I understand that Journal Nodes are like the central repo for all edit logs (no matter which namenode is currently active, all push to journal node). I suppose the QuorumJournalManager handles this pushing from active namenode and syncing the standby namenode. But how does the Journal Nodes work?
- Are they independent processes, who just accept edits from active namenode, and QJM handles sending edits to all independent journal nodes and waits for acknowledgement from atleast quorum of nodes or they have some internal replication mechanism ?
- Do they have some leader/ follower mechanism ?
- If they don’t have any replication mechanism how is data consistency maintained in case of failure of any Journal Node?
I couldn’t find any good documentation on actual working, can you also provide any link to good documentation on actual working of hadoop processes?