  • Works disconnected from the Internet (if you are working while you commute on BART or Caltrain and have spotty or no Internet connection you’ll appreciate this)
  • Enables secure Hadoop configuration (i.e. sets up a local Kerberos infrastructure and uses it)
  • Facilitates the following development cycle:
  1. Configure and install hadoop (make install)
  2. Start hadoop (make start)
  3. Observe bug or opportunity for enhancement
  4. Modify some hadoop code to fix the bug or add the enhancement
  5. Rebuild hadoop (make build)
  6. Shutdown hadoop (make stop)
  7. goto 2 (or 1 if reconfiguration is desired)

Caveats:

  • It does require that your host system run ntpd. But if you are running Mac OS X, it’s already running by default. So no modifications are needed, if your dev machine is a Mac. You may need to configure or start it on other OSes.

  • step 5 (make build) is not yet supported by hadoop-conf. It’s simple to describe: simply go to the top-level hadoop directory and do:

    	   mvn -Pdist package
    

or, opting for certainty of development speed):

  	   mvn -Pdist clean package

High Availability Namenode

I am working on adding support within hadoop-conf for:

Historical Background

The umbrella JIRA is: HDFS-3278 Umbrella Jira for HDFS-HA Phase 2.

The predecessor of this umbrella JIRA is: HDFS-1623 High Availability Framework for HDFS NN

Note that HDFS-3278 mentions that there are two phases to HA NN. The first phase was covered by the predecessor (HDFS-1623). In the description of HDFS-3278 is written:

Branch HDFS-1623 was merged into trunk for tactical reasons even though the work for HA was not complete, Branch HDFS-1623 contained mechanisms for keeping a standby Hot (ie read from shared journal), dual block reports, fencing of DNs, Zookeeper library for leader election etc.



blog comments powered by Disqus

Published

08 December 2012