hadoop入门笔记 （from udacity）

Hadoop生态圈：

Hive: support simple SQL-like language, and interpreted to Map-Reduce system. More used for long batched jobs

PIG: Yet another scripting language that is simpler than writeing MR

Impala: using SQL. directly access HDFS. Optimized, used for low latency queries

HBase: real time DB built on top of HDFS

Sqoop: get traditional DB data and put it to HDFS

Flume: simliar to sqoop

-----------------------------

HDFS:

将data分为默认64MB的chunk，存储于datanode中。每个chunk默认复制3份。 所有chunk的信息（怎么构成原来的data，复制的chunk在哪里）存储于namenode。 

Namenode很重要，可以使用active/standby来保证availability

Map reduce:

有mapreduce任务时，每个data node有一个task track的daemon (程序) 负责任务