Full database sync behaviour with enterprise agent

[timw]

Hi @tglman

I was reviewing how the distributed full database sync operates with the enterprise agent, and was wondering why it still performs a full database backup rather than attempting an incremental backup from a current lsn on the requesting node.

To be clear about my understanding of the reinstall process:

• With or without enterprise agent, a WAL based delta is requested
• If the delta can't be satisfied, then a full database sync is requested
• If "incremental backup" is supported (which effectively means that the enterprise agent is installed), a "full incremental backup" is performed - which is a backup of all the database files and the WAL (excluding Lucene indexes)
• If "incremental backup" isn't supported, then a classic directory based backup of the database is performed

The documentation for the enterprise features states "The incremental backup is used in the Distributed Architecture when a server node restarts. This avoids having to backup and transfer the entire database across the network.", which seems to imply that an actual incremental backup is sent, but it's actually a full database backup in the incremental backup format.

[tglman]

Hi,

OrientDB being multi-master allowing parallel executions means that two transactions that manipulate two different set of records may be applied in different order on the nodes, this means that a specific lsn which is a unique order of the transactions that are applied to the storage of a specific node may not have any correlation to the transaction associated to a lsn with the same value on another node.

Said in a simpler way, lsn are node specific, so cannot be used for sync purpose, can be used for local backup because the backup is associated to the specific node.

The docs are wrong/outdated, the use of incremental sync is exclusively to avoid to block writes on the node sending the sync, the delta read from the WAL serve the purpose to send reduced changes, will fix them.