Short note: Shadow Master and Cluak

Shadow Master

The GFS master is a single point of failure, so its state is replicated.

• The operation log and checkpoints are replicated on several machines. If the master fails, an external monitor starts a new master process quickly.
• Shadow masters are additional replicas of the master that provide read-only access to the file system even when the primary master is down.
• They are "shadows", not mirrors: they may lag the primary slightly (fractions of a second), so clients can see slightly stale metadata.
• A shadow master reads the replicated operation log and applies the changes to its own data structures in the same order as the master. It polls chunkservers at start-up to find chunk locations and exchanges handshakes with them.
• Benefit: improves availability for reads, and reduces load on the master for applications that can tolerate slightly stale data.

Chunkserver services

Chunkservers are the worker nodes that store data.

• Store chunks (64 MB, each with a 64-bit handle) as Linux files on local disks and serve read and write requests from clients directly.
• Each chunk is replicated on three servers; one acts as primary holding a lease to order mutations.
• Send HeartBeat messages to the master to report their state and the chunks they hold, and receive instructions in reply (e.g. delete a chunk, copy a chunk).
• Maintain checksums for each 64 KB block to detect corruption and verify data before returning it.
• Perform chunk replication to other chunkservers when asked by the master.

 Primary master --log--> Shadow masters (read-only)
       |  heartbeat
 +-----v-----+ +-----------+ +-----------+
 |Chunkserver| |Chunkserver| |Chunkserver|
 +-----------+ +-----------+ +-----------+