Awesome

Mongoose's HDFS storage driver

Introduction

The storage driver extends the Mongoose's Abstract NIO Storage Driver and uses the following libraries:

hadoop-common
hadoop-hdfs-client

Features

Authentification: simple
SSL/TLS - TODO
Item types:
- data
- path - TODO
Path listing input
Automatic destination path creation on demand
Data item operation types:
- create, additional modes:
  - copy
- read
  - full
  - random byte ranges
  - fixed byte ranges
  - content verification
- update
  - full (overwrite)
  - fixed byte ranges: append mode only
- delete
- noop
Path item operation types (TODO):
- create, additional modes:
  - copy - ?
  - concatenation - ?
- read
- delete
- noop

Usage

Get the latest pre-built jar file which is available at: http://repo.maven.apache.org/maven2/com/github/emc-mongoose/mongoose-storage-driver-hdfs/ The jar file may be downloaded manually and placed into the <USER_HOME_DIR>/.mongoose/<VERSION>/ext directory of Mongoose to be automatically loaded into the runtime.

java -jar mongoose-<VERSION>.jar \
    --storage-driver-type=hdfs \
    --storage-net-node-addrs=<NODE_IP_ADDRS> \
    --storage-net-node-port=<NODE_PORT> \
    --storage-auth-uid=<USER_ID> \
    ...

Docker

Standalone

docker run \
    --network host \
    emcmongoose/mongoose-storage-driver-hdfs \
    --storage-net-node-addrs=<NODE_IP_ADDRS> \
    --storage-net-node-port=<NODE_PORT> \
    --storage-auth-uid=<USER_ID> \
    ...

Distributed

Additional Node

docker run \
    --network host \
    --expose 1099 \
    emcmongoose/mongoose-storage-driver-hdfs
    --run-node

Entry Node

docker run \
    --network host \
    emcmongoose/mongoose-storage-driver-hdfs \
    --load-step-node-addrs=<ADDR1,ADDR2,...> \
    --storage-net-node-addrs=<NODE_IP_ADDRS> \
    --storage-net-node-port=<NODE_PORT> \
    --storage-auth-uid=<USER_ID> \
    ...

Advanced

Sources

git clone https://github.com/emc-mongoose/mongoose-storage-driver-hdfs.git
cd mongoose-storage-driver-hdfs

Test

./gradlew clean test

Build

./gradlew clean jar

Embedding

compile group: 'com.github.emc-mongoose', name: 'mongoose-storage-driver-hdfs', version: '<VERSION>'

Development Notes

Node's FS browser is available at default port #50070

HDFS default port #8020 (or 9000?)

Basic Testing

Run the pseudo distributed HDFS cluster

docker run -p 22022:22 -p 8020:8020 -p 50010:50010 -p 50020:50020 -p 50070:50070 -p 50075:50075 -d pravega/hdfs

Open the browser and check the HDFS share @ http://127.0.0.1:50070/explorer.html to observe the filesystem
Build the Mongoose HDFS storage driver jar either use the Docker image.
Put the HDFS storage driver jar into the Mongoose's ext directory either use the Docker image with HDFS support.
Run some Mongoose test, for example:

java -jar mongoose-<VERSION>.jar \
    --item-data-size=64MB \
    --item-output-file=hdfs.files.csv \
    --item-output-path=/test \
    --load-op-limit-count=100 \
    --storage-auth-uid=root \
    --storage-driver-limit-concurrency=10 \
    --storage-driver-type=hdfs \
    --storage-net-node-addrs=<HADOOP_NAME_NODE_IP_ADDR> \
    --storage-net-node-port=8020

Operations

The information below describes which particular methods are invoked on the endpoint in each case. The endpoint hereafter is a Hadoop FileSystem instance.

The item types data and path are supported. token type is not supported.

Data

Operations on the data items type are implemented as file operations

Noop

Doesn't invoke anything.

Create

The method create(Path, FsPerm, boolean, int, short, long, null) is invoked with calculated output buffer size. The returned FSDataOutputStream is used to write the data.

Copy

Uses both create and open methods to obtain output and input streams

Concatenation

Note: not supported as far as HDFS doesn't allow to concatenate to the new & empty destination object

concat(Path dst, Path[] srcs) is invoked (doesn't return anything).

Note:

source files ranges concatenation is not supported.

Read

open(Path f, int bufferSize) is invoked. The returned FSDataInputStream instance is used then to read the data.

Partial

The same method used as above, because the FSDataInputStream supports the positioning needed for the partial read.

Random Ranges

Supported

Fixed Ranges

Supported

Update

Overwrite

Supported

Random Ranges

Not supported as far as FSDataOutputStream doesn't allow positioning.

Fixed Ranges

Not supported except the append case.

Append

Supported

Delete

delete(Path f, false) is invoked.

Path

Operations on the path are implemented as directory operations

Create

mkdirs(Path)

Copy

TODO

Concatenation

TODO

Read

listFiles(Path f, false) is invoked returning the RemoteIterator instance which is used to iterate the directory contents.

Delete

delete(Path f, true) is invoked.

Token

Not supported