CSE Roadmap

Specialist tier · about 2 hours · 5 parts

Part 4 of 5 · 2 min read

Rebalancing without downtime

Moving data while the system keeps serving reads and writes requires moving it incrementally, never all at once.

Rebalancing without downtime

Think of renovating one wing of a hospital while the rest keeps treating patients. Equipment and staff move a little at a time, rather than closing the whole building at once.

In plain words: rebalancing without downtime means the system continues serving reads and writes throughout the data movement, rather than pausing to complete the move first. This requires moving data incrementally, a portion at a time, while keeping track of which partition currently holds the authoritative copy of each piece of data being moved, so a query arriving mid-move is still routed correctly. Rebalancing also needs to avoid moving too much data at once, since the movement itself consumes network and disk bandwidth that the system still needs for ordinary reads and writes.

A worked example. A cluster rebalancing 20% of its data moves it in small batches over several hours, continuously updating its routing information so that a query for a specific row always reaches whichever partition currently holds it, whether that is the old location or the new one.

The key lines. No new SQL; avoiding downtime during rebalancing is an operational guarantee the database's rebalancing mechanism must provide, not something a query can request.

Zero-downtime rebalancing

Zero-downtime rebalancing is rebalancing carried out while the system continues serving reads and writes, achieved by moving data incrementally and keeping routing information continuously updated during the move.

Example: renovating one hospital wing at a time while the rest keeps treating patients mirrors incremental, zero-downtime rebalancing.

Remember this

Rebalancing exists because changing the number of nodes changes how data should be spread across them. Automatic rebalancing trades control for convenience; doing it without downtime trades speed for safety, moving data incrementally rather than all at once.

Quick notes

  • Rebalancing moves data between partitions, typically after the number of nodes in a cluster changes.
  • Automatic rebalancing requires no operator action but can act at an inconvenient time.
  • Manual rebalancing gives an operator control but requires them to notice the need and act.
  • Zero-downtime rebalancing moves data incrementally while keeping routing information continuously updated.
  • Moving too much data at once competes with the system's ordinary read and write traffic for bandwidth.