Understanding Redis Partial Resynchronization: Offsets, Backlog Buffer, and Run ID

Components of Partial Resynchronization

Partial resynchronization in Redis consists of three parts:

Master and slave replication offsets

Master's replication backlog buffer

Server run ID (run ID)

Replication Offset

Both master and slave maintain a replication offset. Each time the master sends N bytes to the slave, it increments its offset by N; the slave does the same upon receiving N bytes. For example, if both offsets are 10000 and the master sends 99 bytes, both offsets become 10099.

Scenario Example

Consider a master with three slaves A, B, and C, all at offset 10010. The master sends 33 bytes; B and C receive them, updating their offsets to 10043, while A disconnects and remains at 10010, causing data inconsistency. When A reconnects, it issues a PSYNC command with its offset (10010). The master must decide whether to perform a full or partial resynchronization.

Replication Backlog Buffer

The backlog buffer is a fixed‑length queue maintained by the master (default size 1 MB). Every write command sent to slaves is also enqueued in this buffer, preserving recent commands together with their offsets. When a slave reconnects and sends its offset via PSYNC, the master checks whether the data starting at offset+1 still resides in the backlog. If it does, the master sends only those missing commands (partial resync); otherwise, it performs a full resynchronization.

Applying the Scenario

In the example, slave A reports offset 10010. The master finds that the data after 10010 is still present in the backlog and sends all commands from that point onward, allowing A to catch up without a full sync.

Server Run ID

Each Redis server generates a unique run ID at startup, consisting of 40 random hexadecimal characters. During initial replication, the master transmits its run ID to the slave, which stores it. If a slave disconnects and later reconnects to a master, it sends the saved run ID. If the run ID matches, the slave is reconnecting to the same master and partial resynchronization may be attempted; if it differs, the master must perform a full resynchronization.