Keeps track about the blocks and in which data node they live
Maintains the edit log and logs all the read/write operations
Namenode resiliency
Option 1: backup the metadata both to local disk and a NFS
Option 2: runs a secondary namenode that maintains a merged copy of the edit log
Option 3: HDFS Federation - each namenode manages a specific namespace volume
Option 4: HDFS High Availability - hot standby namenod using a shared edit log. With that, zookeeper tracks the active namenodes. Must ensure that only one namenode is used at a time!
Data Node
Where the chunks of data are actually stored
Data Nodes talk to each other to replicate the chunks of data
Read/Write Flows
Reading
Name Node tells the client where the data lives
Client go and fetch the data directly from the data nodes
Writing
Client tells the Name Node that it's gonna write a file
Client sends data to a single Data Node
Data Node replicates the data across the other data nodes
