Nellie

Away from the circus

Nellie is intended to be a suite of interconnected APIs and commandline tools for accessing and manipulating data in Hadoop clusters from outside of those clusters.

Usage

$ nellie ls /tmp
755 mapred supergroup 0   2013-10-22 12:55:28 mapred/
644 hdfs   supergroup 510 2013-11-16 15:03:28 rain.txt

$ echo "012345" > data.dat
$ nellie put data.dat /tmp
$ nellie ls /tmp
644 hdfs   supergroup 6   2013-11-15 00:18:08 data.dat
755 mapred supergroup 0   2013-10-22 12:55:28 mapred/
644 hdfs   supergroup 510 2013-11-16 15:03:28 rain.txt

$ nellie fstat /tmp/data.dat
Path:        /tmp/data.dat
Type:        FILE
Length:      6
Permission:  644
Owner:       hdfs
Group:       supergroup
Mod. time:   2013-11-15 00:18:08
Access time: 2013-11-15 00:18:08
Block size:  134217728
Replication: 3

$ time nellie get /tmp/data.dat
012345
real    0m0.234s
user    0m0.131s
sys 0m0.052s

Features

• It's FAST
• Runnable outside of a cluster
• Minimal runtime dependencies
• Portability

Promise

• API first

Status

Currently, Nellie is more like a proof of concept than a real thing.

Building

1. Install SBCL

• MacOS (using Homebrew):

  brew install sbcl

• Ubuntu:

  sudo yum install sbcl

• RHEL/CentOS:

Follow these instrucions.

2. Install Quicklisp

curl -O http://beta.quicklisp.org/quicklisp.lisp
sbcl --load quicklisp.lisp \
     --eval '(quicklisp-quickstart:install)' \
     --eval '(ql:add-to-init-file)' \
     --eval '(quit)' 

3. Install Buildapp

curl -O http://www.xach.com/lisp/buildapp.tgz
tar xfz buildapp.tgz
cd buildapp-1.5
sudo make install

IMPORTANT: Even though it's possible, don't install buildapp using quicklisp -- the version coming from QL is outdated and will not work.

4. Build Nellie

cd nellie
./build.sh

This will create a file called nellie in the dist/ directory.

The file is a fully self-contained executable without any external dependencies. The file is fairly large (~80G), so it might make sense to compress it before sending over network, etc.

Despite its size, this binary loads reasonably fast:

$ time dist/nellie
....
real    0m0.024s
user    0m0.004s
sys     0m0.019s

This is so because when a binary is executed, the OS uses memory mapping instead of loading the whole file into memory. In case of SBCL, there is nothing like .jar or .pyc loading phase of any kind as this file is actually a memory dump of a running SBCL virtual machine.

TODO

• Tests
• Documentation
• Code cleanup
• A better build script
• More use cases covered
• Other Hadoop components (HCat, etc..)
• Performance improvements

License

This software is covered by the MIT license.