Hannibal is a tool to help monitor and maintain HBase-Clusters that are configured for manual splitting.
While HBase provides metrics to monitor overall cluster health via JMX or Ganglia, it lacks the ability to monitor single regions in an easy way. This information is essential when your cluster is configured for manual splits, especially when the data growth is not uniform.
This tool tries to fill that gap by answering the following questions:
- How well are regions balanced over the cluster?
- How well are the regions split for each table?
- How do regions evolve over time?
The current version is compatible to HBase 0.90.x in distributed mode only. Support for 0.92.x is available in the 0.92 branch.
Java 6 JDK is required on the machine where this tool is built.
There is a tutorial video on [YouTube][yt], that shows how to install and use Hannibal. [yt]: http://www.youtube.com/watch?v=I4Kto41a5kE&hd=1][
-
Grab the sources from github:
$ git clone https://github.com/sentric/hannibal.git $ cd hannibal -
(only for HBase 0.92.x) Change the branch:
$ git checkout hbase-0.92 -
Copy
conf/hbase-site.template.xmltoconf/hbase-site.xmland adjust it. -
Build the project using the build script inside the root folder of the project:
$ ./build -
Run the start script inside the root folder of the project:
$ ./start
The first time you build, sbt will fetch all dependencies needed to build and run the the
application. It will take a few minutes to build.
When the application has started, you can access the web UI at: http://localhost:9000
Please note that history data about regions is only collected while the application is running, it will need to run for some time until the region detail graphs fill up.
For information about the usage, check out the Usage page on our Wiki.
HBase 0.90.x's API doesn't allow you to query information on running compactions directly, so what we do is to parse the RegionServers' log files directly, which are available through the service interface. HBase 0.92 allows to query compactions directly, but we still collect compactions using the logfile-parsing way, because this way we don't miss any short running compactions. The downside is that this doesn't work without further configuration because either, the url-pattern and the date pattern can differ from system to system. Regardless of which version of HBase you use, you should check those parameters in conf/application.conf:
compactions.logfile-url-pattern = "..."
compactions.logfile-date-format = "yyyy-MM-dd HH:mm:ss,SSS"
Informations about compactions are logged with INFO-Level, so the log levels need to be set at least to INFO.
Hannibal can set the log level to INFO for you, just edit conf/application.conf
and set
compactions.set-loglevels-on-startup = true
if you have problems, please make sure that the url-pattern is correct
compactions.loglevel-url-pattern = "..."
Please let me know if you have trouble with this, or have an idea how we could record information more easily.
More information about the tool can be found in the Wiki
If you need additional features please get in touch with us, maybe we can work something out.
Hannibal is released under MIT License, see LICENSE for details.
This tool is developed at Sentric by: Nils Kübler and Jiayong Ou
With help from: Jean-Pierre König, Christian Gügi and Vadim Kisselmann, Ben Taylor, Stephanie Höhn